winit/src/platform/linux/x11/mod.rs

1373 lines
57 KiB
Rust
Raw Normal View History

#![cfg(any(target_os = "linux", target_os = "dragonfly", target_os = "freebsd", target_os = "openbsd"))]
2015-04-24 09:51:23 +02:00
2018-05-07 17:36:21 -04:00
pub mod ffi;
mod events;
mod monitor;
mod window;
mod xdisplay;
mod dnd;
mod ime;
mod util;
2015-09-24 09:11:59 +02:00
pub use self::monitor::{MonitorId, get_available_monitors, get_primary_monitor};
pub use self::window::{Window2, XWindow};
2015-12-24 10:57:08 +01:00
pub use self::xdisplay::{XConnection, XNotSupported, XError};
use std::{mem, ptr, slice};
2018-05-07 17:36:21 -04:00
use std::sync::{Arc, mpsc, Weak};
use std::sync::atomic::{self, AtomicBool};
use std::cell::RefCell;
use std::collections::HashMap;
use std::ffi::CStr;
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
use std::os::raw::*;
use libc::{self, setlocale, LC_CTYPE};
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
use parking_lot::Mutex;
2018-05-07 17:36:21 -04:00
use {
ControlFlow,
CreationError,
DeviceEvent,
Event,
EventsLoopClosed,
KeyboardInput,
WindowAttributes,
WindowEvent,
};
use events::ModifiersState;
use platform::PlatformSpecificWindowBuilderAttributes;
use self::dnd::{Dnd, DndState};
use self::ime::{ImeReceiver, ImeSender, ImeCreationError, Ime};
pub struct EventsLoop {
display: Arc<XConnection>,
wm_delete_window: ffi::Atom,
dnd: Dnd,
ime_receiver: ImeReceiver,
ime_sender: ImeSender,
ime: RefCell<Ime>,
windows: Arc<Mutex<HashMap<WindowId, WindowData>>>,
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// Please don't laugh at this type signature
shared_state: RefCell<HashMap<WindowId, Weak<Mutex<window::SharedState>>>>,
devices: RefCell<HashMap<DeviceId, Device>>,
xi2ext: XExtension,
pending_wakeup: Arc<AtomicBool>,
root: ffi::Window,
// A dummy, `InputOnly` window that we can use to receive wakeup events and interrupt blocking
// `XNextEvent` calls.
wakeup_dummy_window: ffi::Window,
}
2017-10-25 11:03:57 -07:00
#[derive(Clone)]
pub struct EventsLoopProxy {
pending_wakeup: Weak<AtomicBool>,
display: Weak<XConnection>,
wakeup_dummy_window: ffi::Window,
}
impl EventsLoop {
pub fn new(display: Arc<XConnection>) -> EventsLoop {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let wm_delete_window = unsafe { util::get_atom(&display, b"WM_DELETE_WINDOW\0") }
.expect("Failed to call XInternAtom (WM_DELETE_WINDOW)");
let dnd = Dnd::new(Arc::clone(&display))
.expect("Failed to call XInternAtoms when initializing drag and drop");
let (ime_sender, ime_receiver) = mpsc::channel();
// Input methods will open successfully without setting the locale, but it won't be
// possible to actually commit pre-edit sequences.
unsafe { setlocale(LC_CTYPE, b"\0".as_ptr() as *const _); }
let ime = RefCell::new({
let result = Ime::new(Arc::clone(&display));
if let Err(ImeCreationError::OpenFailure(ref state)) = result {
panic!(format!("Failed to open input method: {:#?}", state));
}
result.expect("Failed to set input method destruction callback")
});
let xi2ext = unsafe {
let mut result = XExtension {
opcode: mem::uninitialized(),
first_event_id: mem::uninitialized(),
first_error_id: mem::uninitialized(),
};
let res = (display.xlib.XQueryExtension)(
display.display,
b"XInputExtension\0".as_ptr() as *const c_char,
&mut result.opcode as *mut c_int,
&mut result.first_event_id as *mut c_int,
&mut result.first_error_id as *mut c_int);
if res == ffi::False {
panic!("X server missing XInput extension");
}
result
};
unsafe {
let mut xinput_major_ver = ffi::XI_2_Major;
let mut xinput_minor_ver = ffi::XI_2_Minor;
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
if (display.xinput2.XIQueryVersion)(
display.display,
&mut xinput_major_ver,
&mut xinput_minor_ver,
) != ffi::Success as libc::c_int {
panic!(
"X server has XInput extension {}.{} but does not support XInput2",
xinput_major_ver,
xinput_minor_ver,
);
}
}
let root = unsafe { (display.xlib.XDefaultRootWindow)(display.display) };
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
util::update_cached_wm_info(&display, root);
let wakeup_dummy_window = unsafe {
let (x, y, w, h) = (10, 10, 10, 10);
let (border_w, border_px, background_px) = (0, 0, 0);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
(display.xlib.XCreateSimpleWindow)(
display.display,
root,
x,
y,
w,
h,
border_w,
border_px,
background_px,
)
};
let result = EventsLoop {
pending_wakeup: Arc::new(AtomicBool::new(false)),
display,
wm_delete_window,
dnd,
ime_receiver,
ime_sender,
ime,
windows: Arc::new(Mutex::new(HashMap::new())),
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
shared_state: RefCell::new(HashMap::new()),
devices: RefCell::new(HashMap::new()),
xi2ext,
root,
wakeup_dummy_window,
};
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// Register for device hotplug events
unsafe {
util::select_xinput_events(
&result.display,
root,
ffi::XIAllDevices,
ffi::XI_HierarchyChangedMask,
)
}.queue(); // The request buffer is flushed during init_device
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
result.init_device(ffi::XIAllDevices);
result
}
/// Returns the `XConnection` of this events loop.
