| commit | 395f6fd0ad5c1a754c12933ae3a1556ef1b95493 | |
| author | tommi@chromium.org <tommi@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98> | |
| Thu, 27 Feb 2014 12:55:13 +0000 (27 12:55 +0000) | ||
| committer | tommi@chromium.org <tommi@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98> | |
| Thu, 27 Feb 2014 12:55:13 +0000 (27 12:55 +0000) | ||
| tree | 2d6bb0d5bda5975210318c81941568d63ed91bc3 | treesnapshot (tar.gz zip) |
| parent | d0c382d108aea041fc6eb8861f8cda0574c2c0fe | commitdiff |
Make sure we don't call DisableKeyPressMonitoring without calling EnableKeyPressMonitoring first.
If we fail to open an audio device, AudioInputDeviceController can call DisableKeyPressMonitoring() and bring the monitoring reference count to below 0. The state of the AudioInputDeviceController instance will be |kEmpty| in this case but the DoClose() code only checks for kClosed.
The class has 2 states that aren't really necessary. kError is never used, so I'm just removing it. kEmpty is only used from construction until successful initialization. It is this state (kEmpty) that is never checked for and can cause us to call DisableKeyPressMonitoring inside DoClose without having first called EnableKeyPressMonitoring.
Since I'm cleaning up the states, I'm also fixing them as far as Chromium style goes.
BUG=347129
TEST=I ran into this while opening apprtc.appspot.com with a debug build of Chrome on Windows where my 'default communication device' was not the same as the 'default device'. These two different devices furthermore did not have the same sample rates, which caused opening the communication device with the incorrect parameters to fail. That's when I hit a DCHECK inside DisableKeyPressMonitoring where key_press_counter_references_ was being set to -1.
Review URL: https://codereview.chromium.org/181553003
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@253762 0039d316-1c4b-4281-b951-d872f2087c98
If we fail to open an audio device, AudioInputDeviceController can call DisableKeyPressMonitoring() and bring the monitoring reference count to below 0. The state of the AudioInputDeviceController instance will be |kEmpty| in this case but the DoClose() code only checks for kClosed.
The class has 2 states that aren't really necessary. kError is never used, so I'm just removing it. kEmpty is only used from construction until successful initialization. It is this state (kEmpty) that is never checked for and can cause us to call DisableKeyPressMonitoring inside DoClose without having first called EnableKeyPressMonitoring.
Since I'm cleaning up the states, I'm also fixing them as far as Chromium style goes.
BUG=347129
TEST=I ran into this while opening apprtc.appspot.com with a debug build of Chrome on Windows where my 'default communication device' was not the same as the 'default device'. These two different devices furthermore did not have the same sample rates, which caused opening the communication device with the incorrect parameters to fail. That's when I hit a DCHECK inside DisableKeyPressMonitoring where key_press_counter_references_ was being set to -1.
Review URL: https://codereview.chromium.org/181553003
git-svn-id: svn://svn.chromium.org/chrome/trunk/src@253762 0039d316-1c4b-4281-b951-d872f2087c98
| media/audio/audio_input_controller.cc | diffblobblamehistory | |
| media/audio/audio_input_controller.h | diffblobblamehistory |