Since we cannot tell how long it'll take until surface flinger
has displayed the data onscreen, we'll just have to guess...
We must not return the buffer to the decoder before it's been displayed.
This change is a complement to the main fix in kernel driver for the same issue (partner change #1250).
It removes clicks sometimes heard after the end of the tones while audio flinger is sending 0s to the audio output stream.
The problem was that the sleep time between two writes was more than the duration of one audio output stream buffer which could cause some underrun.
Also fixed a recent regression in ToneGenerator that made that the end of previous tone was repeated at the beginning of current one under certain timing circumstances when the maximum tone duration was specified.
This currently assumes 44k stereo (won't crash on other formats, but won't give the correct results either), and links statically with libspeex to get FFT data, increasing the size of libmedia by about 45kb.
The problem comes from the fact that the AudioSystem callback indicating that the media server is active again is ignored if it is received before the delayed message indicating media server death. This happens if another application or service running in the system server process makes a request to the AudioSystem in the interval between the death of the media server and the reception of the corresponding delayed message.
The fix consists in resetting mMediaServerOk flags immediately when the death callback is received and not when the delayed message is received.
* changes:
Fix some metadata retriever issues: 1. mRetriever was initialized twice: one in the constructor and the other in the setDataSource 2. cached the mode so that setMode can still be called before setDataSource. mRetriever won't be initialized until data source is set; thus we have to do this
1. mRetriever was initialized twice: one in the constructor and the other in the setDataSource
2. cached the mode so that setMode can still be called before setDataSource.
mRetriever won't be initialized until data source is set; thus we have to do this
When the AudioTrack callback notification size is relatively high (Which is the case on Sholes and over A2DP), it is likely that the end of tone is reached during the first callback. In this case, the AudioTrack is stopped before exiting the callback which causes 2 problems:
- 1: If the AudioFlinger thread is scheduled before we exit the ToneGenerator callback, the track can be stopped and reset before the data is actually marked as present in the buffer by the AudioTrack callback => no audio will be processed by AudioFlinger.
- 2: In this case, the data write index in the AudioTrack buffer is incremented after the track was reset by the AudioFlinger which leaves unplayed data in the buffer. This data will be played the next time the AudioTrack is started if not flushed in between.
The fix consists in adding an intermediate state to ToneGenerator state machine so that we exit the callback function when the stop condition is reached and stop the AudioTrack the next time we execute the callback.
* changes:
Suppress the testPing as we moved the release test target to the userdebug. This test rely on a mock_media_player.so which build under the eng build
* changes:
1) Add the wma / wmv profile reader. 2) Add the flag which check the wma/wmv enable properties. If the codec is not enable, then skip the wma / wmv related test cases. 3) Added a workaround which let the testcodecspecific test case continue even though there is a failure in the middle of the test. Besides, take out the remvoeFile for temporary usage.
