Fixing flakiness of TextureLayerChangeInvisibleMailboxTest

cc/layers/texture_layer.cc

 // Copyright 2010 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "cc/layers/texture_layer.h"
 
 #include "base/bind.h"
 #include "base/callback_helpers.h"
 #include "base/location.h"
 #include "base/memory/ptr_util.h"
@@ skipping 281 matching lines @@
     mailbox_ = TextureMailbox();
     release_callback_ = nullptr;
   }
 }
 
 void TextureLayer::TextureMailboxHolder::ReturnAndReleaseOnImplThread(
     const gpu::SyncToken& sync_token,
     bool is_lost,
     BlockingTaskRunner* main_thread_task_runner) {
   Return(sync_token, is_lost);
-  main_thread_task_runner->PostTask(
-      FROM_HERE, base::Bind(&TextureMailboxHolder::InternalRelease, this));
+  if (main_thread_task_runner->BelongsToCurrentThread()) {

[danakj, 2017/04/04 21:59:37]

You're right this orders it right, but I usually avoid this pattern of

if (on thread) { A } else { post to thread(A) }

The reason is that things that are expecting to be on a fresh call stack
suddenly are only sometimes on a fresh callstack which can lead to unexpected
re-entrancy or other bad side effects (like for instance a recent bug where
during paint we ended up destroying the compositor, which would then crash when
it tried to continue painting but it's destroyed).

So I'd prefer that we keep posting here, and just drop the ack if the frame sink
is released. wdyt?

[Saman Sami, 2017/04/04 22:12:44]

I see. Yes, dropping the ack should work. I sent out a patch doing that. Let's
see how the bots feel.

+    InternalRelease();
+  } else {
+    main_thread_task_runner->PostTask(
+        FROM_HERE, base::Bind(&TextureMailboxHolder::InternalRelease, this));
+  }
 }
 
 }  // namespace cc