GitLab

Bug: 19607784
Change-Id: I55accc49eec725ede0da3d1bfab448320fece926

The C++ class names don't match what the classes do, so rename
ISurfaceTexture to IGraphicBufferProducer, and SurfaceTexture to
GLConsumer.

Bug 7736700

Change-Id: I54f19904e325518a3af8fe23c018d3c0bb6bd807

Allow AudioPlayer to use the platform's 500-2000 range (in
  permille) of playback rate for audio when the data source
  is a URI or FD.
Since play events for those players are simulated through
  monitoring the playback position, scale the time
  between updates so play events are delivered in content
  time, not presentation time (as defined in OpenSL ES
  specification, "positions do not scale with changes in
  playback rate").

Bug 5898620

Change-Id: Ib512caebd7f3983c9e845be62e48dfee8499e59c

Change-Id: Ia98b76b2e0c49ee257e1f31c61c2d6f72a9fe894

See https://android-git.corp.google.com/g/157065

Bug: 5449033
Change-Id: I270cbf7b920d5f2fbb5f345f3c88426697c4391d

Add a media player death notifier to recover from media server crashes.
Downgrade some logs.
Destructors should be protected.

Change-Id: I7774733bcb12fc7fbfe237bdbeffca1adde46e78

Change-Id: I1592d1979ce5f5811ff5a47b4b60db93e8670313

Change-Id: I95aa181f39b375087b30ac65fea37589108c9f16

Change-Id: Ic5c88f8e1c8e7d6226a7a71a2a89a70f0fa876f8

Change-Id: Id23cbc43b56c014f5931cd1e941c12a773dba2b5

mDuration is protected by mSettingsLock because it is accessed from both
the ALooper thread and from the application thread, but only one of the
two "set"s was using the lock, and the "get" was not using the lock.

Also added some comments about the lock, and moved lock closer inside { }.

Change-Id: I7c96186f31baaad1b941d934549cb50d4f82d0c8

Do range-checking of MEDIA_BUFFERING_UPDATE parameters.

The other new logs are LOGV so disabled by default.

Don't forward empty MEDIA_SET_VIDEO_SIZE notifications, as
mediaserver seems to send these even for audio-only tracks.

Change-Id: I49063f83cec7bdfedb768b32f94735983755ebf2

Remove unsafe GenericPlayer::getSampleRate().  getSampleRate() was not
thread-safe as it accessed a member variable updated in the looper thread
without a lock.  If getSampleRate() is ever needed again, the sample rate
is available in the mPcmFormatValues array, which does have a lock.

Replace member variables mChannelMask, mChannelCount, and mSampleRateHz
by local variables for locality and performance.

Change-Id: Ibf92ed090430153d748ded4d0b835e429df47eea

Fix bug where StreamSourceAppProxy destructor wasn't reached,
which caused all sorts of other problems later on.  To see this,
enable the logs StreamSourceAppProxy::~StreamSourceAppProxy and
StreamPlayer::~StreamPlayer.  You'll see that StreamPlayer was destroyed,
but not StreamSourceAppProxy.

As StreamSourceAppProxy is child of StreamPlayer, make the reference
from StreamSourceAppProxy to StreamPlayer a weak reference in case
StreamSourceAppProxy's lifetime exceeds StreamPlayer.  It is not supposed
to any more with this fix, but the wp<> provides extra safety.

StreamPlayer preDestroy no longer bypasses the preDestroy in
GenericMediaPlayer.

Do a full disconnect in GenericMediaPlayer::preDestroy.

Push decremented reference counts for strong pointer through binder
to workaround binder's "optimization".

Extra error-checking in setListener and setBuffers to verify
that mediaserver is calling them correctly.

Use mutex mLock consistently in StreamSourceAppProxy.

Add an explicit StreamSourceAppProxy::disconnect to break
a circular reference, and call it in StreamPlayer destructor.

Make methods private: receivedCmd_l and receivedBuffer_l.

Add explicit clear during preDestroy to give up references earlier.

Warning: setDataSource(NULL) is not supported by NuPlayer yet,
this depends on another change in frameworks/base, so it is
commented out for now.

Rename mPlayerPrepared to mPreparedPlayer to avoid confusion with the
enum mPlayerPrepared.

Change-Id: Ie5f554c206027d22204eb86edd15489c6281b512

If a MEDIA_ERROR event is received during preparation, it indicates
that prepare has completed unsuccessfully.  However, MEDIA_ERROR can
also legitimately be sent after successful prepare. In that case, don't
update the prepare state.

