Overhaul device events API and add gamepad support on Windows #804
Conversation
* When you have multiple windows, you no longer receive duplicate device events * Mouse Device Events now send X-button input * Mouse Device Events now send horizontal scroll wheel input
This helps with properly calculating the deadzone for controller joysticks. One potential issue is that the `Stick` variant isn't used for *all* joysticks, which could be potentially confusing - for example, raw input joysticks will never use the `Stick` variant because we don't understand the semantic meaning of raw input joystick axes.
|
WIll it be more confusing to have Device events broken out individually while WindowEvents remain bundled under a common enum? This might make it more difficult to understand the subtitles between similar events (e.g. MouseMotion vs CursorMoved). |
|
@aloucks That's a good point! It's probably worth taking a closer look at |
|
I finished reading through #695, but I'm still curious as to why it was decided to support gamepads in winit. Regardless of whether people expect winit to support it or not, gilrs already exists for talking to gamepads, and it looks to me like it does this pretty well. It seems like a waste to duplicate this effort. I thought this was previously declared as "out of scope" for winit? |
|
@aleksijuvani It was indeed declared out of scope previously, but that was by a different person that's no longer actively involved with the project and I personally disagree with it. As for why we'd implement this ourselves instead of just directing people to |
|
Once I've rebased this, I'm going to merge this onto the |
|
Can we merge this instead against https://github.com/rust-windowing/winit/tree/dpi-overhaul? Makes my life easier, as I don't have to maintain two copies of glutin. |
|
@zegentzy I'd rather not, since that makes it impossible to implement the DPI changes and gamepad support in separate PRs - you'd have to deal with the error messages from both at once, which sounds like a bad experience for new contributors. Why would you have to maintain additional copies of glutin, anyway? The intent isn't for people to actively use those branches in their applications - it's so we have a centralized place to develop these features, and then merge them onto master. Glutin isn't necessary for either of those processes. (also I'd be surprised if the breaking changes in this patch break things on Glutin's end) |
|
People usually (I think?) only want to implement something if they want to use it in their application. Generally, from my experience at least, you port your application to this shiny new unimplemented api, then fiddle with the dependency till it compiles and works. Like, what better way to test if gampad support works then seeing if it works in your application? Now, if these applications are dependent on glutin, and this PR doesn't break anything in glutin, then I guess it's as simple as a Besides, what if they want to run their application with both the dpi overhaul and gamepad support? I see little joy for the poor soul who was to merge the two branches together (and then demerge them to send in PRs separately), as git is a PITA, nor for the poor soul in the end who has to merge the two branches back, one-by-one, into master, hoping for the changes from each don't conflict with each other. And, of course, now we got to backport changes from master back to two branches, instead of just one, to avoid them getting too out of sync from master. This is like the eventsloopv2-master split, but with twice the pain. |
|
Remember,
I just tested it, and I can confirm that this PR doesn't break anything in Glutin.
I'll agree with that, but see point above.
The difference between these two branches and the EL2 branch is that the changes in these branches are much less fundamental than the changes introduced by the EL2 branch. We aren't in a situation where changes to Besides, these branches should get merged much more quickly than than the EL2 branch did, since the changes are much easier to implement and can be implemented without a deep understanding of Winit's structure.
