Fix the frame freezing issue observed during a config change in the H.264 decoder on Windows

Fix the frame freezing issue observed during a config change in the H.264 decoder on Windows

We need to drain the decoder of any pending output before tearing down the
instance of the decoder and recreating it during a configuration change.

Draining the decoder is done as part of a Flush operation. However this is
a decoder initiated flush which requires the following
1. Don't process pending input in this case as that is what caused the
   config change.
2. Don't notify the client about the completion of the flush operation.

Added a member variable processing_config_change_ to track this state. We
mimic most of the logic we have to process the pending_flush_ state when
this flag is set.

BUG=616349
Committed: https://crrev.com/ec68dd1bde3f2049975d8aa194525983d4a93688
Cr-Commit-Position: refs/heads/master@{#400023}

[ananta, 2016-06-15 02:53:39]

ananta@chromium.org changed reviewers:
+ sandersd@chromium.org

[sandersd (OOO until July 31), 2016-06-15 18:50:55]

I didn't re-read the entire state machine so I wanted to confirm with you that
these two cases work correctly:
  - config change while flushing; if a client calls Decode() while a flush is
still ongoing (I believe Flash does this). This code is correct if that decode
call is simply queued.
  - Flush while config changing; if a client calls Flush() while a config change
Decode() is still ongoing. It's probably fine to process this as a regular
flush.

[ananta, 2016-06-15 20:00:51]

On 2016/06/15 18:50:55, sandersd wrote:
> I didn't re-read the entire state machine so I wanted to confirm with you that
> these two cases work correctly:
>   - config change while flushing; if a client calls Decode() while a flush is
> still ongoing (I believe Flash does this). This code is correct if that decode
> call is simply queued.
>   - Flush while config changing; if a client calls Flush() while a config
change
> Decode() is still ongoing. It's probably fine to process this as a regular
> flush.

Thanks for the comments. I changed the code to reset the config changed flag if
we receive a Flush
with a comment in the code.

[ananta, 2016-06-15 20:39:40]

As per our discussion update the comment in the Flush function to look into
whether a flush would fix the problems with the decoder on windows while
processing config changes.

[sandersd (OOO until July 31), 2016-06-15 20:40:30]

lgtm

[ananta, 2016-06-15 21:11:51]

The CQ bit was checked by ananta@chromium.org

[commit-bot: I haz the power, 2016-06-15 21:12:29]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/2065323002/80001

[commit-bot: I haz the power, 2016-06-15 22:09:12]

Committed patchset #5 (id:80001)

[commit-bot: I haz the power, 2016-06-15 22:09:16]

CQ bit was unchecked

[commit-bot: I haz the power, 2016-06-15 22:12:37]

Description was changed from

==========
Fix the frame freezing issue observed during a config change in the H.264
decoder on Windows

We need to drain the decoder of any pending output before tearing down the
instance of the decoder and recreating it during a configuration change.

Draining the decoder is done as part of a Flush operation. However this is
a decoder initiated flush which requires the following
1. Don't process pending input in this case as that is what caused the
   config change.
2. Don't notify the client about the completion of the flush operation.

Added a member variable processing_config_change_ to track this state. We
mimic most of the logic we have to process the pending_flush_ state when
this flag is set.

BUG=616349
==========

to

==========
Fix the frame freezing issue observed during a config change in the H.264
decoder on Windows

We need to drain the decoder of any pending output before tearing down the
instance of the decoder and recreating it during a configuration change.

Draining the decoder is done as part of a Flush operation. However this is
a decoder initiated flush which requires the following
1. Don't process pending input in this case as that is what caused the
   config change.
2. Don't notify the client about the completion of the flush operation.

Added a member variable processing_config_change_ to track this state. We
mimic most of the logic we have to process the pending_flush_ state when
this flag is set.

BUG=616349

Committed: https://crrev.com/ec68dd1bde3f2049975d8aa194525983d4a93688
Cr-Commit-Position: refs/heads/master@{#400023}
==========

[commit-bot: I haz the power, 2016-06-15 22:12:38]

Patchset 5 (id:??) landed as
https://crrev.com/ec68dd1bde3f2049975d8aa194525983d4a93688
Cr-Commit-Position: refs/heads/master@{#400023}