#[inline]
pub fn x_connection(&self) -> &Arc<XConnection> {
&self.display
}
pub fn create_proxy(&self) -> EventsLoopProxy {
EventsLoopProxy {
pending_wakeup: Arc::downgrade(&self.pending_wakeup),
display: Arc::downgrade(&self.display),
wakeup_dummy_window: self.wakeup_dummy_window,
}
}
pub fn poll_events<F>(&mut self, mut callback: F)
where F: FnMut(Event)
{
let mut xev = unsafe { mem::uninitialized() };
loop {
// Get next event
unsafe {
// Ensure XNextEvent won't block
let count = (self.display.xlib.XPending)(self.display.display);
if count == 0 {
break;
}
(self.display.xlib.XNextEvent)(self.display.display, &mut xev);
}
self.process_event(&mut xev, &mut callback);
}
}
pub fn run_forever<F>(&mut self, mut callback: F)
where F: FnMut(Event) -> ControlFlow
{
let mut xev = unsafe { mem::uninitialized() };
loop {
unsafe { (self.display.xlib.XNextEvent)(self.display.display, &mut xev) }; // Blocks as necessary
let mut control_flow = ControlFlow::Continue;
// Track whether or not `Break` was returned when processing the event.
{
let mut cb = |event| {
if let ControlFlow::Break = callback(event) {
control_flow = ControlFlow::Break;
}
};
self.process_event(&mut xev, &mut cb);
}
if let ControlFlow::Break = control_flow {
break;
}
}
}
fn process_event<F>(&mut self, xev: &mut ffi::XEvent, mut callback: F)
where F: FnMut(Event)
{
let xlib = &self.display.xlib;
// XFilterEvent tells us when an event has been discarded by the input method.
// Specifically, this involves all of the KeyPress events in compose/pre-edit sequences,
// along with an extra copy of the KeyRelease events. This also prevents backspace and
// arrow keys from being detected twice.
if ffi::True == unsafe { (self.display.xlib.XFilterEvent)(
xev,
{ let xev: &ffi::XAnyEvent = xev.as_ref(); xev.window }
) } {
return;
}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let event_type = xev.get_type();
match event_type {
ffi::MappingNotify => {
unsafe { (xlib.XRefreshKeyboardMapping)(xev.as_mut()); }
self.display.check_errors().expect("Failed to call XRefreshKeyboardMapping");
}
ffi::ClientMessage => {
let client_msg: &ffi::XClientMessageEvent = xev.as_ref();
let window = client_msg.window;
let window_id = mkwid(window);
if client_msg.data.get_long(0) as ffi::Atom == self.wm_delete_window {
callback(Event::WindowEvent { window_id, event: WindowEvent::CloseRequested });
} else if client_msg.message_type == self.dnd.atoms.enter {
let source_window = client_msg.data.get_long(0) as c_ulong;
let flags = client_msg.data.get_long(1);
let version = flags >> 24;
self.dnd.version = Some(version);
let has_more_types = flags - (flags & (c_long::max_value() - 1)) == 1;
if !has_more_types {
let type_list = vec![
client_msg.data.get_long(2) as c_ulong,
client_msg.data.get_long(3) as c_ulong,
client_msg.data.get_long(4) as c_ulong
];
self.dnd.type_list = Some(type_list);
} else if let Ok(more_types) = unsafe { self.dnd.get_type_list(source_window) } {
self.dnd.type_list = Some(more_types);
}
} else if client_msg.message_type == self.dnd.atoms.position {
// This event occurs every time the mouse moves while a file's being dragged
// over our window. We emit HoveredFile in response; while the Mac OS X backend
// does that upon a drag entering, XDnD doesn't have access to the actual drop
// data until this event. For parity with other platforms, we only emit
// HoveredFile the first time, though if winit's API is later extended to
// supply position updates with HoveredFile or another event, implementing
// that here would be trivial.
let source_window = client_msg.data.get_long(0) as c_ulong;
// Equivalent to (x << shift) | y
// where shift = mem::size_of::<c_short>() * 8
// Note that coordinates are in "desktop space", not "window space"
// (in x11 parlance, they're root window coordinates)
//let packed_coordinates = client_msg.data.get_long(2);
//let shift = mem::size_of::<libc::c_short>() * 8;
//let x = packed_coordinates >> shift;
//let y = packed_coordinates & !(x << shift);
// By our own state flow, version should never be None at this point.
let version = self.dnd.version.unwrap_or(5);
// Action is specified in versions 2 and up, though we don't need it anyway.
