feat: port libuv threadpool #9
Conversation
| // TODO(?): need help | ||
| // Neither emscripten_dispatch_to_thread_async nor MAIN_THREAD_ASYNC_EM_ASM | ||
| // invoke the async complete callback if there is a printf() in worker thread. | ||
| // This breaks "packages/test/pool" tests. | ||
| // Not sure what happens, maybe has deadlock, | ||
| // and not sure whether this is Emscripten bug or my incorrect usage. | ||
| // BTW emscripten_dispatch_to_thread_async seems | ||
| // not support __wasm64__ V_I64 signature yet | ||
|
|
||
| // pthread_t main_thread = emscripten_main_browser_thread_id(); | ||
| // if (pthread_equal(main_thread, pthread_self())) { | ||
| // NEXT_TICK(callback, data); | ||
| // } else { | ||
| // emscripten_dispatch_to_thread_async(main_thread, | ||
| // __EMNAPI_ASYNC_SEND_CALLBACK_SIG, | ||
| // callback, | ||
| // NULL, | ||
| // data); | ||
| // // or | ||
| // // MAIN_THREAD_ASYNC_EM_ASM({ | ||
| // // emnapiGetDynamicCalls.call_vp($0, $1); | ||
| // // }, callback, data); | ||
| // } |
There was a problem hiding this comment.
@RReverser I left comments here. If you prefer that emnapi should not depend on PThread object, I would appreciate it if you could open another PR to provide help. Sorry for my broken English, forget it if there is some strange or offensive expression.
There was a problem hiding this comment.
@RReverser c3a3cdb1 At last I found emscripten_proxy_async and em_proxying_queue_create works well, but using emscripten_proxy_get_system_queue() will also stuck if call printf in child thread. I could not figure out why.
There was a problem hiding this comment.
If you prefer that emnapi should not depend on PThread object, I would appreciate it if you could open another PR to provide help
Heh unfortunately I'm finding a bit hard to follow how threading works in this library, so not sure I can submit a PR for that yet.
There was a problem hiding this comment.
FWIW one other reason I want it to have its own pool of workers is that upstream there are talks of making it possible to use pthreads without a pthread pool, and I now have a working implementation of that at least for Node.js.
However, it seems that emnapi won't work with that approach in the current form since it manipulates internal arrays and waits until unusedWorkers is not empty. But, if Emscripten will use pthreads without a pthread pool, it will be empty at the start so emnapi will wait forever.
There was a problem hiding this comment.
Ah, looks like I had a slightly outdated dist version and you already removed some more usages. This is looking promising, going to give it a try with the pool-less version of Emscripten.
Fixes #8