GitLab

Using unretained for decoder/selector callbacks is generally not safe /
fragile. For ex, some decoders (e.g. those that offload) will call the
output callback after destruction. Another example: decoder selector
internally posts the Select() callback, so it may run after destruction.

Additionally, sending incoming callbacks as CrossThreadFunctions for
invoking on the media thread is not safe. We may destruct while some
callbacks are still pending, destroying the callbacks internal state
(including bindings to blink GC types) off the main thread. Now we
ensure that all callbacks are saved on the main thread and we extend
the "client" interfaces of the broker for proxying the callbacks safely.

Bug: 1120431

Change-Id: I2ccfe46982cda9b18e47a56424b25f54f8877883
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2366661
Commit-Queue: Chrome Cunningham <chcunningham@chromium.org>
Auto-Submit: Chrome Cunningham <chcunningham@chromium.org>
Reviewed-by: Dan Sanders <sandersd@chromium.org>
Cr-Commit-Position: refs/heads/master@{#800776}