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

1508 lines
55 KiB
Rust
Raw Normal View History

use raw_window_handle::XlibHandle;
use std::{
cmp, env,
ffi::CString,
mem::{self, replace, MaybeUninit},
os::raw::*,
path::Path,
ptr, slice,
sync::Arc,
};
use x11_dl::xlib::TrueColor;
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
2018-05-07 17:36:21 -04:00
use libc;
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
use parking_lot::Mutex;
use crate::{
dpi::{PhysicalPosition, PhysicalSize, Position, Size},
error::{ExternalError, NotSupportedError, OsError as RootOsError},
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
monitor::{MonitorHandle as RootMonitorHandle, VideoMode as RootVideoMode},
platform_impl::{
x11::{ime::ImeContextCreationError, MonitorHandle as X11MonitorHandle},
MonitorHandle as PlatformMonitorHandle, OsError, PlatformSpecificWindowBuilderAttributes,
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
VideoMode as PlatformVideoMode,
},
window::{CursorIcon, Fullscreen, Icon, UserAttentionType, WindowAttributes},
};
Move fullscreen modes to not touch physical resolutions (#270) * Fix X11 screen resolution change using XrandR The previous XF86 resolution switching was broken and everything seems to have moved on to xrandr. Use that instead while cleaning up the code a bit as well. * Use XRandR for actual multiscreen support in X11 * Use actual monitor names in X11 * Get rid of ptr::read usage in X11 * Use a bog standard Vec instead of VecDeque * Get rid of the XRandR mode switching stuff Wayland has made the decision that apps shouldn't change screen resolutions and just take the screens as they've been setup. In the modern world where GPU scaling is cheap and LCD panels are scaling anyway it makes no sense to make "physical" resolution changes when software should be taking care of it. This massively simplifies the code and makes it easier to extend to more niche setups like MST and videowalls. * Rename fullscreen options to match new semantics * Implement XRandR 1.5 support * Get rid of the FullScreen enum Moving to just having two states None and Some(MonitorId) and then being able to set full screen in the current monitor with something like: window.set_fullscreen(Some(window.current_monitor())); * Implement Window::get_current_monitor() Do it by iterating over the available monitors and finding which has the biggest overlap with the window. For this MonitorId needs a new get_position() that needs to be implemented for all platforms. * Add unimplemented get_position() to all MonitorId * Make get_current_monitor() platform specific * Add unimplemented get_current_monitor() to all * Implement proper primary monitor selection in X11 * Shut up some warnings * Remove libxxf86vm package from travis Since we're no longer using XF86 there's no need to keep the package around for CI. * Don't use new struct syntax * Fix indentation * Adjust Android/iOS fullscreen/maximized On Android and iOS we can assume single screen apps that are already fullscreen and maximized so there are a few methods that are implemented by just returning a fixed value or not doing anything. * Mark OSX/Win fullscreen/maximized unimplemented()! These would be safe as no-ops but we should make it explicit so there is more of an incentive to actually implement them.
2017-09-07 09:33:46 +01:00
use super::{
ffi, util, EventLoopWindowTarget, ImeRequest, ImeSender, WakeSender, WindowId, XConnection,
XError,
};
#[derive(Debug)]
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
pub struct SharedState {
2018-05-29 07:48:47 -04:00
pub cursor_pos: Option<(f64, f64)>,
pub size: Option<(u32, u32)>,
pub position: Option<(i32, i32)>,
pub inner_position: Option<(i32, i32)>,
pub inner_position_rel_parent: Option<(i32, i32)>,
pub is_resizable: bool,
pub is_decorated: bool,
pub last_monitor: X11MonitorHandle,
pub dpi_adjusted: Option<(u32, u32)>,
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
pub fullscreen: Option<Fullscreen>,
// Set when application calls `set_fullscreen` when window is not visible
pub desired_fullscreen: Option<Option<Fullscreen>>,
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
// Used to restore position after exiting fullscreen
2018-06-14 19:42:18 -04:00
pub restore_position: Option<(i32, i32)>,
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
// Used to restore video mode after exiting fullscreen
pub desktop_video_mode: Option<(ffi::RRCrtc, ffi::RRMode)>,
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
pub frame_extents: Option<util::FrameExtentsHeuristic>,
pub min_inner_size: Option<Size>,
pub max_inner_size: Option<Size>,
pub resize_increments: Option<Size>,
pub base_size: Option<Size>,
pub visibility: Visibility,
}
#[derive(Copy, Clone, Debug, Eq, PartialEq)]
pub enum Visibility {
No,
Yes,
// Waiting for VisibilityNotify
YesWait,
2018-06-14 19:42:18 -04:00
}
impl SharedState {
fn new(last_monitor: X11MonitorHandle, window_attributes: &WindowAttributes) -> Mutex<Self> {
let visibility = if window_attributes.visible {
Visibility::YesWait
} else {
Visibility::No
};
Mutex::new(SharedState {
last_monitor,
visibility,
is_resizable: window_attributes.resizable,
is_decorated: window_attributes.decorations,
cursor_pos: None,
size: None,
position: None,
inner_position: None,
inner_position_rel_parent: None,
dpi_adjusted: None,
fullscreen: None,
desired_fullscreen: None,
restore_position: None,
desktop_video_mode: None,
frame_extents: None,
min_inner_size: None,
max_inner_size: None,
resize_increments: None,
base_size: None,
})
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
unsafe impl Send for UnownedWindow {}
unsafe impl Sync for UnownedWindow {}
pub struct UnownedWindow {
pub xconn: Arc<XConnection>, // never changes
xwindow: ffi::Window, // never changes
root: ffi::Window, // never changes
screen_id: i32, // never changes
cursor: Mutex<CursorIcon>,
cursor_grabbed: Mutex<bool>,
cursor_visible: Mutex<bool>,
2018-06-14 19:42:18 -04:00
ime_sender: Mutex<ImeSender>,
2018-05-29 07:48:47 -04:00
pub shared_state: Mutex<SharedState>,
redraw_sender: WakeSender<WindowId>,
}
2018-05-29 07:48:47 -04:00
impl UnownedWindow {
pub fn new<T>(
event_loop: &EventLoopWindowTarget<T>,
2018-05-07 17:36:21 -04:00
window_attrs: WindowAttributes,
pl_attribs: PlatformSpecificWindowBuilderAttributes,
) -> Result<UnownedWindow, RootOsError> {
2018-05-29 07:48:47 -04:00
let xconn = &event_loop.xconn;
let root = event_loop.root;
let mut monitors = xconn.available_monitors();
let guessed_monitor = if monitors.is_empty() {
X11MonitorHandle::dummy()
} else {
xconn
.query_pointer(root, util::VIRTUAL_CORE_POINTER)
.ok()
.and_then(|pointer_state| {
let (x, y) = (pointer_state.root_x as i64, pointer_state.root_y as i64);
for i in 0..monitors.len() {
if monitors[i].rect.contains_point(x, y) {
return Some(monitors.swap_remove(i));
}
}
None
})
.unwrap_or_else(|| monitors.swap_remove(0))
};
let scale_factor = guessed_monitor.scale_factor();
info!("Guessed window scale factor: {}", scale_factor);
let max_inner_size: Option<(u32, u32)> = window_attrs
.max_inner_size
.map(|size| size.to_physical::<u32>(scale_factor).into());
let min_inner_size: Option<(u32, u32)> = window_attrs
.min_inner_size
.map(|size| size.to_physical::<u32>(scale_factor).into());
2018-06-14 19:42:18 -04:00
let position = window_attrs
.position
.map(|position| position.to_physical::<i32>(scale_factor));
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let dimensions = {
// x11 only applies constraints when the window is actively resized
// by the user, so we have to manually apply the initial constraints
let mut dimensions: (u32, u32) = window_attrs
.inner_size
.map(|size| size.to_physical::<u32>(scale_factor))
.or_else(|| Some((800, 600).into()))
2018-06-14 19:42:18 -04:00
.map(Into::into)
.unwrap();
if let Some(max) = max_inner_size {
dimensions.0 = cmp::min(dimensions.0, max.0);
dimensions.1 = cmp::min(dimensions.1, max.1);
}
if let Some(min) = min_inner_size {
dimensions.0 = cmp::max(dimensions.0, min.0);
dimensions.1 = cmp::max(dimensions.1, min.1);
}
debug!(
"Calculated physical dimensions: {}x{}",
dimensions.0, dimensions.1
);
dimensions
};
let screen_id = match pl_attribs.screen_id {
Some(id) => id,
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
None => unsafe { (xconn.xlib.XDefaultScreen)(xconn.display) },
};
// creating
let (visual, depth, require_colormap) = match pl_attribs.visual_infos {
Some(vi) => (vi.visual, vi.depth, false),
None if window_attrs.transparent => {
// Find a suitable visual
let mut vinfo = MaybeUninit::uninit();
let vinfo_initialized = unsafe {
(xconn.xlib.XMatchVisualInfo)(
xconn.display,
screen_id,
32,
TrueColor,
vinfo.as_mut_ptr(),
) != 0
};
if vinfo_initialized {
let vinfo = unsafe { vinfo.assume_init() };
(vinfo.visual, vinfo.depth, true)
} else {
debug!("Could not set transparency, because XMatchVisualInfo returned zero for the required parameters");
(
ffi::CopyFromParent as *mut ffi::Visual,