Similarly, in the unlikely case that MEDIA_PREPARED is received after
prepare has already completed successfully or unsuccessfully, ignore it.
This probably indicates a buggy mediaserver, but we wouldn't want that
to take down the client also.

Change-Id: I2a92e98d13b70d1aa5ff6f440ad528f6896a2ead

Change-Id: Ic68d4d7b942ef99120ad30ee3ef7ce48ff1d697a

Workaround for mediaserver bug where it does not send MEDIA_SEEK_COMPLETE
to the MediaPlayerNotifier after a discontinuity is processed.

Instead, we simulate a seek complete event in this case.
A short delay is applied to compensate for shared memory latency.

A longer-term fix will be to have mediaserver send the event.

Change-Id: Id0d01b842a145fa5762d44f03144741152f7b03f

Fixed race due to missing mutex around 2 updates to mPlayerPrepared.

MediaPlayerNotificationClient had a raw reference to GenericMediaPlayer;
if for some reason the binder thread out-lived the GenericMediaPlayer,
then it might reference a destroyed object.  Made it a weak reference,
and issue warning if the GenericMediaPlayer has been destroyed.
This has similar effect as a callback protector.

Minor related:
 - mHasVideo is read without mutex, safe since never changed but make it const
 - GenericMediaPlayer's reference to MediaPlayerNotificationClient is now const also

Change-Id: I67b50e861a055c73fd490fb537a3d6d23d25c3d2

The StreamSource callback thread was continuing to run after the
MediaPlayer object was destroyed.  Fixed by adding a callback protector
and a pre-destroy hook. GenericMediaPlayer::preDestroy now also
calls MediaPlayer::stop just in case.

Change-Id: I5bd771d4d1936f433d2a8c9959593782c96daed9
Miscellaneous:
 - added an explicit destructor on CMediaPlayer::mAVPlayer for consistency
   (probably not a bug since there was a clear on it earlier)
 - updated comments for CallbackProtector
 - made some CallbackProtector fields private since no sub-classes yet

Also fix a bug in decode from fd to PCM buffer queue, where
we were closing the application's fd.

Change-Id: I220cd1c242f2f5e240b4dd1b9e8d79c88878bdcd

Use the utility method IMediaDeathNotifier::getMediaPlayerService()
to get a reference to media player service.  It retries after errors,
and keeps a single global per-process reference to the service rather
than one per object.  If we still can't contact the media player service
after retries, handle it like an unsuccessful prepare.

Miscellaneous:
 - Remove unused field Parcel metadatafilter

Change-Id: I451a64bd76b4a7f157774abe967448ec081f1014

Fixing bug 5126938 will involve configuring the video sink in two places:
at the initial data locator when creating the MediaPlayer object, and
then later in the IAndroidConfiguration interface.  This CL is preparation
for that.

Details:
 - Abstract out common code into android_Player_setNativeWindow.
 - Move the code for checking and initializing the video sink from an
   #ifdef ANDROID in the CMediaPlayer_Realize portable code to into
   platform-specific android_Player_realize.
 - Only set one of mSurface or mSurfaceTexture to non-NULL.
 - For methods which can't fail, replace the SLresult return value by void.

Change-Id: Iec8fd91e1a6de45eb64241b35e3f7d7faf63d727

As the media framework does not directly support marker and periodic
callbacks, we use a retriggerable one-shot timer at application level
to emulate this.  This has the advantage of being faster, but is also
less accurate than if it were in the framework.

Details:
 - Get position is now synchronous and (mostly) lock-free
 - Fix regression in SL_PLAYEVENT_HEADATMARKER and SL_PLAYEVENTHEADATNEWPOS
 - Fix SMP races in GetPlayState, GetCallbackEventsMask, GetMarkerPosition,
   GetPositionUpdatePeriod
 - AudioSfDecoder was not updating mPositionMsec when getPositionUsec failed
 - Updating attributes is relatively expensive, so only do it when a significant change
    in SetMarkerPosition, ClearMarkerPosition, SetPositionUpdatePeriod
 - Error on unknown message type in onMessageReceived
 - Delete an obsolete FIXME
 - "Clearing" a marker is now equivalent to disabling the marker event,
   and the default marker position is cleared.
 - use Android time units and types (int32_t, ANDROID_UNKNOWN_TIME, etc.) where appropriate
   instead of SLmillisecond and SL_TIME_UNKNOWN

Change-Id: Ib78bafa20d883b8d927364769663837389b6ea1d

Change-Id: Ia4d0433b4a6589bfda2f88addbe5b7eb3d0d9789

Details:
 - re-factor volume-related code yet again
 - remove dead variables mAmplFromVolLevel, mAmplFromStereoPos, mDirectLevel
 - add placeholders for kEventPrefetchFillLevelUpdate and kEventPrefetchStatusChange

