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

584 lines
20 KiB
Rust
Raw Normal View History

#![cfg(any(target_os = "linux", target_os = "dragonfly", target_os = "freebsd", target_os = "netbsd", 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;
pub mod util;
mod event_processor;
2018-05-07 17:36:21 -04:00
Event Loop 2.0 API and Windows implementation (#638) * Rename EventsLoop and associated types to EventLoop * Rename WindowEvent::Refresh to WindowEvent::Redraw * Remove second thread from win32 backend * Update run_forever to hijack thread * Replace windows Mutex with parking_lot Mutex * Implement new ControlFlow and associated events * Add StartCause::Init support, timer example * Add ability to send custom user events * Fully invert windows control flow so win32 calls into winit's callback * Add request_redraw * Rename platform to platform_impl * Rename os to platform, add Ext trait postfixes * Add platform::desktop module with EventLoopExt::run_return * Re-organize into module structure * Improve documentation * Small changes to examples * Improve docs for run and run_return * Change instances of "events_loop" to "event_loop" * Rename MonitorId to MonitorHandle * Add CHANGELOG entry * Improve WaitUntil timer precision * When SendEvent is called during event closure, buffer events * Fix resize lag when waiting in some situations * Update send test and errors that broke some examples/APIs * Improve clarity/fix typos in docs * Fix unreachable panic after setting ControlFlow to Poll during some RedrawRequested events. * Fix crash when running in release mode * Remove crossbeam dependency and make drop events work again * Remove serde implementations from ControlFlow * Fix 1.24.1 build * Fix freeze when setting decorations * Replace &EventLoop in callback with &EventLoopWindowTarget * Document and implement Debug for EventLoopWindowTarget * Fix some deadlocks that could occur when changing window state * Fix thread executor not executing closure when called from non-loop thread * Fix buffered events not getting dispatched * Fix crash with runner refcell not getting dropped * Address review feedback * Fix CHANGELOG typo * Catch panics in user callback
2019-02-05 10:30:33 -05:00
pub use self::monitor::MonitorHandle;
2018-05-29 07:48:47 -04:00
pub use self::window::UnownedWindow;
2015-12-24 10:57:08 +01:00
pub use self::xdisplay::{XConnection, XNotSupported, XError};
use std::{mem, slice};
use std::cell::RefCell;
use std::collections::{VecDeque, HashMap, HashSet};
use std::ffi::CStr;
2018-05-29 07:48:47 -04:00
use std::ops::Deref;
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 std::rc::Rc;
use std::sync::{Arc, mpsc, Weak, Mutex};
use libc::{self, setlocale, LC_CTYPE};
use event_loop::{ControlFlow, EventLoopClosed, EventLoopWindowTarget as RootELW};
use event::{WindowEvent, Event};
Event Loop 2.0 API and Windows implementation (#638) * Rename EventsLoop and associated types to EventLoop * Rename WindowEvent::Refresh to WindowEvent::Redraw * Remove second thread from win32 backend * Update run_forever to hijack thread * Replace windows Mutex with parking_lot Mutex * Implement new ControlFlow and associated events * Add StartCause::Init support, timer example * Add ability to send custom user events * Fully invert windows control flow so win32 calls into winit's callback * Add request_redraw * Rename platform to platform_impl * Rename os to platform, add Ext trait postfixes * Add platform::desktop module with EventLoopExt::run_return * Re-organize into module structure * Improve