Add EventLoopExtPumpEvents and EventLoopExtRunOnDemand

This adds two new extensions for running a Winit event loop which will
replace `EventLoopExtRunReturn`

The `run_return` API is trying to solve multiple problems and address
multiple, unrelated, use cases but in doing so it is not succeeding
at addressing any of them fully.

The notable use cases we have are:
1. Applications want to be able to implement their own external
   event loop and call some Winit API to poll / pump events, once
   per iteration of their own loop, without blocking the outer,
   external loop. Addressing rust-windowing#2706
2. Applications want to be able to re-run separate instantiations
   of some Winit-based GUI and want to allow the event loop to exit with
   a status, and then later be able to run the loop again for a new
   instantiation of their GUI. Addressing rust-windowing#2431

It's very notable that these use cases can't be supported across
all platforms and so they are extensions, similar to
`EventLoopExtRunReturn`

The intention is to support these extensions on:
- Windows
- Linux (X11 + Wayland)
- macOS
- Android

These extensions aren't compatible with Web or iOS though.

Each method of running the loop will behave consistently in terms of how
`NewEvents(Init)`, `Resumed` and `LoopDestroyed` events are dispatched
(so portable application code wouldn't necessarily need to have any awareness
of which method of running the loop was being used)

Once all backends have support for these extensions then we can
remove `EventLoopExtRunReturn`

For simplicitly, the extensions are documented with the assumption that
the above platforms will be supported.

This patch makes no functional change, it only introduces these new
extensions so we can then handle adding platform-specific backends
in separate pull requests, so the work can be landed in stages.

Author: rib

src/error.rs

@@ -9,6 +9,10 @@ pub enum ExternalError {
     NotSupported(NotSupportedError),
     /// The OS cannot perform the operation.
     Os(OsError),
+    /// The event loop can't be re-run while it's already running
+    AlreadyRunning,
+    /// Application has exit with an error status.
+    ExitFailure(i32),
 }
 
 /// The error type for when the requested operation is not supported by the backend.
@@ -59,8 +63,10 @@ impl fmt::Display for OsError {
 impl fmt::Display for ExternalError {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
         match self {
+            ExternalError::AlreadyRunning => write!(f, "EventLoop is already running"),
             ExternalError::NotSupported(e) => e.fmt(f),
             ExternalError::Os(e) => e.fmt(f),
+            ExternalError::ExitFailure(status) => write!(f, "Exit Failure: {status}"),
         }
     }
 }

src/event.rs

@@ -329,6 +329,14 @@ pub enum StartCause {
     Init,
 }
 
+/// The return status for `pump_events`
+pub enum PumpStatus {
+    /// Continue running external loop
+    Continue,
+    /// exit external loop
+    Exit(i32),
+}
+
 /// Describes an event from a [`Window`].
 #[derive(Debug, PartialEq)]
 pub enum WindowEvent<'a> {

