Add emscripten_queue_microtask() API #25741
Conversation
juj
force-pushed
the
emscripten_queue_microtask
branch
from
35f230a to
91daa78
Compare
|
I had a version of this API in #24481... but sure how it compares |
test/test_browser.py
Outdated
| self.btest_exit('emscripten_queue_microtask.c') | ||
|
|
||
| @requires_shared_array_buffer | ||
| def test_emscripten_queue_microtask_pthread(self): |
There was a problem hiding this comment.
Other tests in this file do it this way:
@parameterized({
'': ([],),
'proxy_to_pthread': (['-pthread', '-sPROXY_TO_PTHREAD'],),
})
There was a problem hiding this comment.
The reason I didn't, was to get the @requires_shared_array_buffer annotation be precise.
Though now I remember I have a pthread magic regex catch for the same purpose, so changed to this form.
There was a problem hiding this comment.
I think we should be consistent and use @parameterize everywhere.
There must be magic handling that injects @requires_shared_array_buffer I suppose? If you don't like that we could make a new @also_with_proxy_to_pthread that takes care of adding the cflags and also adding requires_shared_array_buffer.
src/lib/libhtml5.js
Outdated
|
|
||
| emscripten_queue_microtask: (cb, userData) => { | ||
| queueMicrotask(() => { | ||
| {{{ makeDynCall('vp', 'cb') }}}(userData); |
There was a problem hiding this comment.
This probably needs to be wrapped in callUserCallback.
There was a problem hiding this comment.
Yeah.. Both the push and pop callback and callUserCallback are nice for some uses, but result in bloat :/
Added these, maybe callUserCallback() can be made zero-cost some day. (Or maybe it is so at least with Closure, not sure)
There was a problem hiding this comment.
We do need callUserCallback whenever we enter user code from the event loop otherwise lots of things can break. A classic one is calling exit() from within the callback.
IIRC the overhead of callUserCallback in MINIMAL_RUNTIME, where we don't have the same requirements is very close to zero. If you want to make reduce it to absolutely zero that seems like a good thing, but unrelated to this specific change.
Maybe just copy the version I had in that PR? Should probably live in libeventloop.js too. I was recently chatting to someone on chrome team about why we this API event exists given that its pretty much indistinguishable from |
| void emscripten_cancel_animation_frame(int requestAnimationFrameId); | ||
| void emscripten_request_animation_frame_loop(bool (*cb)(double time, void *userData), void *userData); | ||
|
|
||
| void emscripten_queue_microtask(void (*cb)(void *userData), void *userData); |
There was a problem hiding this comment.
This should probably go in emscripten/eventloop.h.
2 participants
Add emscripten_queue_microtask() API to enable use case described in this thread: #25670 (comment).