ffi::CopyFromParent,
false,
)
}
}
_ => (
ffi::CopyFromParent as *mut ffi::Visual,
ffi::CopyFromParent,
false,
),
};
let mut set_win_attr = {
let mut swa: ffi::XSetWindowAttributes = unsafe { mem::zeroed() };
swa.colormap = if let Some(vi) = pl_attribs.visual_infos {
unsafe {
let visual = vi.visual;
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
(xconn.xlib.XCreateColormap)(xconn.display, root, visual, ffi::AllocNone)
}
} else if require_colormap {
unsafe { (xconn.xlib.XCreateColormap)(xconn.display, root, visual, ffi::AllocNone) }
} else {
0
};
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
swa.event_mask = ffi::ExposureMask
| ffi::StructureNotifyMask
| ffi::VisibilityChangeMask
| ffi::KeyPressMask
| ffi::KeyReleaseMask
| ffi::KeymapStateMask
| ffi::ButtonPressMask
| ffi::ButtonReleaseMask
| ffi::PointerMotionMask;
swa.border_pixel = 0;
swa.override_redirect = pl_attribs.override_redirect as c_int;
swa
};
let mut window_attributes = ffi::CWBorderPixel | ffi::CWColormap | ffi::CWEventMask;
2015-06-14 22:20:32 +01:00
if pl_attribs.override_redirect {
window_attributes |= ffi::CWOverrideRedirect;
}
// finally creating the window
2018-05-29 07:48:47 -04:00
let xwindow = unsafe {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
(xconn.xlib.XCreateWindow)(
xconn.display,
root,
position.map_or(0, |p: PhysicalPosition<i32>| p.x as c_int),
position.map_or(0, |p: PhysicalPosition<i32>| p.y as c_int),
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
dimensions.0 as c_uint,
dimensions.1 as c_uint,
0,
depth,
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::InputOutput as c_uint,
visual,
window_attributes,
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 set_win_attr,
)
};
let mut window = UnownedWindow {
2018-05-29 07:48:47 -04:00
xconn: Arc::clone(xconn),
xwindow,
root,
screen_id,
2018-05-29 07:48:47 -04:00
cursor: Default::default(),
cursor_grabbed: Mutex::new(false),
cursor_visible: Mutex::new(true),
2018-06-14 19:42:18 -04:00
ime_sender: Mutex::new(event_loop.ime_sender.clone()),
shared_state: SharedState::new(guessed_monitor, &window_attrs),
redraw_sender: WakeSender {
waker: event_loop.redraw_sender.waker.clone(),
sender: event_loop.redraw_sender.sender.clone(),
},
};
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
// Title must be set before mapping. Some tiling window managers (i.e. i3) use the window
// title to determine placement/etc., so doing this after mapping would cause the WM to
// act on the wrong title state.
window.set_title_inner(&window_attrs.title).queue();
window
.set_decorations_inner(window_attrs.decorations)
.queue();
{
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
// Enable drag and drop (TODO: extend API to make this toggleable)
unsafe {
2018-05-27 08:49:35 -04:00
let dnd_aware_atom = xconn.get_atom_unchecked(b"XdndAware\0");
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let version = &[5 as c_ulong]; // Latest version; hasn't changed since 2002
2018-05-27 08:49:35 -04:00
xconn.change_property(
2018-05-29 07:48:47 -04:00
window.xwindow,
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
dnd_aware_atom,
ffi::XA_ATOM,
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
util::PropMode::Replace,
version,
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
)
}
.queue();
2015-12-24 10:57:08 +01:00
// WM_CLASS must be set *before* mapping the window, as per ICCCM!
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
{
let (class, instance) = if let Some(name) = pl_attribs.name {
let instance = CString::new(name.instance.as_str())
.expect("`WM_CLASS` instance contained null byte");
let class = CString::new(name.general.as_str())
.expect("`WM_CLASS` class contained null byte");
(instance, class)
} else {
let class = env::args()
.next()
.as_ref()
// Default to the name of the binary (via argv[0])
.and_then(|path| Path::new(path).file_name())
.and_then(|bin_name| bin_name.to_str())
.map(|bin_name| bin_name.to_owned())
.or_else(|| Some(window_attrs.title.clone()))
.and_then(|string| CString::new(string.as_str()).ok())
.expect("Default `WM_CLASS` class contained null byte");
// This environment variable is extraordinarily unlikely to actually be used...
let instance = env::var("RESOURCE_NAME")
.ok()
.and_then(|instance| CString::new(instance.as_str()).ok())
.or_else(|| Some(class.clone()))
.expect("Default `WM_CLASS` instance contained null byte");
(instance, class)
};
2018-05-27 08:49:35 -04:00
let mut class_hint = xconn.alloc_class_hint();
(*class_hint).res_name = class.as_ptr() as *mut c_char;
(*class_hint).res_class = instance.as_ptr() as *mut c_char;
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
unsafe {
(xconn.xlib.XSetClassHint)(xconn.display, window.xwindow, class_hint.ptr);
} //.queue();
}
if let Some(flusher) = window.set_pid() {
flusher.queue()
}
window.set_window_types(pl_attribs.x11_window_types).queue();
if let Some(variant) = pl_attribs.gtk_theme_variant {
window.set_gtk_theme_variant(variant).queue();
}
// set size hints
{
let mut min_inner_size = window_attrs
.min_inner_size
.map(|size| size.to_physical::<u32>(scale_factor));
let mut max_inner_size = window_attrs
.max_inner_size
.map(|size| size.to_physical::<u32>(scale_factor));
if !window_attrs.resizable {
if util::wm_name_is_one_of(&["Xfwm4"]) {
warn!("To avoid a WM bug, disabling resizing has no effect on Xfwm4");
} else {
max_inner_size = Some(dimensions.into());
min_inner_size = Some(dimensions.into());
let mut shared_state = window.shared_state.get_mut();
shared_state.min_inner_size = window_attrs.min_inner_size;
shared_state.max_inner_size = window_attrs.max_inner_size;
shared_state.resize_increments = pl_attribs.resize_increments;
shared_state.base_size = pl_attribs.base_size;
}
}
2018-06-14 19:42:18 -04:00
let mut normal_hints = util::NormalHints::new(xconn);
normal_hints.set_position(position.map(|PhysicalPosition { x, y }| (x, y)));
2018-06-14 19:42:18 -04:00
normal_hints.set_size(Some(dimensions));
normal_hints.set_min_size(min_inner_size.map(Into::into));
normal_hints.set_max_size(max_inner_size.map(Into::into));
normal_hints.set_resize_increments(
pl_attribs
.resize_increments
.map(|size| size.to_physical::<u32>(scale_factor).into()),
);
normal_hints.set_base_size(
pl_attribs
.base_size
.map(|size| size.to_physical::<u32>(scale_factor).into()),
);
2018-06-14 19:42:18 -04:00
xconn.set_normal_hints(window.xwindow, normal_hints).queue();
}
2018-05-07 17:36:21 -04:00
// Set window icons
if let Some(icon) = window_attrs.window_icon {
window.set_icon_inner(icon).queue();
}
// Opt into handling window close
unsafe {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
(xconn.xlib.XSetWMProtocols)(
xconn.display,
2018-05-29 07:48:47 -04:00
window.xwindow,
2019-06-28 01:40:27 +01:00
&[event_loop.wm_delete_window, event_loop.net_wm_ping] as *const ffi::Atom
as *mut ffi::Atom,
2,
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
);
} //.queue();
// Set visibility (map window)
if window_attrs.visible {
unsafe {
2018-05-29 07:48:47 -04:00
(xconn.xlib.XMapRaised)(xconn.display, window.xwindow);
} //.queue();
}
// Attempt to make keyboard input repeat detectable
unsafe {
let mut supported_ptr = ffi::False;
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
(xconn.xlib.XkbSetDetectableAutoRepeat)(
xconn.display,
ffi::True,
&mut supported_ptr,
);
if supported_ptr == ffi::False {
return Err(os_error!(OsError::XMisc(
"`XkbSetDetectableAutoRepeat` failed"
)));
}
}
// Select XInput2 events
let mask = ffi::XI_MotionMask
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_ButtonPressMask
| ffi::XI_ButtonReleaseMask
//| ffi::XI_KeyPressMask
//| ffi::XI_KeyReleaseMask
| ffi::XI_EnterMask
| ffi::XI_LeaveMask
| ffi::XI_FocusInMask
| ffi::XI_FocusOutMask
| ffi::XI_TouchBeginMask
| ffi::XI_TouchUpdateMask
| ffi::XI_TouchEndMask;
xconn
.select_xinput_events(window.xwindow, ffi::XIAllMasterDevices, mask)
.queue();
2018-06-14 19:42:18 -04:00
{
let result = event_loop
.ime
.borrow_mut()
.create_context(window.xwindow, false);
2018-06-14 19:42:18 -04:00
if let Err(err) = result {
let e = match err {
ImeContextCreationError::XError(err) => OsError::XError(err),
ImeContextCreationError::Null => {
OsError::XMisc("IME Context creation failed")
2019-06-24 12:14:55 -04:00
}
};
return Err(os_error!(e));
2018-06-14 19:42:18 -04:00
}
}
// These properties must be set after mapping
if window_attrs.maximized {
window.set_maximized_inner(window_attrs.maximized).queue();
}
if window_attrs.fullscreen.is_some() {
if let Some(flusher) = window.set_fullscreen_inner(window_attrs.fullscreen.clone())
{
flusher.queue()
}
if let Some(PhysicalPosition { x, y }) = position {
let shared_state = window.shared_state.get_mut();
shared_state.restore_position = Some((x, y));
}
}
if window_attrs.always_on_top {
window
.set_always_on_top_inner(window_attrs.always_on_top)
.queue();
}
}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
// We never want to give the user a broken window, since by then, it's too late to handle.
