Merge to M40:

Reland 720713002: Avoid getting CoreAudio into bad state.

This patch fixes two issues:
1) in audio_low_latency_input_mac.cc, when the number_of_frames is changed on the fly, previously we feed a audio_buffer_list which uses the original number_of_frames_ to CoreAudio, this will put CoreAudio into a bad state, and the consequence is that if the new input stream is trying to use a larger buffer size than what the previous input stream was using, AudioUnitRender() will return -50 for the new stream. Thus it leads to no input audio issue for webrtc.

2) gUM on Mac will create an input stream using 128 samples as buffer size at the beginning, and will re-create the stream using 10ms when a peer connection is connected to a audio track. It discards if audio processing is enabled or not. This patch avoids creating the first stream using 128 samples if audio processing is enabled.

TBR=tommi@chromium.org,henrika@chromium.org

BUG=428706
TEST=https://webrtc-phone.appspot.com changes chrome3940Workaround to false, and make a call lots of time. Verify that the input stream always works.

Review URL: https://codereview.chromium.org/721923002

Cr-Commit-Position: refs/heads/master@{#303909}
(cherry picked from commit 33956f7d3b46358ba7bce3e0ff4e9cecd1e39b0a)

Review URL: https://codereview.chromium.org/730453003

Cr-Commit-Position: refs/branch-heads/2214@{#59}
Cr-Branched-From: 03655fd3f6d72165dc3c9bd2c89807305316fe6c-refs/heads/master@{#303346}

0 files changed, 0 insertions, 0 deletions