src/platform/pump_events.rs

@@ -0,0 +1,152 @@
+use crate::{
+    event::{Event, PumpStatus},
+    event_loop::{ControlFlow, EventLoop, EventLoopWindowTarget},
+};
+
+/// Additional methods on [`EventLoop`] for pumping events within an external event loop
+pub trait EventLoopExtPumpEvents {
+    /// A type provided by the user that can be passed through [`Event::UserEvent`].
+    type UserEvent;
+
+    fn pump_events<F>(&mut self, event_handler: F) -> PumpStatus
+    where
+        F: FnMut(
+            Event<'_, Self::UserEvent>,
+            &EventLoopWindowTarget<Self::UserEvent>,
+            &mut ControlFlow,
+        );
+}
+
+impl<T> EventLoopExtPumpEvents for EventLoop<T> {
+    type UserEvent = T;
+
+    /// Pump the `EventLoop` to check for and dispatch pending events, without blocking
+    ///
+    /// This API is designed to enable applications to integrate Winit into an
+    /// external event loop, for platforms that can support this.
+    ///
+    /// ```rust
+    /// use winit::{event::{Event, PumpStatus, WindowEvent}, event_loop::EventLoop, window::WindowBuilder };
+    /// use winit::platform::pump_events::EventLoopExtPumpEvents;
+    /// let mut event_loop = EventLoop::new();
+    ///
+    /// let window = WindowBuilder::new()
+    ///     .with_title("A fantastic window!")
+    ///     .build(&event_loop)
+    ///     .unwrap();
+    ///
+    /// 'main: loop {
+    ///     let status = event_loop.pump_events(|event, _, control_flow| {
+    ///         match event {
+    ///             Event::WindowEvent {
+    ///                 event: WindowEvent::CloseRequested,
+    ///                 window_id,
+    ///             } if window_id == window.id() => control_flow.set_exit(),
+    ///             Event::MainEventsCleared => {
+    ///                 window.request_redraw();
+    ///             }
+    ///             _ => (),
+    ///         }
+    ///     });
+    ///     if let PumpStatus::Exit(exit_code) = status {
+    ///         break 'main ExitCode::from(exit_code as u8);
+    ///     }
+    ///
+    ///     // Sleep for 1/60 second to simulate application work
+    ///     //
+    ///     // Since `pump_events` doesn't block it will be important to
+    ///     // throttle the loop in the app somehow.
+    ///     println!("Update()");
+    ///     sleep(Duration::from_millis(16));
+    /// }
+    /// ```
+    ///
+    /// **Note:** This is not a portable API, and its usage involves a number of
+    /// caveats and trade offs that should be considered before using this API!
+    ///
+    /// You almost certainly shouldn't use this API, unless you absolutely know it's
+    /// the only practical option you have.
+    ///
+    /// # Synchronous events
+    ///
+    /// Some events _must_ only be handled synchronously via the closure that
+    /// is passed to Winit so that the handler will also be synchronized with
+    /// the window system and operating system.
+    ///
+    /// This is because some events are driven by a window system callback
+    /// where the window systems expects the application to have handled the
+    /// event before returning.
+    ///
+    /// **These events can not be buffered and handled outside of the closure
+    /// passed to Winit**
+    ///
+    /// As a general rule it is not recommended to ever buffer events to handle
+    /// them outside of the closure passed to Winit since it's difficult to
+    /// provide guarantees about which events are safe to buffer across all
+    /// operating systems.
+    ///
+    /// Notable events that will certainly create portability problems if
+    /// buffered and handled outside of Winit include:
+    /// - `RenderRequested` events, used to schedule rendering.
+    ///
+    ///     macOS for example uses a `drawRect` callback to drive rendering
+    /// within applications and expects rendering to be finished before
+    /// the `drawRect` callback returns.
+    ///
+    ///     For portability it's strongly recommended that applications should
+    /// keep their rendering inside the closure provided to Winit.
+    /// - Any lifecycle events, such as `Suspended` / `Resumed`.
+    ///
+    ///     The handling of these events needs to be synchronized with the
+    /// operating system and it would never be appropriate to buffer a
+    /// notification that your application has been suspended or resumed and
+    /// then handled that later since there would always be a chance that
+    /// other lifecycle events occur while the event is buffered.
+    ///
+    /// # Supported Platforms
+    /// - Windows
+    /// - Linux
+    /// - MacOS
+    /// - Android
+    ///
+    /// # Unsupported Platforms
+    /// - Web:  This API is fundamentally incompatible with the event-based way in which
+    /// Web browsers work because it's not possible to have a long-running external
+    /// loop that would block the browser and there is nothing that can be
+    /// polled to ask for new new events. Events are delivered via callbacks based
+    /// on an event loop that is internal to the browser itself.
+    /// - iOS: It's not possible to stop and start an `NSApplication` repeatedly on iOS so
+    /// there's no way to support the same approach to polling as on MacOS.
+    ///
+    /// # Platform-specific
+    /// - **Windows**: The implementation will use `PeekMessage` when checking for
+    ///   window messages to avoid blocking your external event loop.
+    ///
+    /// - **MacOS**: The implementation works in terms of stopping the global `NSApp`
+    ///   whenever the application `RunLoop` indicates that it is preparing to block
+    ///   and wait for new events.
+    ///
+    ///   This is very different to the polling APIs that are available on other
+    ///   platforms (the lower level polling primitives on MacOS are private
+    ///   implementation details for `NSApp` which aren't accessible to application
+    ///   developers)
+    ///
+    ///   It's likely this will be less efficient than polling on other OSs and
+    ///   it also means the `NSApp` is stopped while outside of the Winit
+    ///   event loop - and that's observable (for example to crates like `rfd`)
+    ///   because the `NSApp` is global state.
+    ///
+    ///   If you render outside of Winit you are likely to see window resizing artifacts
+    ///   since MacOS expects applications to render synchronously during any `drawRect`
+    ///   callback.
+    fn pump_events<F>(&mut self, event_handler: F) -> PumpStatus
+    where
+        F: FnMut(
+            Event<'_, Self::UserEvent>,
+            &EventLoopWindowTarget<Self::UserEvent>,
+            &mut ControlFlow,
+        ),
+    {
+        self.event_loop.pump_events(event_handler)
+    }
+}