xconn
.sync_with_server()
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
.map(|_| window)
.map_err(|x_err| os_error!(OsError::XError(x_err)))
}
fn set_pid(&self) -> Option<util::Flusher<'_>> {
2018-05-29 07:48:47 -04:00
let pid_atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_PID\0") };
let client_machine_atom = unsafe { self.xconn.get_atom_unchecked(b"WM_CLIENT_MACHINE\0") };
unsafe {
// 64 would suffice for Linux, but 256 will be enough everywhere (as per SUSv2). For instance, this is
// the limit defined by OpenBSD.
const MAXHOSTNAMELEN: usize = 256;
// `assume_init` is safe here because the array consists of `MaybeUninit` values,
// which do not require initialization.
let mut buffer: [MaybeUninit<c_char>; MAXHOSTNAMELEN] =
MaybeUninit::uninit().assume_init();
let status = libc::gethostname(buffer.as_mut_ptr() as *mut c_char, buffer.len());
if status != 0 {
return None;
}
ptr::write(buffer[MAXHOSTNAMELEN - 1].as_mut_ptr() as *mut u8, b'\0'); // a little extra safety
let hostname_length = libc::strlen(buffer.as_ptr() as *const c_char);
let hostname = slice::from_raw_parts(buffer.as_ptr() as *const c_char, hostname_length);
self.xconn
.change_property(
self.xwindow,
pid_atom,
ffi::XA_CARDINAL,
util::PropMode::Replace,
&[libc::getpid() as util::Cardinal],
)
.queue();
2018-05-29 07:48:47 -04:00
let flusher = self.xconn.change_property(
self.xwindow,
client_machine_atom,
ffi::XA_STRING,
util::PropMode::Replace,
&hostname[0..hostname_length],
);
Some(flusher)
}
}
fn set_window_types(&self, window_types: Vec<util::WindowType>) -> util::Flusher<'_> {
2018-05-29 07:48:47 -04:00
let hint_atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_WINDOW_TYPE\0") };
let atoms: Vec<_> = window_types
.iter()
.map(|t| t.as_atom(&self.xconn))
.collect();
2018-05-29 07:48:47 -04:00
self.xconn.change_property(
self.xwindow,
2018-05-27 08:49:35 -04:00
hint_atom,
ffi::XA_ATOM,
util::PropMode::Replace,
&atoms,
2018-05-27 08:49:35 -04:00
)
}
fn set_gtk_theme_variant(&self, variant: String) -> util::Flusher<'_> {
let hint_atom = unsafe { self.xconn.get_atom_unchecked(b"_GTK_THEME_VARIANT\0") };
let utf8_atom = unsafe { self.xconn.get_atom_unchecked(b"UTF8_STRING\0") };
let variant = CString::new(variant).expect("`_GTK_THEME_VARIANT` contained null byte");
self.xconn.change_property(
self.xwindow,
hint_atom,
utf8_atom,
util::PropMode::Replace,
variant.as_bytes(),
)
}
fn set_netwm(
2018-05-29 07:48:47 -04:00
&self,
operation: util::StateOperation,
properties: (c_long, c_long, c_long, c_long),
) -> util::Flusher<'_> {
2018-05-29 07:48:47 -04:00
let state_atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_STATE\0") };
self.xconn.send_client_msg(
self.xwindow,
self.root,
2018-05-27 08:49:35 -04:00
state_atom,
Some(ffi::SubstructureRedirectMask | ffi::SubstructureNotifyMask),
[
operation as c_long,
properties.0,
properties.1,
properties.2,
properties.3,
],
)
}
fn set_fullscreen_hint(&self, fullscreen: bool) -> util::Flusher<'_> {
let fullscreen_atom =
unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_STATE_FULLSCREEN\0") };
let flusher = self.set_netwm(fullscreen.into(), (fullscreen_atom as c_long, 0, 0, 0));
if fullscreen {
// Ensure that the fullscreen window receives input focus to prevent
// locking up the user's display.
unsafe {
(self.xconn.xlib.XSetInputFocus)(
self.xconn.display,
self.xwindow,
ffi::RevertToParent,
ffi::CurrentTime,
);
}
}
flusher
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
}
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
fn set_fullscreen_inner(&self, fullscreen: Option<Fullscreen>) -> Option<util::Flusher<'_>> {
let mut shared_state_lock = self.shared_state.lock();
match shared_state_lock.visibility {
// Setting fullscreen on a window that is not visible will generate an error.
