Expose winit::platform::emscripten::set_main_loop_callback

[tangmi]

Hi, it seems the only way to use winit on WASM is to use EventLoop::run_forever, which doesn't seem to give the user an opportunity to do work (e.g. render) if no events have come through. Doing the normal loop { events_loop.poll_events(/* ... */) } seems to cause the tab to hang, since we're not getting that sweet, sweet requestAnimationFrame (I'm assuming).

I'm wondering if it'd be possible to export winit::platform::emscripten::set_main_loop_callback publicly, so one can write a main loop for emscripten. (Maybe even better would be to add a method something like pub fn loop<F>(&mut self, mut per_frame: F) where F: FnMut() -> ControlFlow; that could be implemented with set_main_loop_callback on emscripten and something like while per_frame() != ControlFlow::Break {} everywhere else.)

Cheers!

[Osspial]

Thanks for the issue!

The run_forever and poll_events methods are actually implemented with emscripten_set_main_loop. That's actually why loop { events_loop.poll_events(/* ... */) } doesn't work - emscripten_set_main_loop needs your application's main function to return before the browser can run the callback, and Emscripten accomplishes this by throwing a JS exception to unwind the stack. That causes poll_events to never return, what with your stack no longer existing.

However, work is ongoing to address your API concerns! @ryanisaacg has been doing great working getting a WASM backend going with the new event loop API through stdweb (see #797). Unfortunately that isn't quite done yet, but it's rapidly getting there and I expect to see it merged sooner rather than later.

You can check out the API docs for the new API here, and I believe that addresses your issue; you can get periodically woken up for rendering with the new EventLoop::run method and the redesigned ControlFlow enum. Right now we're waiting on the X11/macOS backends to be completed before those changes get on crates.io, but I unfortunately don't have any solid timeline for when those will get done - work on the macOS backend is apparently largely done but frustratingly has stalled for reasons I am not aware of, and I don't know if anybody is currently working on the X11 backend.

[tangmi]

Sounds good, perhaps we should close this issue in favor of eventloop 2.0?

I’m happy to take a stab at X11 (I don’t have a Linux box, but that’s what VMs are for, I suppose), especially if your Windows PR can be used as a reference.

[Osspial]

@vberger you mentioned that you might be doing the X11 backend in this comment - has that gone anywhere yet? If that work hasn't started, @tangmi you should be good to go. Thanks for the offer!

[elinorbgr]

I haven't started yet, I was thinking aybe getting at it in the coming days, though if you want to try & tackle it @tangmi , please do.

Just be aware that the x11 code is very different from the Windows code, and it also currently contains a lot of work-arounds to compensate for X11 gotchas...

[tangmi]

Thanks for the heads up. I'll take a look at it, although no promises yet (ran into issues trying to install a linux system...).

[goddessfreya]

Hey @tangmi, are you still working on X11 support? How far did you get?

[tangmi]

Unfortunately no real progress. I don't have a linux machine or any machine powerful enough to reasonably run a VM :(.

I ended up finding emscripten-specific workarounds to make my project work.