They can either merge the branches themselves, or wait for the changes to land on
I'd argue that merging two WIP branches, fixing the issues raised by both branches, than de-merging them is a gross misuse of git as a tool. |
* Initial implementation * Corrected RAWINPUT buffer sizing * Mostly complete XInput implementation * XInput triggers * Add preliminary CHANGELOG entry. * match unix common API to evl 2.0 * wayland: eventloop2.0 * make EventLoopProxy require T: 'static * Revamp device event API, as well as several misc. fixes on Windows: * When you have multiple windows, you no longer receive duplicate device events * Mouse Device Events now send X-button input * Mouse Device Events now send horizontal scroll wheel input * Add MouseEvent documentation and Device ID debug passthrough * Improve type safety on get_raw_input_data * Remove button_id field from MouseEvent::Button in favor of �utton * Remove regex dependency on Windows * Remove axis filtering in XInput * Make gamepads not use lazy_static * Publicly expose gamepad rumble * Unstack DeviceEvent and fix examples/tests * Add HANDLE retrieval method to DeviceExtWindows * Add distinction between non-joystick axes and joystick axes. This helps with properly calculating the deadzone for controller joysticks. One potential issue is that the `Stick` variant isn't used for *all* joysticks, which could be potentially confusing - for example, raw input joysticks will never use the `Stick` variant because we don't understand the semantic meaning of raw input joystick axes. * Add ability to get gamepad port * Fix xinput controller hot swapping * Add functions for enumerating attached devices * Clamp input to [0.0, 1.0] on gamepad rumble * Expose gamepad rumble errors * Add method to check if device is still connected * Add docs * Rename AxisHint and ButtonHint to GamepadAxis and GamepadButton * Add CHANGELOG entry * Update CHANGELOG.md * Add HidId and MovedAbsolute * Fix xinput deprecation warnings * Add ability to retrieve gamepad battery level * Fix weird imports in gamepad example * Update CHANGELOG.md * Resolve francesca64 comments
* Initial implementation * Corrected RAWINPUT buffer sizing * Mostly complete XInput implementation * XInput triggers * Add preliminary CHANGELOG entry. * match unix common API to evl 2.0 * wayland: eventloop2.0 * make EventLoopProxy require T: 'static * Revamp device event API, as well as several misc. fixes on Windows: * When you have multiple windows, you no longer receive duplicate device events * Mouse Device Events now send X-button input * Mouse Device Events now send horizontal scroll wheel input * Add MouseEvent documentation and Device ID debug passthrough * Improve type safety on get_raw_input_data * Remove button_id field from MouseEvent::Button in favor of �utton * Remove regex dependency on Windows * Remove axis filtering in XInput * Make gamepads not use lazy_static * Publicly expose gamepad rumble * Unstack DeviceEvent and fix examples/tests * Add HANDLE retrieval method to DeviceExtWindows * Add distinction between non-joystick axes and joystick axes. This helps with properly calculating the deadzone for controller joysticks. One potential issue is that the `Stick` variant isn't used for *all* joysticks, which could be potentially confusing - for example, raw input joysticks will never use the `Stick` variant because we don't understand the semantic meaning of raw input joystick axes. * Add ability to get gamepad port * Fix xinput controller hot swapping * Add functions for enumerating attached devices * Clamp input to [0.0, 1.0] on gamepad rumble * Expose gamepad rumble errors * Add method to check if device is still connected * Add docs * Rename AxisHint and ButtonHint to GamepadAxis and GamepadButton * Add CHANGELOG entry * Update CHANGELOG.md * Add HidId and MovedAbsolute * Fix xinput deprecation warnings * Add ability to retrieve gamepad battery level * Fix weird imports in gamepad example * Update CHANGELOG.md * Resolve francesca64 comments
