Run atexit handlers before terminating all threads

Although not defined by the POSIX standard, it seems that on most
systems, atexit handlers are run before threads are terminated.

Author: kleisauke

src/postamble.js

@@ -453,10 +453,6 @@ function exit(status, implicit) {
     }
 #endif // ASSERTIONS
   } else {
-#if USE_PTHREADS
-    PThread.terminateAllThreads();
-#endif
-
     exitRuntime();
 
 #if expectToReceiveOnModule('onExit')

src/postamble_minimal.js

@@ -30,6 +30,9 @@ function run() {
   callRuntimeCallbacks(__ATEXIT__);
   <<< ATEXITS >>>
 #endif
+#if USE_PTHREADS
+  PThread.terminateAllThreads();
+#endif
 
 #if IN_TEST_HARNESS
   // fflush() filesystem stdio for test harness, since there are existing

src/preamble.js

@@ -432,6 +432,9 @@ function exitRuntime() {
 #if EXIT_RUNTIME
   callRuntimeCallbacks(__ATEXIT__);
   <<< ATEXITS >>>
+#endif
+#if USE_PTHREADS
+  PThread.terminateAllThreads();
 #endif
   runtimeExited = true;
 }

tests/pthread/test_pthread_atexit.c

@@ -0,0 +1,42 @@
+#include <assert.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "emscripten.h"
+#include "emscripten/threading.h"
+
+static _Atomic int threadCounter = 0;
+static _Atomic int running = 1;
+static pthread_t thread;
+
+void *workerThread(void* arg) {
+  threadCounter++;
+
+  while (running)
+    emscripten_thread_sleep(1000);
+
+  threadCounter--;
+
+  return NULL;
+}
+
+void terminateThread() {
+  running = 0;
+
+  int res = 0;
+  int rc = pthread_join(thread, (void**)&res);
+  assert(rc == 0);
+  assert(res == 0);
+
+  printf("done waiting - counter is: %d\n", threadCounter);
+}
+
+int main(int argc, char* argv[]) {
+  int rc = atexit(terminateThread);
+  assert(rc == 0);
+
+  rc = pthread_create(&thread, NULL, workerThread, NULL);
+  assert(rc == 0);
+  return 0;
+}

tests/pthread/test_pthread_atexit.out

@@ -0,0 +1 @@
+done waiting - counter is: 0

tests/test_core.py

@@ -2323,6 +2323,12 @@ def test_pthread_equal(self):
   def test_pthread_dispatch_after_exit(self):
     self.do_run_in_out_file_test('pthread/test_pthread_dispatch_after_exit.c')
 
+  @node_pthreads
+  def test_pthread_atexit(self):
+    self.set_setting('EXIT_RUNTIME')
+    self.set_setting('PTHREAD_POOL_SIZE', 1)
+    self.do_run_in_out_file_test('pthread/test_pthread_atexit.c')
+
   @node_pthreads
   def test_pthread_nested_work_queue(self):
     self.set_setting('EXIT_RUNTIME')