Worker. WASM C code ending callback

[DiachenkoIgor]

Hi, when I create thread in C, new worker is created. JS script with name *.worker.js call function 'Module['invokeEntryPoint']' which start C procedure for given thread. However, sometimes this function is blocking and releases after C code execution end , another time function is released without ending C code. Can some, please, explain how to predict behaviour of this function or advice other way of defining end of execution of C code in worker?

[sbc100]

Can you give some more details. What does "function is released without ending C code" mean? Can you give an example? Can you share the link flags you are using?

[DiachenkoIgor]

Hi, I reproduced such behaviour.
test.cpp

#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>
#include <unistd.h>
extern "C" int test_asyncify();
void *print_message_function( void *ptr );

int main()
{
     pthread_t thread1;

     pthread_create( &thread1, NULL, print_message_function, (void*) NULL);

     pthread_join(thread1, NULL);

     printf("END...\n");

     return 0;
}

void *print_message_function( void *ptr )
{
     for(int i = 0; i < 5; i++){
          printf("print_message...\n");
          sleep(1);
     }
     test_asyncify();

     printf("After test_asyncify...\n");

     return NULL;
}

test_library.js

mergeInto(LibraryManager.library, {

test_asyncify: function () {
 return Asyncify.handleAsync(async ()=>{await new Promise(resolve => {
    setTimeout(() => {
      resolve('resolved');
    }, 5000);
  })});
}
});

compile command
emcc test.cpp -s PROXY_TO_PTHREAD -s ASYNCIFY -s ASYNCIFY_IMPORTS=test_asyncify --js-library ./test_library.js -sPTHREAD_POOL_SIZE=1 -sUSE_PTHREADS=1 -pthread -o test.html

pthread_join return before thread method end execution.

[sbc100]

Ah yes, this looks like an issue with ASYNCIFY. The way ASYNCIFY works is that it re-writes the wasm file to return early on each suspend and then ti re-winds to where it left off after the suspend. I'm not sure ASYNCIFY works with pthreads today, If this is true we should make this more clear. @kripken do you know if ASYNCIFY is supposed to work with pthreads today?

[kripken]

I don't believe we've tested asyncify+pthreads so far. Very possible things are not working yet, though I can't think offhand of a particular issue relevant here.

[DiachenkoIgor]

Is there other way for implementation such tasks?

[mmarczell-graphisoft]

I think we've tried asyncify + pthreads at my company and the only thing that prevented us from using it in production was #16291 and that was since fixed.