Visibility::No | Visibility::YesWait => {
shared_state_lock.desired_fullscreen = Some(fullscreen);
return None;
}
Visibility::Yes => (),
}
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
let old_fullscreen = shared_state_lock.fullscreen.clone();
if old_fullscreen == fullscreen {
return None;
}
shared_state_lock.fullscreen = fullscreen.clone();
match (&old_fullscreen, &fullscreen) {
// Store the desktop video mode before entering exclusive
// fullscreen, so we can restore it upon exit, as XRandR does not
// provide a mechanism to set this per app-session or restore this
// to the desktop video mode as macOS and Windows do
(
&None,
&Some(Fullscreen::Exclusive(RootVideoMode {
video_mode: PlatformVideoMode::X(ref video_mode),
})),
)
| (
&Some(Fullscreen::Borderless(_)),
&Some(Fullscreen::Exclusive(RootVideoMode {
video_mode: PlatformVideoMode::X(ref video_mode),
})),
) => {
let monitor = video_mode.monitor.as_ref().unwrap();
shared_state_lock.desktop_video_mode =
Some((monitor.id, self.xconn.get_crtc_mode(monitor.id)));
}
// Restore desktop video mode upon exiting exclusive fullscreen
(&Some(Fullscreen::Exclusive(_)), &None)
| (&Some(Fullscreen::Exclusive(_)), &Some(Fullscreen::Borderless(_))) => {
let (monitor_id, mode_id) = shared_state_lock.desktop_video_mode.take().unwrap();
self.xconn
.set_crtc_config(monitor_id, mode_id)
.expect("failed to restore desktop video mode");
}
_ => (),
}
drop(shared_state_lock);
match fullscreen {
Move fullscreen modes to not touch physical resolutions (#270) * Fix X11 screen resolution change using XrandR The previous XF86 resolution switching was broken and everything seems to have moved on to xrandr. Use that instead while cleaning up the code a bit as well. * Use XRandR for actual multiscreen support in X11 * Use actual monitor names in X11 * Get rid of ptr::read usage in X11 * Use a bog standard Vec instead of VecDeque * Get rid of the XRandR mode switching stuff Wayland has made the decision that apps shouldn't change screen resolutions and just take the screens as they've been setup. In the modern world where GPU scaling is cheap and LCD panels are scaling anyway it makes no sense to make "physical" resolution changes when software should be taking care of it. This massively simplifies the code and makes it easier to extend to more niche setups like MST and videowalls. * Rename fullscreen options to match new semantics * Implement XRandR 1.5 support * Get rid of the FullScreen enum Moving to just having two states None and Some(MonitorId) and then being able to set full screen in the current monitor with something like: window.set_fullscreen(Some(window.current_monitor())); * Implement Window::get_current_monitor() Do it by iterating over the available monitors and finding which has the biggest overlap with the window. For this MonitorId needs a new get_position() that needs to be implemented for all platforms. * Add unimplemented get_position() to all MonitorId * Make get_current_monitor() platform specific * Add unimplemented get_current_monitor() to all * Implement proper primary monitor selection in X11 * Shut up some warnings * Remove libxxf86vm package from travis Since we're no longer using XF86 there's no need to keep the package around for CI. * Don't use new struct syntax * Fix indentation * Adjust Android/iOS fullscreen/maximized On Android and iOS we can assume single screen apps that are already fullscreen and maximized so there are a few methods that are implemented by just returning a fixed value or not doing anything. * Mark OSX/Win fullscreen/maximized unimplemented()! These would be safe as no-ops but we should make it explicit so there is more of an incentive to actually implement them.
2017-09-07 09:33:46 +01:00
None => {
2018-06-14 19:42:18 -04:00
let flusher = self.set_fullscreen_hint(false);
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
let mut shared_state_lock = self.shared_state.lock();
if let Some(position) = shared_state_lock.restore_position.take() {
drop(shared_state_lock);
2018-06-14 19:42:18 -04:00
self.set_position_inner(position.0, position.1).queue();
}
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
Some(flusher)
2019-06-24 12:14:55 -04:00
}
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
Some(fullscreen) => {
let (video_mode, monitor) = match fullscreen {
Fullscreen::Exclusive(RootVideoMode {
video_mode: PlatformVideoMode::X(ref video_mode),
}) => (Some(video_mode), video_mode.monitor.clone().unwrap()),
Fullscreen::Borderless(Some(RootMonitorHandle {
inner: PlatformMonitorHandle::X(monitor),
})) => (None, monitor),
Fullscreen::Borderless(None) => (None, self.current_monitor()),
#[cfg(feature = "wayland")]
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
_ => unreachable!(),
};
// Don't set fullscreen on an invalid dummy monitor handle
if monitor.is_dummy() {
return None;
}
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
if let Some(video_mode) = video_mode {
// FIXME: this is actually not correct if we're setting the
// video mode to a resolution higher than the current
// desktop resolution, because XRandR does not automatically
// reposition the monitors to the right and below this
// monitor.
//
// What ends up happening is we will get the fullscreen
// window showing up on those monitors as well, because
// their virtual position now overlaps with the monitor that
// we just made larger..
//
// It'd be quite a bit of work to handle this correctly (and
// nobody else seems to bother doing this correctly either),
// so we're just leaving this broken. Fixing this would
// involve storing all CRTCs upon entering fullscreen,
// restoring them upon exit, and after entering fullscreen,
// repositioning displays to the right and below this
// display. I think there would still be edge cases that are
// difficult or impossible to handle correctly, e.g. what if
// a new monitor was plugged in while in fullscreen?
//
// I think we might just want to disallow setting the video
// mode higher than the current desktop video mode (I'm sure
// this will make someone unhappy, but it's very unusual for
// games to want to do this anyway).
self.xconn
.set_crtc_config(monitor.id, video_mode.native_mode)
.expect("failed to set video mode");
}
let window_position = self.outer_position_physical();
self.shared_state.lock().restore_position = Some(window_position);
let monitor_origin: (i32, i32) = monitor.position().into();
self.set_position_inner(monitor_origin.0, monitor_origin.1)
.queue();
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
Some(self.set_fullscreen_hint(true))
2019-06-24 12:14:55 -04:00
}
}
}
#[inline]
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
pub fn fullscreen(&self) -> Option<Fullscreen> {
let shared_state = self.shared_state.lock();
shared_state
.desired_fullscreen
.clone()
.unwrap_or_else(|| shared_state.fullscreen.clone())
}
2018-06-14 19:42:18 -04:00
#[inline]
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
pub fn set_fullscreen(&self, fullscreen: Option<Fullscreen>) {
if let Some(flusher) = self.set_fullscreen_inner(fullscreen) {
flusher
.sync()
Add exclusive fullscreen mode (#925) * Add exclusive fullscreen mode * Add `WindowExtMacOS::set_fullscreen_presentation_options` * Capture display for exclusive fullscreen on macOS * Fix applying video mode on macOS after a fullscreen cycle * Fix compilation on iOS * Set monitor appropriately for fullscreen on macOS * Fix exclusive to borderless fullscreen transitions on macOS * Fix borderless to exclusive fullscreen transition on macOS * Sort video modes on Windows * Fix fullscreen issues on Windows * Fix video mode changes during exclusive fullscreen on Windows * Add video mode sorting for macOS and iOS * Fix monitor `ns_screen` returning `None` after video mode change * Fix "multithreaded" example on macOS * Restore video mode upon closing an exclusive fullscreen window * Fix "multithreaded" example closing multiple windows at once * Fix compilation on Linux * Update FEATURES.md * Don't care about logical monitor groups on X11 * Add exclusive fullscreen for X11 * Update FEATURES.md * Fix transitions between exclusive and borderless fullscreen on X11 * Update CHANGELOG.md * Document that Wayland doesn't support exclusive fullscreen * Replace core-graphics display mode bindings on macOS * Use `panic!()` instead of `unreachable!()` in "fullscreen" example * Fix fullscreen "always on top" flag on Windows * Track current monitor for fullscreen in "multithreaded" example * Fix exclusive fullscreen sometimes not positioning window properly * Format * More formatting and fix CI issues * Fix formatting * Fix changelog formatting
2019-07-29 21:16:14 +03:00
.expect("Failed to change window fullscreen state");
self.invalidate_cached_frame_extents();
}
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
}
// Called by EventProcessor when a VisibilityNotify event is received
pub(crate) fn visibility_notify(&self) {
let mut shared_state = self.shared_state.lock();
match shared_state.visibility {
Visibility::No => unsafe {
(self.xconn.xlib.XUnmapWindow)(self.xconn.display, self.xwindow);
},
Visibility::Yes => (),
Visibility::YesWait => {
shared_state.visibility = Visibility::Yes;
if let Some(fullscreen) = shared_state.desired_fullscreen.take() {
drop(shared_state);
self.set_fullscreen(fullscreen);
}
}
}
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn current_monitor(&self) -> X11MonitorHandle {
self.shared_state.lock().last_monitor.clone()
Move fullscreen modes to not touch physical resolutions (#270) * Fix X11 screen resolution change using XrandR The previous XF86 resolution switching was broken and everything seems to have moved on to xrandr. Use that instead while cleaning up the code a bit as well. * Use XRandR for actual multiscreen support in X11 * Use actual monitor names in X11 * Get rid of ptr::read usage in X11 * Use a bog standard Vec instead of VecDeque * Get rid of the XRandR mode switching stuff Wayland has made the decision that apps shouldn't change screen resolutions and just take the screens as they've been setup. In the modern world where GPU scaling is cheap and LCD panels are scaling anyway it makes no sense to make "physical" resolution changes when software should be taking care of it. This massively simplifies the code and makes it easier to extend to more niche setups like MST and videowalls. * Rename fullscreen options to match new semantics * Implement XRandR 1.5 support * Get rid of the FullScreen enum Moving to just having two states None and Some(MonitorId) and then being able to set full screen in the current monitor with something like: window.set_fullscreen(Some(window.current_monitor())); * Implement Window::get_current_monitor() Do it by iterating over the available monitors and finding which has the biggest overlap with the window. For this MonitorId needs a new get_position() that needs to be implemented for all platforms. * Add unimplemented get_position() to all MonitorId * Make get_current_monitor() platform specific * Add unimplemented get_current_monitor() to all * Implement proper primary monitor selection in X11 * Shut up some warnings * Remove libxxf86vm package from travis Since we're no longer using XF86 there's no need to keep the package around for CI. * Don't use new struct syntax * Fix indentation * Adjust Android/iOS fullscreen/maximized On Android and iOS we can assume single screen apps that are already fullscreen and maximized so there are a few methods that are implemented by just returning a fixed value or not doing anything. * Mark OSX/Win fullscreen/maximized unimplemented()! These would be safe as no-ops but we should make it explicit so there is more of an incentive to actually implement them.
