Record async "task backtraces"

base/threading/worker_pool_posix.cc

Index: base/threading/worker_pool_posix.cc
diff --git a/base/threading/worker_pool_posix.cc b/base/threading/worker_pool_posix.cc
index 32a5e92b91c7fa99f241ead417fcc231d70ca539..a10ac8e05139afead4614e26cab29e86ce80e990 100644
--- a/base/threading/worker_pool_posix.cc
+++ b/base/threading/worker_pool_posix.cc
@@ -140,7 +140,9 @@ void PosixDynamicThreadPool::Terminate() {
 void PosixDynamicThreadPool::PostTask(
     const tracked_objects::Location& from_here,
     const base::Closure& task) {
-  PendingTask pending_task(from_here, task);
+  PendingTask pending_task(from_here,
+                           MessageLoop::current()->current_pending_task_,

[danakj, 2015/04/03 17:18:35]

The use of MessageLoop::current()->current_pending_task_ worries me. I think
maybe we need to ask for this off the MessageLoop's task_runner() instead. But
I'm not totally sure..

It's easy to get lost in MessageLoop. Ryan do you have a good position to
comment on this being right/wrong?

I can dig a whole bunch more if you don't have some guidance here :)

+                           task);
   AddTask(&pending_task);
 }
 
@@ -168,7 +170,7 @@ PendingTask PosixDynamicThreadPool::WaitForTask() {
   AutoLock locked(lock_);
 
   if (terminated_)
-    return PendingTask(FROM_HERE, base::Closure());
+    return PendingTask(FROM_HERE, NULL, base::Closure());
 
   if (pending_tasks_.empty()) {  // No work available, wait for work.
     num_idle_threads_++;
@@ -182,7 +184,7 @@ PendingTask PosixDynamicThreadPool::WaitForTask() {
     if (pending_tasks_.empty()) {
       // We waited for work, but there's still no work.  Return NULL to signal
       // the thread to terminate.
-      return PendingTask(FROM_HERE, base::Closure());
+      return PendingTask(FROM_HERE, NULL, base::Closure());
     }
   }