documentation * Small changes to examples * Improve docs for run and run_return * Change instances of "events_loop" to "event_loop" * Rename MonitorId to MonitorHandle * Add CHANGELOG entry * Improve WaitUntil timer precision * When SendEvent is called during event closure, buffer events * Fix resize lag when waiting in some situations * Update send test and errors that broke some examples/APIs * Improve clarity/fix typos in docs * Fix unreachable panic after setting ControlFlow to Poll during some RedrawRequested events. * Fix crash when running in release mode * Remove crossbeam dependency and make drop events work again * Remove serde implementations from ControlFlow * Fix 1.24.1 build * Fix freeze when setting decorations * Replace &EventLoop in callback with &EventLoopWindowTarget * Document and implement Debug for EventLoopWindowTarget * Fix some deadlocks that could occur when changing window state * Fix thread executor not executing closure when called from non-loop thread * Fix buffered events not getting dispatched * Fix crash with runner refcell not getting dropped * Address review feedback * Fix CHANGELOG typo * Catch panics in user callback
2019-02-05 10:30:33 -05:00
use platform_impl::PlatformSpecificWindowBuilderAttributes;
use platform_impl::platform::sticky_exit_callback;
use window::{CreationError, WindowAttributes};
use self::dnd::{Dnd, DndState};
use self::ime::{ImeReceiver, ImeSender, ImeCreationError, Ime};
use self::event_processor::EventProcessor;
pub struct EventLoopWindowTarget<T> {
2018-05-29 07:48:47 -04:00
xconn: Arc<XConnection>,
wm_delete_window: ffi::Atom,
ime_sender: ImeSender,
root: ffi::Window,
ime: RefCell<Ime>,
2018-05-29 07:48:47 -04:00
windows: RefCell<HashMap<WindowId, Weak<UnownedWindow>>>,
pending_redraws: Arc<Mutex<HashSet<WindowId>>>,
_marker: ::std::marker::PhantomData<T>
}
pub struct EventLoop<T: 'static> {
inner_loop: ::calloop::EventLoop<()>,
_x11_source: ::calloop::Source<::calloop::generic::Generic<::calloop::generic::EventedRawFd>>,
_user_source: ::calloop::Source<::calloop::channel::Channel<T>>,
pending_user_events: Rc<RefCell<VecDeque<T>>>,
user_sender: ::calloop::channel::Sender<T>,
pending_events: Rc<RefCell<VecDeque<Event<T>>>>,
target: Rc<RootELW<T>>
}
2017-10-25 11:03:57 -07:00
#[derive(Clone)]
pub struct EventLoopProxy<T: 'static> {
user_sender: ::calloop::channel::Sender<T>,
}
impl<T: 'static> EventLoop<T> {
pub fn new(xconn: Arc<XConnection>) -> EventLoop<T> {
2018-05-29 07:48:47 -04:00
let root = unsafe { (xconn.xlib.XDefaultRootWindow)(xconn.display) };
2018-05-29 07:48:47 -04:00
let wm_delete_window = unsafe { xconn.get_atom_unchecked(b"WM_DELETE_WINDOW\0") };
2018-05-29 07:48:47 -04:00
let dnd = Dnd::new(Arc::clone(&xconn))
.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({
2018-05-29 07:48:47 -04:00
let result = Ime::new(Arc::clone(&xconn));
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")
});
2018-06-14 19:42:18 -04:00
let randr_event_offset = xconn.select_xrandr_input(root)
.expect("Failed to query XRandR extension");
let xi2ext = unsafe {
let mut result = XExtension {
opcode: mem::uninitialized(),
first_event_id: mem::uninitialized(),
first_error_id: mem::uninitialized(),
};
2018-05-29 07:48:47 -04:00
let res = (xconn.xlib.XQueryExtension)(
xconn.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;
2018-05-29 07:48:47 -04:00
if (xconn.xinput2.XIQueryVersion)(
xconn.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
&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,
);
}
}
2018-05-29 07:48:47 -04:00