2017-09-07 09:33:46 +01:00
}
pub fn available_monitors(&self) -> Vec<X11MonitorHandle> {
self.xconn.available_monitors()
}
pub fn primary_monitor(&self) -> X11MonitorHandle {
self.xconn.primary_monitor()
}
fn set_minimized_inner(&self, minimized: bool) -> util::Flusher<'_> {
unsafe {
if minimized {
let screen = (self.xconn.xlib.XDefaultScreen)(self.xconn.display);
(self.xconn.xlib.XIconifyWindow)(self.xconn.display, self.xwindow, screen);
util::Flusher::new(&self.xconn)
} else {
let atom = self.xconn.get_atom_unchecked(b"_NET_ACTIVE_WINDOW\0");
self.xconn.send_client_msg(
self.xwindow,
self.root,
atom,
Some(ffi::SubstructureRedirectMask | ffi::SubstructureNotifyMask),
[1, ffi::CurrentTime as c_long, 0, 0, 0],
)
}
}
}
#[inline]
pub fn set_minimized(&self, minimized: bool) {
self.set_minimized_inner(minimized)
.flush()
.expect("Failed to change window minimization");
}
#[inline]
pub fn is_maximized(&self) -> bool {
let state_atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_STATE\0") };
let state = self
.xconn
.get_property(self.xwindow, state_atom, ffi::XA_ATOM);
let horz_atom = unsafe {
self.xconn
.get_atom_unchecked(b"_NET_WM_STATE_MAXIMIZED_HORZ\0")
};
let vert_atom = unsafe {
self.xconn
.get_atom_unchecked(b"_NET_WM_STATE_MAXIMIZED_VERT\0")
};
match state {
Ok(atoms) => {
let horz_maximized = atoms.iter().any(|atom: &ffi::Atom| *atom == horz_atom);
let vert_maximized = atoms.iter().any(|atom: &ffi::Atom| *atom == vert_atom);
horz_maximized && vert_maximized
}
_ => false,
}
}
fn set_maximized_inner(&self, maximized: bool) -> util::Flusher<'_> {
let horz_atom = unsafe {
self.xconn
.get_atom_unchecked(b"_NET_WM_STATE_MAXIMIZED_HORZ\0")
};
let vert_atom = unsafe {
self.xconn
.get_atom_unchecked(b"_NET_WM_STATE_MAXIMIZED_VERT\0")
};
self.set_netwm(
maximized.into(),
(horz_atom as c_long, vert_atom as c_long, 0, 0),
)
}
2018-06-14 19:42:18 -04:00
#[inline]
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
pub fn set_maximized(&self, maximized: bool) {
self.set_maximized_inner(maximized)
.flush()
.expect("Failed to change window maximization");
self.invalidate_cached_frame_extents();
Move fullscreen modes to not touch physical resolutions (#270) * Fix X11 screen resolution change using XrandR The previous XF86 resolution switching was broken and everything seems to have moved on to xrandr. Use that instead while cleaning up the code a bit as well. * Use XRandR for actual multiscreen support in X11 * Use actual monitor names in X11 * Get rid of ptr::read usage in X11 * Use a bog standard Vec instead of VecDeque * Get rid of the XRandR mode switching stuff Wayland has made the decision that apps shouldn't change screen resolutions and just take the screens as they've been setup. In the modern world where GPU scaling is cheap and LCD panels are scaling anyway it makes no sense to make "physical" resolution changes when software should be taking care of it. This massively simplifies the code and makes it easier to extend to more niche setups like MST and videowalls. * Rename fullscreen options to match new semantics * Implement XRandR 1.5 support * Get rid of the FullScreen enum Moving to just having two states None and Some(MonitorId) and then being able to set full screen in the current monitor with something like: window.set_fullscreen(Some(window.current_monitor())); * Implement Window::get_current_monitor() Do it by iterating over the available monitors and finding which has the biggest overlap with the window. For this MonitorId needs a new get_position() that needs to be implemented for all platforms. * Add unimplemented get_position() to all MonitorId * Make get_current_monitor() platform specific * Add unimplemented get_current_monitor() to all * Implement proper primary monitor selection in X11 * Shut up some warnings * Remove libxxf86vm package from travis Since we're no longer using XF86 there's no need to keep the package around for CI. * Don't use new struct syntax * Fix indentation * Adjust Android/iOS fullscreen/maximized On Android and iOS we can assume single screen apps that are already fullscreen and maximized so there are a few methods that are implemented by just returning a fixed value or not doing anything. * Mark OSX/Win fullscreen/maximized unimplemented()! These would be safe as no-ops but we should make it explicit so there is more of an incentive to actually implement them.
2017-09-07 09:33:46 +01:00
}
fn set_title_inner(&self, title: &str) -> util::Flusher<'_> {
2018-05-29 07:48:47 -04:00
let wm_name_atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_NAME\0") };
let utf8_atom = unsafe { self.xconn.get_atom_unchecked(b"UTF8_STRING\0") };
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let title = CString::new(title).expect("Window title contained null byte");
unsafe {
2018-05-29 07:48:47 -04:00
(self.xconn.xlib.XStoreName)(
self.xconn.display,
self.xwindow,
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
title.as_ptr() as *const c_char,
);
2018-05-29 07:48:47 -04:00
self.xconn.change_property(
self.xwindow,
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
wm_name_atom,
utf8_atom,
util::PropMode::Replace,
title.as_bytes(),
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
)
}
}
2015-12-24 10:57:08 +01:00
2018-06-14 19:42:18 -04:00
#[inline]
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
pub fn set_title(&self, title: &str) {
self.set_title_inner(title)
.flush()
.expect("Failed to set window title");
}
fn set_decorations_inner(&self, decorations: bool) -> util::Flusher<'_> {
self.shared_state.lock().is_decorated = decorations;
let mut hints = self.xconn.get_motif_hints(self.xwindow);
hints.set_decorations(decorations);
self.xconn.set_motif_hints(self.xwindow, &hints)
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
}
2018-06-14 19:42:18 -04:00
#[inline]
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
pub fn set_decorations(&self, decorations: bool) {
self.set_decorations_inner(decorations)
.flush()
.expect("Failed to set decoration state");
self.invalidate_cached_frame_extents();
}
#[inline]
pub fn is_decorated(&self) -> bool {
self.shared_state.lock().is_decorated
}
fn set_maximizable_inner(&self, maximizable: bool) -> util::Flusher<'_> {
let mut hints = self.xconn.get_motif_hints(self.xwindow);
hints.set_maximizable(maximizable);
self.xconn.set_motif_hints(self.xwindow, &hints)
}
fn set_always_on_top_inner(&self, always_on_top: bool) -> util::Flusher<'_> {
2018-05-29 07:48:47 -04:00
let above_atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_STATE_ABOVE\0") };
self.set_netwm(always_on_top.into(), (above_atom as c_long, 0, 0, 0))
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn set_always_on_top(&self, always_on_top: bool) {
self.set_always_on_top_inner(always_on_top)
.flush()
.expect("Failed to set always-on-top state");
}
fn set_icon_inner(&self, icon: Icon) -> util::Flusher<'_> {
2018-05-29 07:48:47 -04:00
let icon_atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_ICON\0") };
2018-05-07 17:36:21 -04:00
let data = icon.to_cardinals();
2018-05-29 07:48:47 -04:00
self.xconn.change_property(
self.xwindow,
2018-05-27 08:49:35 -04:00
icon_atom,
ffi::XA_CARDINAL,
util::PropMode::Replace,
data.as_slice(),
)
2018-05-07 17:36:21 -04:00
}
fn unset_icon_inner(&self) -> util::Flusher<'_> {
2018-05-29 07:48:47 -04:00
let icon_atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_ICON\0") };
2018-05-07 17:36:21 -04:00
let empty_data: [util::Cardinal; 0] = [];
2018-05-29 07:48:47 -04:00
self.xconn.change_property(
self.xwindow,
2018-05-27 08:49:35 -04:00
icon_atom,
ffi::XA_CARDINAL,
util::PropMode::Replace,
&empty_data,
)
2018-05-07 17:36:21 -04:00
}
2018-06-14 19:42:18 -04:00
#[inline]
2018-05-07 17:36:21 -04:00
pub fn set_window_icon(&self, icon: Option<Icon>) {
match icon {
Some(icon) => self.set_icon_inner(icon),
None => self.unset_icon_inner(),
}
.flush()
.expect("Failed to set icons");
2018-05-07 17:36:21 -04:00
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn set_visible(&self, visible: bool) {
let mut shared_state = self.shared_state.lock();
match (visible, shared_state.visibility) {
(true, Visibility::Yes) | (true, Visibility::YesWait) | (false, Visibility::No) => {
return
}
_ => (),
}
if visible {
unsafe {
(self.xconn.xlib.XMapRaised)(self.xconn.display, self.xwindow);
}
self.xconn
.flush_requests()
.expect("Failed to call XMapRaised");
shared_state.visibility = Visibility::YesWait;
} else {
unsafe {
(self.xconn.xlib.XUnmapWindow)(self.xconn.display, self.xwindow);
}
self.xconn
.flush_requests()
.expect("Failed to call XUnmapWindow");
shared_state.visibility = Visibility::No;