* Initial implementation * Corrected RAWINPUT buffer sizing * Mostly complete XInput implementation * XInput triggers * Add preliminary CHANGELOG entry. * match unix common API to evl 2.0 * wayland: eventloop2.0 * make EventLoopProxy require T: 'static * Revamp device event API, as well as several misc. fixes on Windows: * When you have multiple windows, you no longer receive duplicate device events * Mouse Device Events now send X-button input * Mouse Device Events now send horizontal scroll wheel input * Add MouseEvent documentation and Device ID debug passthrough * Improve type safety on get_raw_input_data * Remove button_id field from MouseEvent::Button in favor of �utton * Remove regex dependency on Windows * Remove axis filtering in XInput * Make gamepads not use lazy_static * Publicly expose gamepad rumble * Unstack DeviceEvent and fix examples/tests * Add HANDLE retrieval method to DeviceExtWindows * Add distinction between non-joystick axes and joystick axes. This helps with properly calculating the deadzone for controller joysticks. One potential issue is that the `Stick` variant isn't used for *all* joysticks, which could be potentially confusing - for example, raw input joysticks will never use the `Stick` variant because we don't understand the semantic meaning of raw input joystick axes. * Add ability to get gamepad port * Fix xinput controller hot swapping * Add functions for enumerating attached devices * Clamp input to [0.0, 1.0] on gamepad rumble * Expose gamepad rumble errors * Add method to check if device is still connected * Add docs * Rename AxisHint and ButtonHint to GamepadAxis and GamepadButton * Add CHANGELOG entry * Update CHANGELOG.md * Add HidId and MovedAbsolute * Fix xinput deprecation warnings * Add ability to retrieve gamepad battery level * Fix weird imports in gamepad example * Update CHANGELOG.md * Resolve francesca64 comments
* Initial implementation * Corrected RAWINPUT buffer sizing * Mostly complete XInput implementation * XInput triggers * Add preliminary CHANGELOG entry. * match unix common API to evl 2.0 * wayland: eventloop2.0 * make EventLoopProxy require T: 'static * Revamp device event API, as well as several misc. fixes on Windows: * When you have multiple windows, you no longer receive duplicate device events * Mouse Device Events now send X-button input * Mouse Device Events now send horizontal scroll wheel input * Add MouseEvent documentation and Device ID debug passthrough * Improve type safety on get_raw_input_data * Remove button_id field from MouseEvent::Button in favor of �utton * Remove regex dependency on Windows * Remove axis filtering in XInput * Make gamepads not use lazy_static * Publicly expose gamepad rumble * Unstack DeviceEvent and fix examples/tests * Add HANDLE retrieval method to DeviceExtWindows * Add distinction between non-joystick axes and joystick axes. This helps with properly calculating the deadzone for controller joysticks. One potential issue is that the `Stick` variant isn't used for *all* joysticks, which could be potentially confusing - for example, raw input joysticks will never use the `Stick` variant because we don't understand the semantic meaning of raw input joystick axes. * Add ability to get gamepad port * Fix xinput controller hot swapping * Add functions for enumerating attached devices * Clamp input to [0.0, 1.0] on gamepad rumble * Expose gamepad rumble errors * Add method to check if device is still connected * Add docs * Rename AxisHint and ButtonHint to GamepadAxis and GamepadButton * Add CHANGELOG entry * Update CHANGELOG.md * Add HidId and MovedAbsolute * Fix xinput deprecation warnings * Add ability to retrieve gamepad battery level * Fix weird imports in gamepad example * Update CHANGELOG.md * Resolve francesca64 comments
* Initial implementation * Corrected RAWINPUT buffer sizing * Mostly complete XInput implementation * XInput triggers * Add preliminary CHANGELOG entry. * match unix common API to evl 2.0 * wayland: eventloop2.0 * make EventLoopProxy require T: 'static * Revamp device event API, as well as several misc. fixes on Windows: * When you have multiple windows, you no longer receive duplicate device events * Mouse Device Events now send X-button input * Mouse Device Events now send horizontal scroll wheel input * Add MouseEvent documentation and Device ID debug passthrough * Improve type safety on get_raw_input_data * Remove button_id field from MouseEvent::Button in favor of �utton * Remove regex dependency on Windows * Remove axis filtering in XInput * Make gamepads not use lazy_static * Publicly expose gamepad rumble * Unstack DeviceEvent and fix examples/tests * Add HANDLE retrieval method to DeviceExtWindows * Add distinction between non-joystick