//let action = client_msg.data.get_long(4);
let accepted = if let Some(ref type_list) = self.dnd.type_list {
type_list.contains(&self.dnd.atoms.uri_list)
} else {
false
};
if accepted {
self.dnd.source_window = Some(source_window);
unsafe {
if self.dnd.result.is_none() {
let time = if version >= 1 {
client_msg.data.get_long(3) as c_ulong
} else {
// In version 0, time isn't specified
ffi::CurrentTime
};
// This results in the SelectionNotify event below
self.dnd.convert_selection(window, time);
}
self.dnd.send_status(window, source_window, DndState::Accepted)
.expect("Failed to send XDnD status message.");
}
} else {
unsafe {
self.dnd.send_status(window, source_window, DndState::Rejected)
.expect("Failed to send XDnD status message.");
self.dnd.send_finished(window, source_window, DndState::Rejected)
.expect("Failed to send XDnD finished message.");
}
self.dnd.reset();
}
} else if client_msg.message_type == self.dnd.atoms.drop {
if let Some(source_window) = self.dnd.source_window {
if let Some(Ok(ref path_list)) = self.dnd.result {
for path in path_list {
callback(Event::WindowEvent {
window_id,
event: WindowEvent::DroppedFile(path.clone()),
});
}
}
unsafe {
self.dnd.send_finished(window, source_window, DndState::Accepted)
.expect("Failed to send XDnD finished message.");
}
}
self.dnd.reset();
} else if client_msg.message_type == self.dnd.atoms.leave {
self.dnd.reset();
callback(Event::WindowEvent {
window_id,
event: WindowEvent::HoveredFileCancelled,
});
} else if self.pending_wakeup.load(atomic::Ordering::Relaxed) {
self.pending_wakeup.store(false, atomic::Ordering::Relaxed);
callback(Event::Awakened);
}
}
ffi::SelectionNotify => {
let xsel: &ffi::XSelectionEvent = xev.as_ref();
let window = xsel.requestor;
let window_id = mkwid(window);
if xsel.property == self.dnd.atoms.selection {
let mut result = None;
// This is where we receive data from drag and drop
if let Ok(mut data) = unsafe { self.dnd.read_data(window) } {
let parse_result = self.dnd.parse_data(&mut data);
if let Ok(ref path_list) = parse_result {
for path in path_list {
callback(Event::WindowEvent {
window_id,
event: WindowEvent::HoveredFile(path.clone()),
});
}
}
result = Some(parse_result);
}
self.dnd.result = result;
}
}
ffi::ConfigureNotify => {
let xev: &ffi::XConfigureEvent = xev.as_ref();
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// So apparently...
// XSendEvent (synthetic ConfigureNotify) -> position relative to root
// XConfigureNotify (real ConfigureNotify) -> position relative to parent
// https://tronche.com/gui/x/icccm/sec-4.html#s-4.1.5
// We don't want to send Moved when this is true, since then every Resized
// (whether the window moved or not) is accompanied by an extraneous Moved event
// that has a position relative to the parent window.
let is_synthetic = xev.send_event == ffi::True;
let window = xev.window;
let window_id = mkwid(window);
let new_size = (xev.width, xev.height);
let new_position = (xev.x, xev.y);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let (resized, moved) = {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let mut windows = self.windows.lock();
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
if let Some(window_data) = windows.get_mut(&WindowId(window)) {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let (mut resized, mut moved) = (false, false);
if window_data.config.size.is_none() {
window_data.config.size = Some(new_size);
resized = true;
}
if window_data.config.size.is_none() && is_synthetic {
window_data.config.position = Some(new_position);
moved = true;
}
if !resized {
if window_data.config.size != Some(new_size) {
window_data.config.size = Some(new_size);
resized = true;
}
}
if !moved && is_synthetic {
if window_data.config.position != Some(new_position) {
window_data.config.position = Some(new_position);
moved = true;
}
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
if !is_synthetic
&& window_data.config.inner_position != Some(new_position) {
window_data.config.inner_position = Some(new_position);
// This way, we get sent Moved when the decorations are toggled.
window_data.config.position = None;
self.shared_state.borrow().get(&WindowId(window)).map(|window_state| {
if let Some(window_state) = window_state.upgrade() {
// Extra insurance against stale frame extents
(*window_state.lock()).frame_extents.take();
}
});
}
(resized, moved)
} else {
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
return;
}
};
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
if resized {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let (width, height) = (xev.width as u32, xev.height as u32);
callback(Event::WindowEvent {
window_id,
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
event: WindowEvent::Resized(width, height),
});
}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
if moved {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// We need to convert client area position to window position.
self.shared_state.borrow().get(&WindowId(window)).map(|window_state| {
if let Some(window_state) = window_state.upgrade() {
let (x, y) = {
let (inner_x, inner_y) = (xev.x as i32, xev.y as i32);
let mut window_state_lock = window_state.lock();
if (*window_state_lock).frame_extents.is_some() {
(*window_state_lock).frame_extents
.as_ref()
.unwrap()
.inner_pos_to_outer(inner_x, inner_y)
} else {
let extents = util::get_frame_extents_heuristic(
&self.display,
window,
self.root,
);
let outer_pos = extents.inner_pos_to_outer(inner_x, inner_y);
(*window_state_lock).frame_extents = Some(extents);
outer_pos
}
};
callback(Event::WindowEvent {
window_id,
event: WindowEvent::Moved(x, y),
});
}
});
}
}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
ffi::ReparentNotify => {
let xev: &ffi::XReparentEvent = xev.as_ref();
let window = xev.window;
// This is generally a reliable way to detect when the window manager's been
// replaced, though this event is only fired by reparenting window managers
// (which is almost all of them). Failing to correctly update WM info doesn't
// really have much impact, since on the WMs affected (xmonad, dwm, etc.) the only
// effect is that we waste some time trying to query unsupported properties.
util::update_cached_wm_info(&self.display, self.root);
self.shared_state
.borrow()
.get(&WindowId(window))
.map(|window_state| {
if let Some(window_state) = window_state.upgrade() {
(*window_state.lock()).frame_extents.take();
}
});
}
ffi::DestroyNotify => {
let xev: &ffi::XDestroyWindowEvent = xev.as_ref();
let window = xev.window;
let window_id = mkwid(window);
// In the event that the window's been destroyed without being dropped first, we
// cleanup again here.