}
}
#[inline]
pub fn is_visible(&self) -> Option<bool> {
Some(self.shared_state.lock().visibility == Visibility::Yes)
}
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
fn update_cached_frame_extents(&self) {
let extents = self
.xconn
.get_frame_extents_heuristic(self.xwindow, self.root);
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
(*self.shared_state.lock()).frame_extents = Some(extents);
}
2018-06-14 19:42:18 -04:00
pub(crate) fn invalidate_cached_frame_extents(&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
(*self.shared_state.lock()).frame_extents.take();
}
pub(crate) fn outer_position_physical(&self) -> (i32, i32) {
2018-06-14 19:42:18 -04:00
let extents = (*self.shared_state.lock()).frame_extents.clone();
if let Some(extents) = extents {
let (x, y) = self.inner_position_physical();
extents.inner_pos_to_outer(x, y)
2018-06-14 19:42:18 -04:00
} else {
self.update_cached_frame_extents();
self.outer_position_physical()
2018-06-14 19:42:18 -04:00
}
}
2015-09-21 14:42:05 +02:00
#[inline]
pub fn outer_position(&self) -> Result<PhysicalPosition<i32>, NotSupportedError> {
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
let extents = (*self.shared_state.lock()).frame_extents.clone();
if let Some(extents) = extents {
let (x, y) = self.inner_position_physical();
Ok(extents.inner_pos_to_outer(x, y).into())
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 {
self.update_cached_frame_extents();
self.outer_position()
X11: General cleanup (#491) * X11: General cleanup This is almost entirely internal changes, and as usual, doesn't actually fix any problems people have complained about. - `XSetInputFocus` can't be called before the window is visible. This was previously handled by looping (with a sleep) and querying for the window's state until it was visible. Now we use `XIfEvent`, which blocks until we receive `VisibilityNotify`. Note that this can't be replaced with an `XSync` (I tried). - We now call `XSync` at the end of window creation and check for errors, assuring that broken windows are never returned. When creating invisible windows, this is the only time the output buffer is flushed during the entire window creation process (AFAIK). For visible windows, `XIfEvent` will generally flush, but window creation has overall been reduced to the minimum number of flushes. - `check_errors().expect()` has been a common pattern throughout the backend, but it seems that people (myself included) didn't make a distinction between using it after synchronous requests and asynchronous requests. Now we only use it after async requests if we flush first, though this still isn't correct (since the request likely hasn't been processed yet). The only real solution (besides forcing a sync *every time*) is to handle asynchronous errors *asynchronously*. For future work, I plan on adding logging, though I don't plan on actually *handling* those errors; that's more of something to hope for in the hypothetical async/await XCB paradise. - We now flush whenever it makes sense to. `util::Flusher` was added to force contributors to be aware of the output buffer. - `Window::get_position`, `Window::get_inner_position`, `Window::get_inner_size`, and `Window::get_outer_size` previously all required *several* round-trips. On my machine, it took an average of around 80µs. They've now been reduced to one round-trip each, which reduces my measurement to 16µs. This was accomplished simply by caching the frame extents, which are expensive to calculate (due to various queries and heuristics), but change infrequently and predictably. I still recommend that application developers use these methods sparingly and generally prefer storing the values from `Resized`/`Moved`, as that's zero overhead. - The above change enabled me to change the `Moved` event to supply window positions, rather than client area positions. Additionally, we no longer generate `Moved` for real (as in, not synthetic) `ConfigureNotify` events. Real `ConfigureNotify` events contain positions relative to the parent window, which are typically constant and useless. Since that position would be completely different from the root-relative positions supplied by synthetic `ConfigureNotify` events (which are the vast majority of them), that meant real `ConfigureNotify` events would *always* be detected as the position having changed, so the resultant `Moved` was multiple levels of misleading. In practice, this meant a garbage `Moved` would be sent every time the window was resized; now a resize has to actually change the window's position to be accompanied by `Moved`. - Every time we processed an `XI_Enter` event, we would leak 4 bytes via `util::query_pointer` (`XIQueryPointer`). `XIButtonState` contains a dynamically-allocated mask field which we weren't freeing. As this event occurs with fairly high frequency, long-running applications could easily accumulate substantial leaks. `util::PointerState::drop` now takes care of this. - The `util` module has been split up into several sub-modules, as it was getting rather lengthy. This accounts for a significant part of this diff, unfortunately. - Atoms are now cached. Xlib caches them too, so `XInternAtom` wouldn't typically be a round-trip anyway, but the added complexity is negligible. - Switched from `std::sync::Mutex` to `parking_lot::Mutex` (within this backend). There appears to be no downside to this, but if anyone finds one, this would be easy to revert. - The WM name and supported hints are now global to the application, and are updated upon `ReparentNotify`, which should detect when the WM was replaced (assuming a reparenting WM was involved, that is). Previously, these values were per-window and would never update, meaning replacing the WM could potentially lead to (admittedly very minor) problems. - The result of `Window2::create_empty_cursor` will now only be used if it actually succeeds. - `Window2::load_cursor` no longer re-allocates the cursor name. - `util::lookup_utf8` previously allocated a 16-byte buffer on the heap. Now it allocates a 1024-byte buffer on the stack, and falls back to dynamic allocation if the buffer is too small. This base buffer size is admittedly gratuitous, but less so if you're using IME. - `with_c_str` was finally removed. - Added `util::Format` enum to help prevent goofs when dealing with format arguments. - `util::get_property`, something I added way back in my first winit PR, only calculated offsets correctly for `util::Format::Char`. This was concealed by the accomodating buffer size, as it would be very rare for the offset to be needed; however, testing with a buffer size of 1, `util::Format::Long` would read from the same offset multiple times, and `util::Format::Short` would miss data. This function now works correctly for all formats, relying on the simple fact that the offset increases by the buffer size on each iteration. We also account for the extra byte that `XGetWindowProperty` allocates at the end of the buffer, and copy data from the buffer instead of moving it and taking ownership of the pointer. - Drag and drop now reliably works in release mode. This is presumably related to the `util::get_property` changes. - `util::change_property` now exists, which should make it easier to add features in the future. - The `EventsLoop` device map is no longer in a mutex. - `XConnection` now implements `Debug`. - Valgrind no longer complains about anything related to winit (with either the system allocator or jemalloc, though "not having valgrind complain about jemalloc" isn't something to strive for). * X11: Add better diagnostics when initialization fails * X11: Handle XIQueryDevice failure * X11: Use correct types in error handler
2018-05-03 09:15:49 -04:00
}
}
pub(crate) fn inner_position_physical(&self) -> (i32, i32) {
// This should be okay to unwrap since the only error XTranslateCoordinates can return
// is BadWindow, and if the window handle is bad we have bigger problems.
self.xconn
.translate_coords(self.xwindow, self.root)
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
.map(|coords| (coords.x_rel_root, coords.y_rel_root))
.unwrap()
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn inner_position(&self) -> Result<PhysicalPosition<i32>, NotSupportedError> {
Ok(self.inner_position_physical().into())
2018-06-14 19:42:18 -04:00
}
pub(crate) fn set_position_inner(&self, mut x: i32, mut y: i32) -> util::Flusher<'_> {
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
// There are a few WMs that set client area position rather than window position, so
// we'll translate for consistency.