xconn.update_cached_wm_info(root);
let target = Rc::new(RootELW{
p: super::EventLoopWindowTarget::X(EventLoopWindowTarget {
ime,
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
root,
windows: Default::default(),
_marker: ::std::marker::PhantomData,
ime_sender,
xconn,
wm_delete_window,
pending_redraws: Default::default(),
}),
_marker: ::std::marker::PhantomData
});
// A calloop event loop to drive us
let inner_loop = ::calloop::EventLoop::new().unwrap();
// Handle user events
let pending_user_events = Rc::new(RefCell::new(VecDeque::new()));
let pending_user_events2 = pending_user_events.clone();
let (user_sender, user_channel) = ::calloop::channel::channel();
let _user_source = inner_loop.handle().insert_source(user_channel, move |evt, &mut()| {
if let ::calloop::channel::Event::Msg(msg) = evt {
pending_user_events2.borrow_mut().push_back(msg);
}
}).unwrap();
// Handle X11 events
let pending_events: Rc<RefCell<VecDeque<_>>> = Default::default();
let mut processor = EventProcessor {
target: target.clone(),
dnd,
2018-05-29 07:48:47 -04:00
devices: Default::default(),
randr_event_offset,
ime_receiver,
xi2ext,
};
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
2018-05-27 08:49:35 -04:00
// (The request buffer is flushed during `init_device`)
get_xtarget(&target).xconn.select_xinput_events(
2018-05-29 07:48:47 -04:00
root,
ffi::XIAllDevices,
ffi::XI_HierarchyChangedMask,
).queue();
processor.init_device(ffi::XIAllDevices);
// Setup the X11 event source
let mut x11_events = ::calloop::generic::Generic::from_raw_fd(get_xtarget(&target).xconn.x11_fd);
x11_events.set_interest(::calloop::mio::Ready::readable());
let _x11_source = inner_loop.handle().insert_source(
x11_events,
{
let pending_events = pending_events.clone();
let mut callback = move |event| {
pending_events.borrow_mut().push_back(event);
};
move |evt, &mut ()| {
if evt.readiness.is_readable() {
// process all pending events
let mut xev = unsafe { mem::uninitialized() };
while unsafe { processor.poll_one_event(&mut xev) } {
processor.process_event(&mut xev, &mut callback);
}
}
}
}
).unwrap();
let result = EventLoop {
inner_loop,
pending_events,
_x11_source,
_user_source,
user_sender,
pending_user_events,
target
};
result
}
/// Returns the `XConnection` of this events loop.
#[inline]
pub fn x_connection(&self) -> &Arc<XConnection> {
&get_xtarget(&self.target).xconn
}
pub fn create_proxy(&self) -> EventLoopProxy<T> {
Event Loop 2.0 API and Windows implementation (#638) * Rename EventsLoop and associated types to EventLoop * Rename WindowEvent::Refresh to WindowEvent::Redraw * Remove second thread from win32 backend * Update run_forever to hijack thread * Replace windows Mutex with parking_lot Mutex * Implement new ControlFlow and associated events * Add StartCause::Init support, timer example * Add ability to send custom user events * Fully invert windows control flow so win32 calls into winit's callback * Add request_redraw * Rename platform to platform_impl * Rename os to platform, add Ext trait postfixes * Add platform::desktop module with EventLoopExt::run_return * Re-organize into module structure * Improve documentation * Small changes to examples * Improve docs for run and run_return * Change instances of "events_loop" to "event_loop" * Rename MonitorId to MonitorHandle * Add CHANGELOG entry * Improve WaitUntil timer precision * When SendEvent is called during event closure, buffer events * Fix resize lag when waiting in some situations * Update send test and errors that broke some examples/APIs * Improve clarity/fix typos in docs * Fix unreachable