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
self.windows.lock().remove(&WindowId(window));
// Since all XIM stuff needs to happen from the same thread, we destroy the input
// context here instead of when dropping the window.
self.ime
.borrow_mut()
.remove_context(window)
.expect("Failed to destroy input context");
callback(Event::WindowEvent { window_id, event: WindowEvent::Destroyed });
}
ffi::Expose => {
let xev: &ffi::XExposeEvent = xev.as_ref();
let window = xev.window;
let window_id = mkwid(window);
callback(Event::WindowEvent { window_id, event: WindowEvent::Refresh });
}
ffi::KeyPress | ffi::KeyRelease => {
use events::ElementState::{Pressed, Released};
// Note that in compose/pre-edit sequences, this will always be Released.
let state = if xev.get_type() == ffi::KeyPress {
Pressed
} else {
Released
};
let xkev: &mut ffi::XKeyEvent = xev.as_mut();
let window = xkev.window;
let window_id = mkwid(window);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// Standard virtual core keyboard ID. XInput2 needs to be used to get a reliable
// value, though this should only be an issue under multiseat configurations.
let device = 3;
let device_id = mkdid(device);
// When a compose sequence or IME pre-edit is finished, it ends in a KeyPress with
// a keycode of 0.
if xkev.keycode != 0 {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let modifiers = ModifiersState {
alt: xkev.state & ffi::Mod1Mask != 0,
shift: xkev.state & ffi::ShiftMask != 0,
ctrl: xkev.state & ffi::ControlMask != 0,
logo: xkev.state & ffi::Mod4Mask != 0,
};
let keysym = unsafe {
let mut keysym = 0;
(self.display.xlib.XLookupString)(
xkev,
ptr::null_mut(),
0,
&mut keysym,
ptr::null_mut(),
);
self.display.check_errors().expect("Failed to lookup keysym");
keysym
};
let virtual_keycode = events::keysym_to_element(keysym as c_uint);
callback(Event::WindowEvent {
window_id,
event: WindowEvent::KeyboardInput {
device_id,
input: KeyboardInput {
state,
scancode: xkev.keycode - 8,
virtual_keycode,
modifiers,
},
}
});
}
if state == Pressed {
let written = if let Some(ic) = self.ime.borrow().get_context(window) {
unsafe { util::lookup_utf8(&self.display, ic, xkev) }
} else {
return;
};
for chr in written.chars() {
let event = Event::WindowEvent {
window_id,
event: WindowEvent::ReceivedCharacter(chr),
};
callback(event);
}
}
}
ffi::GenericEvent => {
let guard = if let Some(e) = GenericEventCookie::from_event(&self.display, *xev) { e } else { return };
let xev = &guard.cookie;
if self.xi2ext.opcode != xev.extension {
return;
}
use events::WindowEvent::{Focused, CursorEntered, MouseInput, CursorLeft, CursorMoved, MouseWheel, AxisMotion};
use events::ElementState::{Pressed, Released};
use events::MouseButton::{Left, Right, Middle, Other};
use events::MouseScrollDelta::LineDelta;
use events::{Touch, TouchPhase};
match xev.evtype {
ffi::XI_ButtonPress | ffi::XI_ButtonRelease => {
let xev: &ffi::XIDeviceEvent = unsafe { &*(xev.data as *const _) };
let window_id = mkwid(xev.event);
let device_id = mkdid(xev.deviceid);
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
if (xev.flags & ffi::XIPointerEmulated) != 0 {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let windows = self.windows.lock();
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
if let Some(window_data) = windows.get(&WindowId(xev.event)) {
if window_data.multitouch {
// Deliver multi-touch events instead of emulated mouse events.
return;
}
} else {
return;
}
}
let modifiers = ModifiersState::from(xev.mods);
let state = if xev.evtype == ffi::XI_ButtonPress {
Pressed
} else {
Released
};
match xev.detail as u32 {
ffi::Button1 => callback(Event::WindowEvent {
window_id,
event: MouseInput {
device_id,
state,
button: Left,
modifiers,
},
}),
ffi::Button2 => callback(Event::WindowEvent {
window_id,
event: MouseInput {
device_id,
state,
button: Middle,
modifiers,
},
}),
ffi::Button3 => callback(Event::WindowEvent {
window_id,
event: MouseInput {
device_id,
state,
button: Right,
modifiers,
},
}),
// Suppress emulated scroll wheel clicks, since we handle the real motion events for those.
// In practice, even clicky scroll wheels appear to be reported by evdev (and XInput2 in
// turn) as axis motion, so we don't otherwise special-case these button presses.