if util::wm_name_is_one_of(&["Enlightenment", "FVWM"]) {
let extents = (*self.shared_state.lock()).frame_extents.clone();
if let Some(extents) = extents {
x += extents.frame_extents.left as i32;
y += extents.frame_extents.top as i32;
} else {
self.update_cached_frame_extents();
2018-06-14 19:42:18 -04:00
return self.set_position_inner(x, y);
}
}
unsafe {
(self.xconn.xlib.XMoveWindow)(self.xconn.display, self.xwindow, x as c_int, y as c_int);
2018-06-14 19:42:18 -04:00
}
util::Flusher::new(&self.xconn)
}
pub(crate) fn set_position_physical(&self, x: i32, y: i32) {
self.set_position_inner(x, y)
.flush()
.expect("Failed to call `XMoveWindow`");
}
2015-09-21 14:42:05 +02:00
#[inline]
pub fn set_outer_position(&self, position: Position) {
let (x, y) = position.to_physical::<i32>(self.scale_factor()).into();
2018-06-14 19:42:18 -04:00
self.set_position_physical(x, y);
}
pub(crate) fn inner_size_physical(&self) -> (u32, u32) {
// This should be okay to unwrap since the only error XGetGeometry can return
// is BadWindow, and if the window handle is bad we have bigger problems.
self.xconn
.get_geometry(self.xwindow)
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
.map(|geo| (geo.width, geo.height))
.unwrap()
}
2015-09-21 14:42:05 +02:00
#[inline]
pub fn inner_size(&self) -> PhysicalSize<u32> {
self.inner_size_physical().into()
2018-06-14 19:42:18 -04:00
}
2015-09-21 14:42:05 +02:00
#[inline]
pub fn outer_size(&self) -> PhysicalSize<u32> {
2018-06-14 19:42:18 -04:00
let extents = self.shared_state.lock().frame_extents.clone();
if let Some(extents) = extents {
let (width, height) = self.inner_size_physical();
extents.inner_size_to_outer(width, height).into()
2018-06-14 19:42:18 -04:00
} else {
self.update_cached_frame_extents();
self.outer_size()
2018-06-14 19:42:18 -04:00
}
}
pub(crate) fn set_inner_size_physical(&self, width: u32, height: u32) {
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
unsafe {
2018-05-29 07:48:47 -04:00
(self.xconn.xlib.XResizeWindow)(
self.xconn.display,
self.xwindow,
width as c_uint,
height as c_uint,
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
self.xconn.flush_requests()
}
.expect("Failed to call `XResizeWindow`");
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn set_inner_size(&self, size: Size) {
let scale_factor = self.scale_factor();
let size = size.to_physical::<u32>(scale_factor).into();
if !self.shared_state.lock().is_resizable {
self.update_normal_hints(|normal_hints| {
normal_hints.set_min_size(Some(size));
normal_hints.set_max_size(Some(size));
})
.expect("Failed to call `XSetWMNormalHints`");
}
self.set_inner_size_physical(size.0, size.1);
2018-06-14 19:42:18 -04:00
}
fn update_normal_hints<F>(&self, callback: F) -> Result<(), XError>
where
F: FnOnce(&mut util::NormalHints<'_>),
{
2018-06-14 19:42:18 -04:00
let mut normal_hints = self.xconn.get_normal_hints(self.xwindow)?;
callback(&mut normal_hints);
self.xconn
.set_normal_hints(self.xwindow, normal_hints)
.flush()
2018-06-14 19:42:18 -04:00
}
pub(crate) fn set_min_inner_size_physical(&self, dimensions: Option<(u32, u32)>) {
2018-06-14 19:42:18 -04:00
self.update_normal_hints(|normal_hints| normal_hints.set_min_size(dimensions))
.expect("Failed to call `XSetWMNormalHints`");
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn set_min_inner_size(&self, dimensions: Option<Size>) {
self.shared_state.lock().min_inner_size = dimensions;
let physical_dimensions =
dimensions.map(|dimensions| dimensions.to_physical::<u32>(self.scale_factor()).into());
self.set_min_inner_size_physical(physical_dimensions);
2018-06-14 19:42:18 -04:00
}
pub(crate) fn set_max_inner_size_physical(&self, dimensions: Option<(u32, u32)>) {
2018-06-14 19:42:18 -04:00
self.update_normal_hints(|normal_hints| normal_hints.set_max_size(dimensions))
.expect("Failed to call `XSetWMNormalHints`");
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn set_max_inner_size(&self, dimensions: Option<Size>) {
self.shared_state.lock().max_inner_size = dimensions;
let physical_dimensions =
dimensions.map(|dimensions| dimensions.to_physical::<u32>(self.scale_factor()).into());
self.set_max_inner_size_physical(physical_dimensions);
}
2018-06-14 19:42:18 -04:00
pub(crate) fn adjust_for_dpi(
&self,
old_scale_factor: f64,
new_scale_factor: f64,
width: u32,
height: u32,
shared_state: &SharedState,
) -> (u32, u32) {
let scale_factor = new_scale_factor / old_scale_factor;
2018-06-14 19:42:18 -04:00
self.update_normal_hints(|normal_hints| {
let dpi_adjuster =
|size: Size| -> (u32, u32) { size.to_physical::<u32>(new_scale_factor).into() };
let max_size = shared_state.max_inner_size.map(&dpi_adjuster);
let min_size = shared_state.min_inner_size.map(&dpi_adjuster);
let resize_increments = shared_state.resize_increments.map(&dpi_adjuster);
let base_size = shared_state.base_size.map(&dpi_adjuster);
2018-06-14 19:42:18 -04:00
normal_hints.set_max_size(max_size);
normal_hints.set_min_size(min_size);
normal_hints.set_resize_increments(resize_increments);
normal_hints.set_base_size(base_size);
})
.expect("Failed to update normal hints");
let new_width = (width as f64 * scale_factor).round() as u32;
let new_height = (height as f64 * scale_factor).round() as u32;
(new_width, new_height)
}
pub fn set_resizable(&self, resizable: bool) {
if util::wm_name_is_one_of(&["Xfwm4"]) {
// Making the window unresizable on Xfwm prevents further changes to `WM_NORMAL_HINTS` from being detected.
// This makes it impossible for resizing to be re-enabled, and also breaks DPI scaling. As such, we choose
// the lesser of two evils and do nothing.
warn!("To avoid a WM bug, disabling resizing has no effect on Xfwm4");
return;
}
2018-06-14 19:42:18 -04:00
let (min_size, max_size) = if resizable {
2018-06-14 19:42:18 -04:00
let shared_state_lock = self.shared_state.lock();
(
shared_state_lock.min_inner_size,
shared_state_lock.max_inner_size,
)
} else {
let window_size = Some(Size::from(self.inner_size()));
(window_size, window_size)
2018-06-14 19:42:18 -04:00
};
self.shared_state.lock().is_resizable = resizable;
2018-06-14 19:42:18 -04:00
self.set_maximizable_inner(resizable).queue();
let scale_factor = self.scale_factor();
let min_inner_size = min_size
.map(|size| size.to_physical::<u32>(scale_factor))
2018-06-14 19:42:18 -04:00
.map(Into::into);
let max_inner_size = max_size
.map(|size| size.to_physical::<u32>(scale_factor))
2018-06-14 19:42:18 -04:00
.map(Into::into);
self.update_normal_hints(|normal_hints| {
normal_hints.set_min_size(min_inner_size);
normal_hints.set_max_size(max_inner_size);
})
.expect("Failed to call `XSetWMNormalHints`");
}
#[inline]
pub fn is_resizable(&self) -> bool {
self.shared_state.lock().is_resizable
}
#[inline]
pub fn xlib_display(&self) -> *mut c_void {
2018-05-29 07:48:47 -04:00
self.xconn.display as _
}
2016-10-19 19:11:02 +03:00
#[inline]
pub fn xlib_screen_id(&self) -> c_int {
2018-05-29 07:48:47 -04:00
self.screen_id
2016-10-19 19:11:02 +03:00
}
#[inline]
pub fn xlib_xconnection(&self) -> Arc<XConnection> {
2018-05-29 07:48:47 -04:00
Arc::clone(&self.xconn)
2016-10-19 19:11:02 +03:00
}
#[inline]
pub fn xlib_window(&self) -> c_ulong {
2018-05-29 07:48:47 -04:00
self.xwindow
}
2015-09-21 14:42:05 +02:00
#[inline]
pub fn xcb_connection(&self) -> *mut c_void {
unsafe { (self.xconn.xlib_xcb.XGetXCBConnection)(self.xconn.display) as *mut _ }
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn set_cursor_icon(&self, cursor: CursorIcon) {
let old_cursor = replace(&mut *self.cursor.lock(), cursor);
if cursor != old_cursor && *self.cursor_visible.lock() {
self.xconn.set_cursor_icon(self.xwindow, Some(cursor));
}
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn set_cursor_grab(&self, grab: bool) -> Result<(), ExternalError> {
let mut grabbed_lock = self.cursor_grabbed.lock();
if grab == *grabbed_lock {
return Ok(());
}
unsafe {
// We ungrab before grabbing to prevent passive grabs from causing `AlreadyGrabbed`.
