Create a MediaCodecVideoTrackRenderer.shouldDropFrame function that can be overriden by a child class. The YouTube override prevents a frame drop if we haven't rendered anything in the last 35 ms.
The YouTube override is off at the moment, I plan to use a server side flag to do a slow and controlled experiment.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=142190774
Cue has been modified to optionally accept a fill color and a toggle specifying when to use the fill color. When the fill color toggle is set, then SubtitlePainter will use the fill color value instead of the color specified by the device's Accessibility settings. Cea708Decoder has also been modified to propagate that value, as well as cleaned up (in terms of documentation) to prepare it for inclusion in the open-source project.
There is also a new Cea708Cue, extending Cue, which holds the Cue's priority, which is used to sort potentially overlapping cues/windows.
Note that I've left the @ClosedSource annotation and logging in this CL. I intend to start testing the 608 and 708 functionality in the Fiber app to ensure that it works as expected on a wide-range of channels (as opposed to the single channel in ExoPlayer Demo) before removing these.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=142173264
It's a nicer place for it to live once it starts being passed
around more widely (e.g. through the video renderer, for
tunneling)
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=142158460
Although the underlying platform AudioTrack is capable of
writing the AV sync header from M onward, I've opted not
to use the functionality since it appears to allocate an
unnecessary (and large) number of ByteBuffers. We can swap
over from O if this is addressed.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=142138988
Also clarify when getNextLoadPositionUs and continueLoading
can be called.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=142124497
These currently lead to cryptic ArrayIndexOutOfBoundsExceptions being thrown from System.arraycopy() so my proposal is to throw a more useful ParserException instead.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=142087132
ClippingMediaSource wraps a single period/window video-on-demand source and
exposes a specified time range within it.
Issue: #1988
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141991215
This is particularly problematic for subtitle tracks where adjustment
can be broken. Now, the primary url can change when clients ask for a
variant snapshot, instead of happening on chunk load as before.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141933689
Define what element is returned if inclusive is true and there are multiple
elements equal to the input value.
Clarify/fix javadoc.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141891741
Not sure what I think about this, but we're getting quite
a lot of issues reported where streams play fine but capabilities
indicate they wont. It's probably best just to cross our fingers
and hope for the best in such cases, as was the case in V1 when
using ExtractorSampleSource.
Issue: #2157
Issue: #2034
Issue: #2007
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141758070
This requires knowing the seek time in Extractor.seek, so that it's possible to
pick the latest synchronization sample at/before the seek time for each track
(rather than the earliest synchronization sample after the seek position).
Also remove the STATE_AFTER_SEEK state which should no longer be needed.
Issue: #2167
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141432598
When the primary url is blacklisted (due to a 404, for example) or
the selected variant is different from primary url, allow the tracker
to change the url.
Issue:#87
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141291435
Pending improvement:
* Peek just the required priv frame. Avoid decoding all id3 information.
* Sniff the used container format instead of using the extension.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141181781
This is a first step toward supporting tunneled playback
Issue: #1688
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141167501
- Fix bug where we'd try and call replaceStream having already
notified the renderer that the current stream is final. This
could occur if a period was added to the end of the playlist.
- If the current period being played is removed and a new period
to play cannot be resolved, assume we've gone off the end of
the playlist and transition to the ended state. This allows
the current source to be re-used (unlike the previous behavior
of considering it an error). Treat valid seeks that cannot be
resolved due to concurrent timeline update similarly.
- Add seek sanity check back to ExoPlayerImpl. Meh. It's probably
best to keep this, since it stops the exposed window index
being invalid w.r.t the exposed timeline.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141167151
- Infer likely left/center/right alignment for pop-on captions.
This makes the rendering much better in practice, particularly
when the captions were intended to be center aligned.
- Fix line anchoring.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141156222
CEA-608 specifies a "safe-area" in which the captions should be rendered. This change adjusts all of the cue positions so they are within that safe area. It also fixes a minor issue in which roll up captions would drift towards the last row if a preamble address code had the "next row toggle" bit set.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141077433
Seeks are permitted when there's no timeline, and in this
case playback necessarily fails on the playback thread when
the timeline becomes known. Pushing failure to the same
place when the timeline is known means there's only one
failure path for this case. A later CL will likely introduce
an InvalidSeekPositionException for this case.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=141037708
This is implemented in the same way as in MediaCodecRenderer.
Move codec != null guard into feedInputBuffer to handle the case of an input
format change with no dequeued buffers where the codec can't be reinitialized
immediately.
Issue: #2111
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140864942
These additions are useful for sources that need to track the playback position
and control playback.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140828310
Prevented readFile() setting "changed" boolean to true every time. It's set only if encryption key is set but the index file isn't encrypted.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140714173
If a period has no enabled tracks it is considered to be fully buffered, even
if its duration is unknown. This would cause the offset of the next loading
period to be set based on the unset duration of the preceding period, in turn
causing the from of the player to expose a position based on an unset value.
Only load the next period when the current one has a known duration. If a
period has no enabled tracks and an unknown duration this causes the player to
play the period indefinitely.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140613858
When CEA-608 is embedded in MP4 each packet consists of
cc_data_1 and cc_data_2 only. The marker_bits, cc_valid
and cc_type are implicit. As a result playback of CEA-608
embedded in MP4 broke when we started passing the extra
byte for the TS case (and adjusted the decoder to assume
the byte was present).
