API contract for wgpuInstanceProcessEvents() is unclear

[litherum]

Its not clear whether it supposed to block or not, and how it integrates with an already-existing runloop.

[kvark]

In wgpu-native this call (also referred to as Tick() sometimes) would do the following:

• check in-flight GPU work and free resources no longer used by GPU
• issue mapping callbacks
• not block on any GPU work

[Kangz]

Same for Dawn (although it's currently wgpu::Device::Tick), it:

• Checks for completion of GPU commands.
• Fires any callbacks that need firing.
• Frees resources.
• (does an implicit submit of pending commands, but that's unobservable).
• Returns immediately without blocking.

[litherum]

I suppose it has to be reentrant, then, huh.

[Kangz]

Reentrant? But yeah during the processing you have to support more callbacks being added to the queue (and first move the queue, then process it, to avoid iterator invalidation)

[kainino0x]

I don't think wgpuInstanceProcessEvents() needs to be callable from within a callback running inside wgpuInstanceProcessEvents(). But yes to the stuff Corentin said.