4 | 5 | 6 | 7 => if xev.flags & ffi::XIPointerEmulated == 0 {
callback(Event::WindowEvent {
window_id,
event: MouseWheel {
device_id,
delta: match xev.detail {
4 => LineDelta(0.0, 1.0),
5 => LineDelta(0.0, -1.0),
6 => LineDelta(-1.0, 0.0),
7 => LineDelta(1.0, 0.0),
_ => unreachable!(),
},
phase: TouchPhase::Moved,
modifiers,
},
});
},
x => callback(Event::WindowEvent {
window_id,
event: MouseInput {
device_id,
state,
button: Other(x as u8),
modifiers,
},
}),
}
}
ffi::XI_Motion => {
let xev: &ffi::XIDeviceEvent = unsafe { &*(xev.data as *const _) };
let device_id = mkdid(xev.deviceid);
let window_id = mkwid(xev.event);
let new_cursor_pos = (xev.event_x, xev.event_y);
let modifiers = ModifiersState::from(xev.mods);
// Gymnastics to ensure self.windows isn't locked when we invoke callback
if {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let mut windows = self.windows.lock();
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
let window_data = {
if let Some(window_data) = windows.get_mut(&WindowId(xev.event)) {
window_data
} else {
return;
}
};
if Some(new_cursor_pos) != window_data.cursor_pos {
window_data.cursor_pos = Some(new_cursor_pos);
true
} else { false }
} {
callback(Event::WindowEvent {
window_id,
event: CursorMoved {
device_id,
position: new_cursor_pos,
modifiers,
},
});
}
// More gymnastics, for self.devices
let mut events = Vec::new();
{
let mask = unsafe { slice::from_raw_parts(xev.valuators.mask, xev.valuators.mask_len as usize) };
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let mut devices = self.devices.borrow_mut();
let physical_device = match devices.get_mut(&DeviceId(xev.sourceid)) {
Some(device) => device,
None => return,
};
let mut value = xev.valuators.values;
for i in 0..xev.valuators.mask_len*8 {
if ffi::XIMaskIsSet(mask, i) {
let x = unsafe { *value };
if let Some(&mut (_, ref mut info)) = physical_device.scroll_axes.iter_mut().find(|&&mut (axis, _)| axis == i) {
let delta = (x - info.position) / info.increment;
info.position = x;
events.push(Event::WindowEvent {
window_id,
event: MouseWheel {
device_id,
delta: match info.orientation {
ScrollOrientation::Horizontal => LineDelta(delta as f32, 0.0),
// X11 vertical scroll coordinates are opposite to winit's
ScrollOrientation::Vertical => LineDelta(0.0, -delta as f32),
},
phase: TouchPhase::Moved,
modifiers,
},
});
} else {
events.push(Event::WindowEvent {
window_id,
event: AxisMotion {
device_id,
axis: i as u32,
value: unsafe { *value },
},
});
}
value = unsafe { value.offset(1) };
}
}
}
for event in events {
callback(event);
}
}
ffi::XI_Enter => {
let xev: &ffi::XIEnterEvent = unsafe { &*(xev.data as *const _) };
2017-07-09 17:47:52 +10:00
let window_id = mkwid(xev.event);
let device_id = mkdid(xev.deviceid);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
if let Some(all_info) = DeviceInfo::get(&self.display, ffi::XIAllDevices) {
let mut devices = self.devices.borrow_mut();
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
for device_info in all_info.iter() {
if device_info.deviceid == xev.sourceid
// This is needed for resetting to work correctly on i3, and
// presumably some other WMs. On those, `XI_Enter` doesn't include
// the physical device ID, so both `sourceid` and `deviceid` are
// the virtual device.
|| device_info.attachment == xev.sourceid {
let device_id = DeviceId(device_info.deviceid);
if let Some(device) = devices.get_mut(&device_id) {
device.reset_scroll_position(device_info);
}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
}
2017-07-09 17:47:52 +10:00
}
}
callback(Event::WindowEvent {
window_id,
event: CursorEntered { device_id },
});
2017-07-09 17:47:52 +10:00
let new_cursor_pos = (xev.event_x, xev.event_y);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// The mods field on this event isn't actually populated, so query the
// pointer device. In the future, we can likely remove this round-trip by
// relying on Xkb for modifier values.
let modifiers = unsafe {
util::query_pointer(
&self.display,
xev.event,
xev.deviceid,
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
)
}.expect("Failed to query pointer device").get_modifier_state();
callback(Event::WindowEvent { window_id, event: CursorMoved {
device_id,
position: new_cursor_pos,
modifiers,
}})
}
ffi::XI_Leave => {
let xev: &ffi::XILeaveEvent = unsafe { &*(xev.data as *const _) };
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
// Leave, FocusIn, and FocusOut can be received by a window that's already
// been destroyed, which the user presumably doesn't want to deal with.
let window_closed = self.windows
.lock()
.get(&WindowId(xev.event))
.is_none();
if !window_closed {
callback(Event::WindowEvent {
window_id: mkwid(xev.event),
event: CursorLeft { device_id: mkdid(xev.deviceid) },
});
}
}
ffi::XI_FocusIn => {
let xev: &ffi::XIFocusInEvent = unsafe { &*(xev.data as *const _) };
let window_id = mkwid(xev.event);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
if let None = self.windows.lock().get(&WindowId(xev.event)) {
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
return;
}
self.ime
.borrow_mut()
.focus(xev.event)
.expect("Failed to focus input context");
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
callback(Event::WindowEvent { window_id, event: Focused(true) });
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
// The deviceid for this event is for a keyboard instead of a pointer,
// so we have to do a little extra work.