Known issues:
 - MPEG-2 TS doesn't yet implement the ability to query channel count, so default to stereo

Change-Id: Ic0f2297b267dc3d380755e8d314e1d5f0f659d7c

Change-Id: If4e233f9007d7e48225a342a43f06df7f66cdea7

Change-Id: I754085b4d1e43c4d10bea36e0fb683d8113844e9

Change-Id: I1624c744168233c4ef79f2cd3090447b8aa7d53b

Aux effects on MediaPlayer audio players (non-AudioTrack) were broken

Change-Id: I5b97bee08d809b12cb9815d6d277934123780581

Change-Id: I9f1541b37ad5cda21dc49388e42c7fb08b22fc5b

Change-Id: I0cccee0d59fe3eefc6ef94b22bb6100d96bccdff

Change-Id: I9ae27ed4e977568a9687b9dd18eb998651569b24

Change-Id: I71c5e016147002fa84a56fc703da52e7b5c8417a

The play interface was only in place for routing calls to an
 AudioPlayer in OpenSL ES, not in OpenMAX AL for a MediaPlayer.

Change-Id: Icf514be428396264252874faf1060c76ddee9545

AudioPlayer (OpenSL ES) and MediaPlayer (OpenMAX AL) that play
 media from a URI or FD are implemented on top of an Android
 MediaPlayer. This CL implements requesting the current position
 from the underlying MediaPlayer object inside the event loop.
 At this stage, GetPosition cannot be called from inside a
 callback as the callback is made on the same thread that will
 process the asynchronous part of the GetPosition implementation.
Add test app for GetPosition.

Change-Id: Ife768bc40e37fc60cc19f6ba84bbe6cb06cc6590

through listener during video playback.
For feature request 800939.

Change-Id: I14aa18d2f803163a62c05f9e6bd990518c58cd3f

When an AudioPlayer is created, its playstate is set to STOPPED
 three times. A stop command is translated into a pause + seek(0).
 When a seek is underway, any new seek command was postponed. So
 when playback would start after a first seek(0), the other seeks
 would be executed, causing the stutterring.
This CL verifies that, when seeking to a certain time, any new
 seek command to the same point in time received while seeking
 is ignored.
It also correct the update of mStateFlags, which should be done
 BEFORE the command is issued to the player, in case the seek
 complete notification happens syncronously with the seek command.

This CL does NOT fix why a newly created player sees its playstate
 updated to STOPPED three times in a test app like slesTest_playUri
 before playback starts.

Change-Id: Ic7627678735f559b911ea3c5a0e1e00d7d7534d2

ISeek::SetLoop was returning undefined status.
Khronos Bug 7614 - sizeof(long) != 32 bits on x86_64.
Fix bug in IID hashing due to the above.
Fix bug in GenericMediaPlayer::onLoop; was using mPlayer without checking for NULL.
Fix wrong API level in tests/sandbox/object.c.
Decrease log level from I to V in MediaPlayerNotificationClient::notify
    and GenericPlayer::~GenericPlayer.
Fix memory leak in tests/sandbox/engine.c.
Comment out overly verbose log in tests/sandbox/streamSource/slesTestPlayStream.cpp.

Change-Id: I322fa9e19ee95ea1e18870a20f6a3c03b8f21208

When an object like GenericPlayer is stored in CAudioPlayer or
 CMediaPlayer by a sp, don't use a pointer to GenericPlayer when
 the object is passed as a function parameter, but rather use
 a reference to the sp that maintains it.

Change-Id: I49ff219618ac4b657deb43cfffb65f0e0dd6d58d

The SfPlayer class was implementing URI and FD decode and playback
 in the application process. This CL removes this class entirely
 and moves its functionality under the GenericPlayer class. This
 means that the playback of URIs and FDs is now implemented through
 an android::MediaPlayer object running in the media service.

The SfPlayer header had many StageFright and system includes, which
 are now moved to the relevant files where they are required,
 instead of automagically coming from sles_allinclusive.h.

Note that this CL breaks support for the following OpenSL ES features
due to missing Android MediaPlayer features:
 SLPlayItf.GetPosition
           HEADATMARKER
           HEADATNEWPOSITION
 SLVolumeItf.EnableStereoPosition
           SetStereoPosition
 SLMuteSoloItf
 SLPlaybackRateItf

 Support for those features will be reintegrated in subsequent CLs.

Change-Id: I8d7f6ea006eb7b876ef1ca1909d74b4517335850