// Therefore, this is common to both codepaths.
(self.xconn.xlib.XUngrabPointer)(self.xconn.display, ffi::CurrentTime);
2016-03-21 18:42:54 +00:00
}
let result = if grab {
let result = unsafe {
(self.xconn.xlib.XGrabPointer)(
self.xconn.display,
self.xwindow,
ffi::True,
(ffi::ButtonPressMask
| ffi::ButtonReleaseMask
| ffi::EnterWindowMask
| ffi::LeaveWindowMask
| ffi::PointerMotionMask
| ffi::PointerMotionHintMask
| ffi::Button1MotionMask
| ffi::Button2MotionMask
| ffi::Button3MotionMask
| ffi::Button4MotionMask
| ffi::Button5MotionMask
| ffi::ButtonMotionMask
| ffi::KeymapStateMask) as c_uint,
ffi::GrabModeAsync,
ffi::GrabModeAsync,
self.xwindow,
0,
ffi::CurrentTime,
)
};
2016-03-21 18:42:54 +00:00
match result {
ffi::GrabSuccess => Ok(()),
ffi::AlreadyGrabbed => {
Err("Cursor could not be grabbed: already grabbed by another client")
2019-06-24 12:14:55 -04:00
}
ffi::GrabInvalidTime => Err("Cursor could not be grabbed: invalid time"),
ffi::GrabNotViewable => {
Err("Cursor could not be grabbed: grab location not viewable")
2019-06-24 12:14:55 -04:00
}
ffi::GrabFrozen => Err("Cursor could not be grabbed: frozen by another client"),
_ => unreachable!(),
}
.map_err(|err| ExternalError::Os(os_error!(OsError::XMisc(err))))
} else {
self.xconn
.flush_requests()
.map_err(|err| ExternalError::Os(os_error!(OsError::XError(err))))
};
if result.is_ok() {
*grabbed_lock = grab;
2016-03-21 18:42:54 +00:00
}
result
}
#[inline]
pub fn set_cursor_visible(&self, visible: bool) {
let mut visible_lock = self.cursor_visible.lock();
if visible == *visible_lock {
return;
}
let cursor = if visible {
Some(*self.cursor.lock())
} else {
None
};
*visible_lock = visible;
drop(visible_lock);
self.xconn.set_cursor_icon(self.xwindow, cursor);
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn scale_factor(&self) -> f64 {
self.current_monitor().scale_factor
}
pub fn set_cursor_position_physical(&self, x: i32, y: i32) -> Result<(), ExternalError> {
unsafe {
(self.xconn.xlib.XWarpPointer)(self.xconn.display, 0, self.xwindow, 0, 0, 0, 0, x, y);
self.xconn
.flush_requests()
.map_err(|e| ExternalError::Os(os_error!(OsError::XError(e))))
}
}
2018-06-14 19:42:18 -04:00
#[inline]
pub fn set_cursor_position(&self, position: Position) -> Result<(), ExternalError> {
let (x, y) = position.to_physical::<i32>(self.scale_factor()).into();
2018-06-14 19:42:18 -04:00
self.set_cursor_position_physical(x, y)
}
#[inline]
pub fn set_cursor_hittest(&self, _hittest: bool) -> Result<(), ExternalError> {
Err(ExternalError::NotSupported(NotSupportedError::new()))
}
pub fn drag_window(&self) -> Result<(), ExternalError> {
let pointer = self
.xconn
.query_pointer(self.xwindow, util::VIRTUAL_CORE_POINTER)
.map_err(|err| ExternalError::Os(os_error!(OsError::XError(err))))?;
let window = self.inner_position().map_err(ExternalError::NotSupported)?;
let message = unsafe { self.xconn.get_atom_unchecked(b"_NET_WM_MOVERESIZE\0") };
// we can't use `set_cursor_grab(false)` here because it doesn't run `XUngrabPointer`
// if the cursor isn't currently grabbed
let mut grabbed_lock = self.cursor_grabbed.lock();
unsafe {
(self.xconn.xlib.XUngrabPointer)(self.xconn.display, ffi::CurrentTime);
}
self.xconn
.flush_requests()
.map_err(|err| ExternalError::Os(os_error!(OsError::XError(err))))?;
*grabbed_lock = false;
// we keep the lock until we are done
self.xconn
.send_client_msg(
self.xwindow,
self.root,
message,
Some(ffi::SubstructureRedirectMask | ffi::SubstructureNotifyMask),
[
(window.x as c_long + pointer.win_x as c_long),
(window.y as c_long + pointer.win_y as c_long),
8, // _NET_WM_MOVERESIZE_MOVE
ffi::Button1 as c_long,
1,
],
)
.flush()
.map_err(|err| ExternalError::Os(os_error!(OsError::XError(err))))
}
#[inline]
pub fn set_ime_position(&self, spot: Position) {
let (x, y) = spot.to_physical::<i32>(self.scale_factor()).into();
let _ = self
.ime_sender
2018-06-14 19:42:18 -04:00
.lock()
.send(ImeRequest::Position(self.xwindow, x, y));
2018-06-14 19:42:18 -04:00
}
#[inline]
pub fn set_ime_allowed(&self, allowed: bool) {
let _ = self
.ime_sender
.lock()
.send(ImeRequest::Allow(self.xwindow, allowed));
2018-06-14 19:42:18 -04:00
}
2021-05-19 18:39:53 +02:00
#[inline]
pub fn focus_window(&self) {
let state_atom = unsafe { self.xconn.get_atom_unchecked(b"WM_STATE\0") };
let state_type_atom = unsafe { self.xconn.get_atom_unchecked(b"CARD32\0") };
let is_minimized = if let Ok(state) =
self.xconn
.get_property(self.xwindow, state_atom, state_type_atom)
{
state.contains(&(ffi::IconicState as c_ulong))
} else {
false
};
let is_visible = match self.shared_state.lock().visibility {
Visibility::Yes => true,
Visibility::YesWait | Visibility::No => false,
};
if is_visible && !is_minimized {
let atom = unsafe { self.xconn.get_atom_unchecked(b"_NET_ACTIVE_WINDOW\0") };
let flusher = self.xconn.send_client_msg(
self.xwindow,
self.root,
atom,
Some(ffi::SubstructureRedirectMask | ffi::SubstructureNotifyMask),
[1, ffi::CurrentTime as c_long, 0, 0, 0],
);
if let Err(e) = flusher.flush() {
log::error!(
"`flush` returned an error when focusing the window. Error was: {}",
e
);
}
}
}
#[inline]
pub fn request_user_attention(&self, request_type: Option<UserAttentionType>) {
let mut wm_hints = self
.xconn
.get_wm_hints(self.xwindow)
.expect("`XGetWMHints` failed");
if request_type.is_some() {
(*wm_hints).flags |= ffi::XUrgencyHint;
} else {
(*wm_hints).flags &= !ffi::XUrgencyHint;
}
self.xconn
.set_wm_hints(self.xwindow, wm_hints)
.flush()
.expect("Failed to set urgency hint");
}
#[inline]
pub fn id(&self) -> WindowId {
WindowId(self.xwindow)
}
#[inline]
pub fn request_redraw(&self) {
self.redraw_sender
.sender
.send(WindowId(self.xwindow))
.unwrap();
self.redraw_sender.waker.wake().unwrap();
}
#[inline]
pub fn raw_window_handle(&self) -> XlibHandle {
let mut handle = XlibHandle::empty();
handle.window = self.xlib_window();
handle.display = self.xlib_display();
handle
}
}