axes and joystick axes. This helps with properly calculating the deadzone for controller joysticks. One potential issue is that the `Stick` variant isn't used for *all* joysticks, which could be potentially confusing - for example, raw input joysticks will never use the `Stick` variant because we don't understand the semantic meaning of raw input joystick axes. * Add ability to get gamepad port * Fix xinput controller hot swapping * Add functions for enumerating attached devices * Clamp input to [0.0, 1.0] on gamepad rumble * Expose gamepad rumble errors * Add method to check if device is still connected * Add docs * Rename AxisHint and ButtonHint to GamepadAxis and GamepadButton * Add CHANGELOG entry * Update CHANGELOG.md * Add HidId and MovedAbsolute * Fix xinput deprecation warnings * Add ability to retrieve gamepad battery level * Fix weird imports in gamepad example * Update CHANGELOG.md * Resolve francesca64 comments
* Initial implementation * Corrected RAWINPUT buffer sizing * Mostly complete XInput implementation * XInput triggers * Add preliminary CHANGELOG entry. * match unix common API to evl 2.0 * wayland: eventloop2.0 * make EventLoopProxy require T: 'static * Revamp device event API, as well as several misc. fixes on Windows: * When you have multiple windows, you no longer receive duplicate device events * Mouse Device Events now send X-button input * Mouse Device Events now send horizontal scroll wheel input * Add MouseEvent documentation and Device ID debug passthrough * Improve type safety on get_raw_input_data * Remove button_id field from MouseEvent::Button in favor of �utton * Remove regex dependency on Windows * Remove axis filtering in XInput * Make gamepads not use lazy_static * Publicly expose gamepad rumble * Unstack DeviceEvent and fix examples/tests * Add HANDLE retrieval method to DeviceExtWindows * Add distinction between non-joystick axes and joystick axes. This helps with properly calculating the deadzone for controller joysticks. One potential issue is that the `Stick` variant isn't used for *all* joysticks, which could be potentially confusing - for example, raw input joysticks will never use the `Stick` variant because we don't understand the semantic meaning of raw input joystick axes. * Add ability to get gamepad port * Fix xinput controller hot swapping * Add functions for enumerating attached devices * Clamp input to [0.0, 1.0] on gamepad rumble * Expose gamepad rumble errors * Add method to check if device is still connected * Add docs * Rename AxisHint and ButtonHint to GamepadAxis and GamepadButton * Add CHANGELOG entry * Update CHANGELOG.md * Add HidId and MovedAbsolute * Fix xinput deprecation warnings * Add ability to retrieve gamepad battery level * Fix weird imports in gamepad example * Update CHANGELOG.md * Resolve francesca64 comments
CHANGELOG.mdif knowledge of this change could be valuable to usersThis started out as an attempt to merge @francesca64's gamepad integration (❤️) from #624 onto the EventLoop 2.0 branch, but ballooned to a complete overhaul of the device event API since it was the easiest way to expose a clean rumble API and we agreed that it needed to be done anyway.
This is a draft PR because it makes absolutely no attempt to to implement the device event changes on non-Windows platforms, which needs to be done before this can be merged. This also means that it's waiting on X11 and MacOS EL2.0 implementations, since meaningful work to get this working on those backends can't start until those backends work.
Feedback on the API is encouraged, and the API docs are available here. I expect the non-breaking changes to be fairly uncontroversial, and these are my reasons for the breaking changes:
DeviceEventintoMouseEvent,KeyboardEvent, andGamepadEvent:DeviceEventenum was a bit of a mess; there were several different event sources aggregated under one type, most of which could only be emitted from one specific kind of device. Splitting the API up clarifies which events get emitted by which types of devices and makes it more easily learnable.DeviceEvent::Textvariant:WindowEvent::ReceivedCharacterwas questionable.DeviceId:DeviceEvent. Also, I had to split offGamepadHandleso I could add arumblefunction, and splitting upMouseIdandKeyboardIdmakes the API more consistent.WindowEventvariants:Closes #750, closes #624, closes #119, closes #212 (?).