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let pointer_id = self.devices
.borrow()
.get(&DeviceId(xev.deviceid))
.map(|device| device.attachment)
.unwrap_or(2);
callback(Event::WindowEvent {
window_id,
event: CursorMoved {
device_id: mkdid(pointer_id),
position: (xev.event_x, xev.event_y),
modifiers: ModifiersState::from(xev.mods),
}
});
}
ffi::XI_FocusOut => {
let xev: &ffi::XIFocusOutEvent = unsafe { &*(xev.data as *const _) };
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
if let None = self.windows.lock().get(&WindowId(xev.event)) {
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
return;
}
self.ime
.borrow_mut()
.unfocus(xev.event)
.expect("Failed to unfocus input context");
x11: Destroy dropped windows; handle WM_DELETE_WINDOW (#416) Fixes #79 #414 This changes the implementation of Drop for Window to send a WM_DELETE_WINDOW ClientMessage, offloading all the cleanup and window destruction to the event loop. Unsurprisingly, this entails that the event loop now handles WM_DELETE_WINDOW using the behavior that was previously contained in Window's Drop implementation, along with destroying the Window. Not only does this mean that dropped windows are closed, but also that clicking the × button on the window actually closes it now. The previous implemention of Drop was also broken, as the event loop would be (seemingly permenanently) frozen after its invocation. That was caused specifically by the mutex locking, and is no longer an issue now that the locking is done in the event loop. While I don't have full confidence that it makes sense for the Drop implementation to behave this way, this is nonetheless a significant improvement. The previous behavior led to inconsistent state, panics, and event loop breakage, along with not actually destroying the window. This additionally makes the assumption that users don't need Focused or CursorLeft events for the destroyed window, as Closed is adequate to indicate unfocus, and users may not expect to receive events for closed/dropped windows. In my testing, those specific events were sent immediately after the window was destroyed, though this sort of behavior could be WM-specific. I've opted to explicitly suppress those events in the case of the window no longer existing.
2018-03-23 05:31:31 -04:00
callback(Event::WindowEvent {
window_id: mkwid(xev.event),
event: Focused(false),
})
}
ffi::XI_TouchBegin | ffi::XI_TouchUpdate | ffi::XI_TouchEnd => {
let xev: &ffi::XIDeviceEvent = unsafe { &*(xev.data as *const _) };
let window_id = mkwid(xev.event);
let phase = match xev.evtype {
ffi::XI_TouchBegin => TouchPhase::Started,
ffi::XI_TouchUpdate => TouchPhase::Moved,
ffi::XI_TouchEnd => TouchPhase::Ended,
_ => unreachable!()
};
callback(Event::WindowEvent {
window_id,
event: WindowEvent::Touch(Touch {
device_id: mkdid(xev.deviceid),
phase,
location: (xev.event_x, xev.event_y),
id: xev.detail as u64,
},
)})
}
ffi::XI_RawButtonPress | ffi::XI_RawButtonRelease => {
let xev: &ffi::XIRawEvent = unsafe { &*(xev.data as *const _) };
if xev.flags & ffi::XIPointerEmulated == 0 {
callback(Event::DeviceEvent { device_id: mkdid(xev.deviceid), event: DeviceEvent::Button {
2017-07-01 02:20:13 -07:00
button: xev.detail as u32,
state: match xev.evtype {
ffi::XI_RawButtonPress => Pressed,
ffi::XI_RawButtonRelease => Released,
_ => unreachable!(),
},
}});
}
}
ffi::XI_RawMotion => {
let xev: &ffi::XIRawEvent = unsafe { &*(xev.data as *const _) };
let did = mkdid(xev.deviceid);
let mask = unsafe { slice::from_raw_parts(xev.valuators.mask, xev.valuators.mask_len as usize) };
let mut value = xev.raw_values;
let mut mouse_delta = (0.0, 0.0);
let mut scroll_delta = (0.0, 0.0);
for i in 0..xev.valuators.mask_len*8 {
if ffi::XIMaskIsSet(mask, i) {
let x = unsafe { *value };
// We assume that every XInput2 device with analog axes is a pointing device emitting
// relative coordinates.
match i {
0 => mouse_delta.0 = x,
1 => mouse_delta.1 = x,
2 => scroll_delta.0 = x as f32,
3 => scroll_delta.1 = x as f32,
_ => {},
}
callback(Event::DeviceEvent { device_id: did, event: DeviceEvent::Motion {
2017-07-01 02:20:13 -07:00
axis: i as u32,
value: x,
}});
value = unsafe { value.offset(1) };
}
}
if mouse_delta != (0.0, 0.0) {
callback(Event::DeviceEvent { device_id: did, event: DeviceEvent::MouseMotion {
delta: mouse_delta,
}});
}
if scroll_delta != (0.0, 0.0) {
callback(Event::DeviceEvent { device_id: did, event: DeviceEvent::MouseWheel {
delta: LineDelta(scroll_delta.0, scroll_delta.1),
}});
}
}
ffi::XI_RawKeyPress | ffi::XI_RawKeyRelease => {
let xev: &ffi::XIRawEvent = unsafe { &*(xev.data as *const _) };
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let state = match xev.evtype {
ffi::XI_RawKeyPress => Pressed,
ffi::XI_RawKeyRelease => Released,
_ => unreachable!(),
};
let device_id = xev.sourceid;
let keycode = xev.detail;
if keycode < 8 { return; }
let scancode = (keycode - 8) as u32;
let keysym = unsafe {
(self.display.xlib.XKeycodeToKeysym)(
self.display.display,
xev.detail as ffi::KeyCode,
0,
)
};
self.display.check_errors().expect("Failed to lookup raw keysym");
let virtual_keycode = events::keysym_to_element(keysym as c_uint);
callback(Event::DeviceEvent {
device_id: mkdid(device_id),
event: DeviceEvent::Key(KeyboardInput {
scancode,
virtual_keycode,
state,
// So, in an ideal world we can use libxkbcommon to get modifiers.