panic after setting ControlFlow to Poll during some RedrawRequested events. * Fix crash when running in release mode * Remove crossbeam dependency and make drop events work again * Remove serde implementations from ControlFlow * Fix 1.24.1 build * Fix freeze when setting decorations * Replace &EventLoop in callback with &EventLoopWindowTarget * Document and implement Debug for EventLoopWindowTarget * Fix some deadlocks that could occur when changing window state * Fix thread executor not executing closure when called from non-loop thread * Fix buffered events not getting dispatched * Fix crash with runner refcell not getting dropped * Address review feedback * Fix CHANGELOG typo * Catch panics in user callback
2019-02-05 10:30:33 -05:00
EventLoopProxy {
user_sender: self.user_sender.clone(),
}
}
pub(crate) fn window_target(&self) -> &RootELW<T> {
&self.target
}
pub fn run_return<F>(&mut self, mut callback: F)
where F: FnMut(Event<T>, &RootELW<T>, &mut ControlFlow)
{
let mut control_flow = ControlFlow::default();
let wt = get_xtarget(&self.target);
loop {
// Empty the event buffer
{
let mut guard = self.pending_events.borrow_mut();
for evt in guard.drain(..) {
sticky_exit_callback(evt, &self.target, &mut control_flow, &mut callback);
}
}
// Empty the user event buffer
{
let mut guard = self.pending_user_events.borrow_mut();
for evt in guard.drain(..) {
sticky_exit_callback(
::event::Event::UserEvent(evt),
&self.target,
&mut control_flow,
&mut callback
);
}
}
// Empty the redraw requests
{
let mut guard = wt.pending_redraws.lock().unwrap();
for wid in guard.drain() {
sticky_exit_callback(
Event::WindowEvent {
window_id: ::window::WindowId(super::WindowId::X(wid)),
event: WindowEvent::RedrawRequested
},
&self.target,
&mut control_flow,
&mut callback
);
}
}
// send Events cleared
{
sticky_exit_callback(
::event::Event::EventsCleared,
&self.target,
&mut control_flow,
&mut callback
);
}
// flush the X11 connection
unsafe { (wt.xconn.xlib.XFlush)(wt.xconn.display); }
match control_flow {
ControlFlow::Exit => break,
ControlFlow::Poll => {
// non-blocking dispatch
self.inner_loop.dispatch(Some(::std::time::Duration::from_millis(0)), &mut ()).unwrap();
control_flow = ControlFlow::default();
callback(::event::Event::NewEvents(::event::StartCause::Poll), &self.target, &mut control_flow);
},
ControlFlow::Wait => {
self.inner_loop.dispatch(None, &mut ()).unwrap();
control_flow = ControlFlow::default();
callback(
::event::Event::NewEvents(::event::StartCause::WaitCancelled {
start: ::std::time::Instant::now(),
requested_resume: None
}),
&self.target,
&mut control_flow
);
},
ControlFlow::WaitUntil(deadline) => {
let start = ::std::time::Instant::now();
// compute the blocking duration
let duration = deadline.duration_since(::std::cmp::max(deadline, start));
self.inner_loop.dispatch(Some(duration), &mut ()).unwrap();
control_flow = ControlFlow::default();
let now = std::time::Instant::now();
if now < deadline {
callback(
::event::Event::NewEvents(::event::StartCause::WaitCancelled {
start,
requested_resume: Some(deadline)
}),
&self.target,
&mut control_flow
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
);
} else {
callback(
::event::Event::NewEvents(::event::StartCause::ResumeTimeReached {
start,
requested_resume: deadline
}),
&self.target,
&mut control_flow
2018-06-14 19:42:18 -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
}
}
2018-05-29 07:48:47 -04:00
callback(::event::Event::LoopDestroyed, &self.target, &mut control_flow);
2018-05-29 07:48:47 -04:00
}
pub fn run<F>(mut self, callback: F) -> !