src/platform/run_ondemand.rs

@@ -0,0 +1,79 @@
+use crate::{
+    error::ExternalError,
+    event::Event,
+    event_loop::{ControlFlow, EventLoop, EventLoopWindowTarget},
+};
+
+/// Additional methods on [`EventLoop`] to return control flow to the caller.
+pub trait EventLoopExtRunOnDemand {
+    /// A type provided by the user that can be passed through [`Event::UserEvent`].
+    type UserEvent;
+
+    /// Runs the event loop in the calling thread and calls the given `event_handler` closure
+    /// to dispatch any window system events.
+    ///
+    /// Unlike [`EventLoop::run`], this function accepts non-`'static` (i.e. non-`move`) closures
+    /// and it is possible to return control back to the caller without
+    /// consuming the `EventLoop` (by setting the `control_flow` to [`ControlFlow::Exit`]) and
+    /// so the event loop can be re-run after it has exit.
+    ///
+    /// It's expected that each run of the loop will be for orthogonal instantiations of your
+    /// Winit application, but internally each instantiation may re-use some common window
+    /// system resources, such as a display server connection.
+    ///
+    /// This API is not designed to run an event loop in bursts that you can exit from and return
+    /// to while maintaining the full state of your application. (If you need something like this
+    /// you can look at the [`pump_events()`] API)
+    ///
+    /// Each time `run_ondemand` is called the `event_handler` can expect to receive a
+    /// `NewEvents(Init)` and `Resumed` event (even on platforms that have no suspend/resume
+    /// lifecycle) - which can be used to consistently initialize application state.
+    ///
+    /// See the [`ControlFlow`] docs for information on how changes to `&mut ControlFlow` impact the
+    /// event loop's behavior.
+    ///
+    /// # Caveats
+    /// - This extension isn't available on all platforms, since it's not always possible to
+    ///   return to the caller (specifically this is impossible on iOS and Web - though with
+    ///   the Web backend it is possible to use `spawn()` more than once instead).
+    /// - No [`Window`] state can be carried between separate runs of the event loop.
+    ///
+    /// You are strongly encouraged to use `run` for portability, unless you specifically need
+    /// the ability to re-run a single event loop more than once
+    ///
+    /// # Supported Platforms
+    /// - Windows
+    /// - Linux
+    /// - macOS
+    /// - Android
+    ///
+    /// # Unsupported Platforms
+    /// - Web:  This API is fundamentally incompatible with the event-based way in which
+    /// Web browsers work because it's not possible to have a long-running external
+    /// loop that would block the browser and there is nothing that can be
+    /// polled to ask for new new events. Events are delivered via callbacks based
+    /// on an event loop that is internal to the browser itself.
+    /// - iOS: It's not possible to stop and start an `NSApplication` repeatedly on iOS.
+    fn run_ondemand<F>(&mut self, event_handler: F) -> Result<(), ExternalError>
+    where
+        F: FnMut(
+            Event<'_, Self::UserEvent>,
+            &EventLoopWindowTarget<Self::UserEvent>,
+            &mut ControlFlow,
+        );
+}
+
+impl<T> EventLoopExtRunOnDemand for EventLoop<T> {
+    type UserEvent = T;
+
+    fn run_ondemand<F>(&mut self, event_handler: F) -> Result<(), ExternalError>
+    where
+        F: FnMut(
+            Event<'_, Self::UserEvent>,
+            &EventLoopWindowTarget<Self::UserEvent>,
+            &mut ControlFlow,
+        ),
+    {
+        self.event_loop.run_ondemand(event_handler)
+    }
+}