// However, libxkbcommon-x11 isn't as commonly installed as one
// would hope. We can still use the Xkb extension to get
// comprehensive keyboard state updates, but interpreting that
// info manually is going to be involved.
modifiers: ModifiersState::default(),
}),
});
}
ffi::XI_HierarchyChanged => {
let xev: &ffi::XIHierarchyEvent = unsafe { &*(xev.data as *const _) };
for info in unsafe { slice::from_raw_parts(xev.info, xev.num_info as usize) } {
if 0 != info.flags & (ffi::XISlaveAdded | ffi::XIMasterAdded) {
self.init_device(info.deviceid);
callback(Event::DeviceEvent { device_id: mkdid(info.deviceid), event: DeviceEvent::Added });
} else if 0 != info.flags & (ffi::XISlaveRemoved | ffi::XIMasterRemoved) {
callback(Event::DeviceEvent { device_id: mkdid(info.deviceid), event: DeviceEvent::Removed });
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let mut devices = self.devices.borrow_mut();
devices.remove(&DeviceId(info.deviceid));
}
}
}
_ => {}
}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
},
_ => (),
}
match self.ime_receiver.try_recv() {
Ok((window_id, x, y)) => {
self.ime.borrow_mut().send_xim_spot(window_id, x, y);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
},
Err(_) => (),
}
}
fn init_device(&self, device: c_int) {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let mut devices = self.devices.borrow_mut();
if let Some(info) = DeviceInfo::get(&self.display, device) {
for info in info.iter() {
devices.insert(DeviceId(info.deviceid), Device::new(&self, info));
}
}
}
}
impl EventsLoopProxy {
pub fn wakeup(&self) -> Result<(), EventsLoopClosed> {
// Update the `EventsLoop`'s `pending_wakeup` flag.
let display = match (self.pending_wakeup.upgrade(), self.display.upgrade()) {
(Some(wakeup), Some(display)) => {
wakeup.store(true, atomic::Ordering::Relaxed);
display
},
_ => return Err(EventsLoopClosed),
};
// Push an event on the X event queue so that methods run_forever will advance.
//
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// NOTE: This design is taken from the old `WindowProxy::wakeup` implementation. It
// assumes that X11 is thread safe. Is this true?
// (WARNING: it's probably not true)
unsafe {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
util::send_client_msg(
&display,
self.wakeup_dummy_window,
self.wakeup_dummy_window,
0,
None,
(0, 0, 0, 0, 0),
)
}.flush().expect("Failed to call XSendEvent after wakeup");
Ok(())
}
}
struct DeviceInfo<'a> {
display: &'a XConnection,
info: *const ffi::XIDeviceInfo,
count: usize,
}
impl<'a> DeviceInfo<'a> {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
fn get(display: &'a XConnection, device: c_int) -> Option<Self> {
unsafe {
let mut count = mem::uninitialized();
let info = (display.xinput2.XIQueryDevice)(display.display, device, &mut count);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
display.check_errors()
.ok()
.and_then(|_| {
if info.is_null() || count == 0 {
None
} else {
Some(DeviceInfo {
display,
info,
count: count as usize,
})
}
})
}
}
}
impl<'a> Drop for DeviceInfo<'a> {
fn drop(&mut self) {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
assert!(!self.info.is_null());
unsafe { (self.display.xinput2.XIFreeDeviceInfo)(self.info as *mut _) };
}
}
impl<'a> ::std::ops::Deref for DeviceInfo<'a> {
type Target = [ffi::XIDeviceInfo];
fn deref(&self) -> &Self::Target {
unsafe { slice::from_raw_parts(self.info, self.count) }
}
}
#[derive(Debug, Copy, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct WindowId(ffi::Window);
#[derive(Debug, Copy, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct DeviceId(c_int);
pub struct Window {
pub window: Arc<Window2>,
display: Weak<XConnection>,
windows: Weak<Mutex<HashMap<WindowId, WindowData>>>,
ime_sender: Mutex<ImeSender>,
}
impl ::std::ops::Deref for Window {
type Target = Window2;
#[inline]
fn deref(&self) -> &Window2 {
&*self.window
}
}
impl Window {
pub fn new(
x_events_loop: &EventsLoop,
2018-05-07 17:36:21 -04:00
attribs: WindowAttributes,
pl_attribs: PlatformSpecificWindowBuilderAttributes
) -> Result<Self, CreationError> {
2018-05-07 17:36:21 -04:00
let multitouch = attribs.multitouch;
let win = Arc::new(Window2::new(&x_events_loop, attribs, pl_attribs)?);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
x_events_loop.shared_state
.borrow_mut()
.insert(win.id(), Arc::downgrade(&win.shared_state));
x_events_loop.ime
.borrow_mut()
.create_context(win.id().0)
.expect("Failed to create input context");
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
x_events_loop.windows.lock().insert(win.id(), WindowData {
config: Default::default(),
2018-05-07 17:36:21 -04:00
multitouch,
cursor_pos: None,
});
Ok(Window {
window: win,
windows: Arc::downgrade(&x_events_loop.windows),
display: Arc::downgrade(&x_events_loop.display),
ime_sender: Mutex::new(x_events_loop.ime_sender.clone()),
})
}
#[inline]
pub fn id(&self) -> WindowId {
self.window.id()
}
#[inline]
pub fn send_xim_spot(&self, x: i16, y: i16) {
let _ = self.ime_sender
.lock()
.send((self.window.id().0, x, y));
}
}
impl Drop for Window {