where F: 'static + FnMut(Event<T>, &RootELW<T>, &mut ControlFlow)
{
self.run_return(callback);
::std::process::exit(0);
2018-05-29 07:48:47 -04:00
}
}
fn get_xtarget<T>(rt: &RootELW<T>) -> &EventLoopWindowTarget<T> {
if let super::EventLoopWindowTarget::X(ref target) = rt.p {
target
} else {
unreachable!();
}
}
impl<T: 'static> EventLoopProxy<T> {
pub fn send_event(&self, event: T) -> Result<(), EventLoopClosed> {
self.user_sender.send(event).map_err(|_| EventLoopClosed)
}
}
struct DeviceInfo<'a> {
2018-05-29 07:48:47 -04:00
xconn: &'a XConnection,
info: *const ffi::XIDeviceInfo,
count: usize,
}
impl<'a> DeviceInfo<'a> {
2018-05-29 07:48:47 -04:00
fn get(xconn: &'a XConnection, device: c_int) -> Option<Self> {
unsafe {
let mut count = mem::uninitialized();
2018-05-29 07:48:47 -04:00
let info = (xconn.xinput2.XIQueryDevice)(xconn.display, device, &mut count);
xconn.check_errors()
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
.ok()
.and_then(|_| {
if info.is_null() || count == 0 {
None
} else {
Some(DeviceInfo {
2018-05-29 07:48:47 -04:00
xconn,
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
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());
2018-05-29 07:48:47 -04:00
unsafe { (self.xconn.xinput2.XIFreeDeviceInfo)(self.info as *mut _) };
}
}
2018-05-29 07:48:47 -04:00
impl<'a> 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);
impl WindowId {
pub unsafe fn dummy() -> Self {
WindowId(0)
}
}
#[derive(Debug, Copy, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct DeviceId(c_int);
impl DeviceId {
pub unsafe fn dummy() -> Self {
DeviceId(0)
}
}
2018-06-14 19:42:18 -04:00
pub struct Window(Arc<UnownedWindow>);
2018-05-29 07:48:47 -04:00
impl Deref for Window {
type Target = UnownedWindow;
#[inline]
2018-05-29 07:48:47 -04:00
fn deref(&self) -> &UnownedWindow {
2018-06-14 19:42:18 -04:00
&*self.0
}
}
impl Window {
pub fn new<T>(
event_loop: &EventLoopWindowTarget<T>,
2018-05-07 17:36:21 -04:00
attribs: WindowAttributes,
pl_attribs: PlatformSpecificWindowBuilderAttributes
) -> Result<Self, CreationError> {
2018-05-29 07:48:47 -04:00
let window = Arc::new(UnownedWindow::new(&event_loop, attribs, pl_attribs)?);
event_loop.windows
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
.borrow_mut()
2018-05-29 07:48:47 -04:00
.insert(window.id(), Arc::downgrade(&window));
2018-06-14 19:42:18 -04:00
Ok(Window(window))
}
}
impl Drop for Window {
fn drop(&mut self) {
2018-06-14 19:42:18 -04:00
let window = self.deref();
let xconn = &window.xconn;
2018-05-29 07:48:47 -04:00
unsafe {
2018-06-14 19:42:18 -04:00
(xconn.xlib.XDestroyWindow)(xconn.display, window.id().0);
2018-05-29 07:48:47 -04:00
// If the window was somehow already destroyed, we'll get a `BadWindow` error, which we don't care about.
let _ = xconn.check_errors();
}
}
}
/// 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> {
2018-05-29 07:48:47 -04:00
xconn: &'a XConnection,
cookie: ffi::XGenericEventCookie
}
impl<'a> GenericEventCookie<'a> {
2018-05-29 07:48:47 -04:00
fn from_event<'b>(xconn: &'b XConnection, event: ffi::XEvent) -> Option<GenericEventCookie<'b>> {
unsafe {
let mut cookie: ffi::XGenericEventCookie = From::from(event);
2018-05-29 07:48:47 -04:00
if (xconn.xlib.XGetEventData)(xconn.display, &mut cookie) == ffi::True {
Some(GenericEventCookie { xconn, cookie })
} else {
None
}
}
}
}
impl<'a> Drop for GenericEventCookie<'a> {
fn drop(&mut self) {
unsafe {
2018-05-29 07:48:47 -04:00
(self.xconn.xlib.XFreeEventData)(self.xconn.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) -> ::window::WindowId { ::window::WindowId(::platform_impl::WindowId::X(WindowId(w))) }
fn mkdid(w: c_int) -> ::event::DeviceId { ::event::DeviceId(::platform_impl::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 {
fn new<T: 'static>(el: &EventProcessor<T>, 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();
let wt = get_xtarget(&el.target);
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;
2018-05-27 08:49:35 -04:00
// The request buffer is flushed when we poll for events
wt.xconn.select_xinput_events(wt.root, info.deviceid, mask).queue();
// 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) }
}
}