fn drop(&mut self) {
if let (Some(windows), Some(display)) = (self.windows.upgrade(), self.display.upgrade()) {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
if let Some(_) = windows.lock().remove(&self.window.id()) {
unsafe {
(display.xlib.XDestroyWindow)(display.display, self.window.id().0);
}
}
}
}
}
/// State maintained for translating window-related events
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
#[derive(Debug)]
struct WindowData {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
config: WindowConfig,
multitouch: bool,
cursor_pos: Option<(f64, f64)>,
}
// Required by ffi members
unsafe impl Send for WindowData {}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
#[derive(Debug, Default)]
struct WindowConfig {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
pub size: Option<(c_int, c_int)>,
pub position: Option<(c_int, c_int)>,
pub inner_position: Option<(c_int, c_int)>,
}
/// XEvents of type GenericEvent store their actual data in an XGenericEventCookie data structure. This is a wrapper to
/// extract the cookie from a GenericEvent XEvent and release the cookie data once it has been processed
struct GenericEventCookie<'a> {
display: &'a XConnection,
cookie: ffi::XGenericEventCookie
}
impl<'a> GenericEventCookie<'a> {
fn from_event<'b>(display: &'b XConnection, event: ffi::XEvent) -> Option<GenericEventCookie<'b>> {
unsafe {
let mut cookie: ffi::XGenericEventCookie = From::from(event);
if (display.xlib.XGetEventData)(display.display, &mut cookie) == ffi::True {
Some(GenericEventCookie{display: display, cookie: cookie})
} else {
None
}
}
}
}
impl<'a> Drop for GenericEventCookie<'a> {
fn drop(&mut self) {
unsafe {
let xlib = &self.display.xlib;
(xlib.XFreeEventData)(self.display.display, &mut self.cookie);
}
}
}
#[derive(Debug, Copy, Clone)]
struct XExtension {
opcode: c_int,
first_event_id: c_int,
first_error_id: c_int,
}
fn mkwid(w: ffi::Window) -> ::WindowId { ::WindowId(::platform::WindowId::X(WindowId(w))) }
fn mkdid(w: c_int) -> ::DeviceId { ::DeviceId(::platform::DeviceId::X(DeviceId(w))) }
#[derive(Debug)]
struct Device {
name: String,
scroll_axes: Vec<(i32, ScrollAxis)>,
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// For master devices, this is the paired device (pointer <-> keyboard).
// For slave devices, this is the master.
attachment: c_int,
}
#[derive(Debug, Copy, Clone)]
struct ScrollAxis {
increment: f64,
orientation: ScrollOrientation,
position: f64,
}
#[derive(Debug, Copy, Clone)]
enum ScrollOrientation {
Vertical,
Horizontal,
}
impl Device {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
fn new(el: &EventsLoop, info: &ffi::XIDeviceInfo) -> Self {
let name = unsafe { CStr::from_ptr(info.name).to_string_lossy() };
2017-07-09 17:47:52 +10:00
let mut scroll_axes = Vec::new();
2017-07-09 17:47:52 +10:00
if Device::physical_device(info) {
// Register for global raw events
let mask = ffi::XI_RawMotionMask
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
| ffi::XI_RawButtonPressMask
| ffi::XI_RawButtonReleaseMask
| ffi::XI_RawKeyPressMask
| ffi::XI_RawKeyReleaseMask;
unsafe {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
util::select_xinput_events(
&el.display,
el.root,
info.deviceid,
mask,
)
}.queue(); // The request buffer is flushed when we poll for events
// Identify scroll axes
2017-07-09 17:47:52 +10:00
for class_ptr in Device::classes(info) {
let class = unsafe { &**class_ptr };
match class._type {
ffi::XIScrollClass => {
let info = unsafe { mem::transmute::<&ffi::XIAnyClassInfo, &ffi::XIScrollClassInfo>(class) };
scroll_axes.push((info.number, ScrollAxis {
increment: info.increment,
orientation: match info.scroll_type {
ffi::XIScrollTypeHorizontal => ScrollOrientation::Horizontal,
ffi::XIScrollTypeVertical => ScrollOrientation::Vertical,
_ => { unreachable!() }
},
position: 0.0,
}));
}
_ => {}
}
}
2017-07-09 17:47:52 +10:00
}
let mut device = Device {
name: name.into_owned(),
scroll_axes: scroll_axes,
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
attachment: info.attachment,
2017-07-09 17:47:52 +10:00
};
device.reset_scroll_position(info);
device
}
fn reset_scroll_position(&mut self, info: &ffi::XIDeviceInfo) {
if Device::physical_device(info) {
for class_ptr in Device::classes(info) {
let class = unsafe { &**class_ptr };
match class._type {
ffi::XIValuatorClass => {
let info = unsafe { mem::transmute::<&ffi::XIAnyClassInfo, &ffi::XIValuatorClassInfo>(class) };
2017-07-09 17:47:52 +10:00
if let Some(&mut (_, ref mut axis)) = self.scroll_axes.iter_mut().find(|&&mut (axis, _)| axis == info.number) {
axis.position = info.value;
}
}
_ => {}
}
}
}
2017-07-09 17:47:52 +10:00
}
2017-07-09 17:47:52 +10:00
#[inline]
fn physical_device(info: &ffi::XIDeviceInfo) -> bool {
info._use == ffi::XISlaveKeyboard || info._use == ffi::XISlavePointer || info._use == ffi::XIFloatingSlave
}
#[inline]
fn classes(info: &ffi::XIDeviceInfo) -> &[*const ffi::XIAnyClassInfo] {
unsafe { slice::from_raw_parts(info.classes as *const *const ffi::XIAnyClassInfo, info.num_classes as usize) }
}
}