This change introduces a special mimeType for the case
where the byte is implicit (!). An alternative option
was to insert the extra byte every 2 bytes in the MP4
extractor, but this is really quite fiddly to get right.
Also made the loops in the 608/708 decoders robust against
input of the wrong length.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140609304
Encryption key in index file is read by DataInputStream.read() which may return less bytes than required. Replaced it with readFully() which should read full length of data.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140597693
Leaving disabled by default for now, but we may want to consider
enabling it by default. Note that in V1 for ExtractorSampleSource
the behaviour was equivalent to this option being enabled, since
we didn't perform capabilities checks. For DASH/SS/HLS the V1
behaviour was equivalent to this option being disabled.
The type in PlayerActivity has been changed just to make it
easier to add a line that calls setParameters.
Issue: #2034
Issue: #2007
Issue: #2078
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140477568
I can't see how this would ever make a difference, but there's no
point in returning true. Either we've really reached EOS (in which
case outputStreamEnded will be true and the next drainOutputBuffer
will be turned into a no-op) or we've re-initialized the codec (in
which case there wont be anything to drain since we wont have fed
anything to the codec yet).
This change should also prevent the hypothetical NPE described in
issue #2096, although we're unsure how that NPE would occur unless
MediaCodecRenderer has been extended in an unusual way.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140338581
This allows ID3 PRIV timestamp extraction and Extractor Sniffing.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140209568
Select the timestamp master depending on track availability. If
a variant is being loaded, then that is the timestmap master.
Otherwise, if an audio track is being loaded, then the responsible
chunk source is the timestmap master. If no variant or audio
rendition is enabled, then a subtitle chunk source is selected as
timestamp master. This CL will become specially relevant once
ID3 PRIV timestamps are used for audio renditions.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140201385
Note that we still have null timelines in ExoPlayerImplInternal.
This is deliberate; and is likely necessary to distinguish between
the no-timeline-yet and timeline-is-empty cases (we want to try
and process a pending seek for the latter, but not the former).
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140200980
The refresh handler in HlsPlaylistTracker was being instantiated in the
same thread as the MediaSource (i.e. Main thread).
Issue:#2108
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140197553
- Allow extension and overriding of renderer creation.
Several developers have asked for this, so that they
can use their own renderers (typically extensions to
the core ones) without losing the ability to use
SimpleExoPlayer.
- Add option to not attempt extension renderer creation,
for efficiency.
- Align build variants for internal and external demo
apps. This is slightly unfortunate, but convergence
seems necessary for useExtensionRenderers.
- Fix DASH playback tests to use the debug video
renderer.
Issue #2102
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140140915
This issue affects ExtractorMediaSource only. We shouldn't
start loading in the case that we're prepared and have no
enabled tracks, since there's nothing that we need to load.
This was causing an assertion failure in startLoading.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140030650
This behavior is defined in ISO-13818-1, section 2.4.3.3(continuity_counter).
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140029161
- Fix handling of the currently playing period being
removed. This didn't do the right thing at all.
- Relax rule on seekToPeriodPosition re-using an
existing holder.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140014791
Note that multi-segment fetching is only possible in the
case that segments in a representation are defined to have
the same Uri and adjacent ranges (this is very rarely true
for live streams, but is quite often true for on-demand).
In the case that merging is requested but not possible,
the implementation will request one at a time.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=140012443
Also allow custom DashManifestParser injection, to
support parsing of custom elements and attributes that
service providers may wish to include in their manifests
(e.g. #2058).
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=139813108
- This should be a no-op change
- Inline attemptRestart
- Clean up processing of pending seeks
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=139804630
This is techically not allowed by the spec[1] but might still occur in
certain scenarios. New playlists with older media sequence numbers are
ignored.
[1]: HLS draft version 20, section-6.2.1
Issue:#2059
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=139564889
continueLoading shouldn't return true unless it's done
something. Always returning true if endOfStream was
causing CompositeSequenceableLoader.continueLoading to
loop forever.
It looks like the same issue exists in ChunkSampleStream
as well, although I can't seem to provoke DASH or SS
playbacks into doing anything bad as a result.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=139559834
ReusableBufferedOutputStream is a subclass of BufferedOutputStream with a reset(OutputStream) that allows an instance to be re-used with another underlying output stream.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=139505999
I'll have a more thorough refactor of some of this class
fairly soon!
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=139451693
Use BufferedOutputStream and BufferedInputStream while writing / reading.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=139446039
This removes duplication from SimpleDecoderAudioRenderer and
MediaCodecAudioRenderer.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=139187254
Clean up AtomicFile and make it return a custom FileOutputStream
for writing which handles IOException automatically during write
operations.
It also syncs the file descriptor and deletes the backup file on
close() call. This fixes the order of flush and close operations
when the fileoutputstream is wrapped by another OutputStream.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138779187
This can be used by the app for showing arbitrary UI on top of the player (for
example, UI elements associated with an ad).
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138610733
These variables are never read, since the underlying control
view reads them directly from the attrs.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138528246
This fixes VOD->Live transitions, with the caveat that
the Live portion still ends up further behind the live
edge than intended. This will be fixed in a following
change.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138407066
This change fixes the race condition where the internal
timeline is different to the externally visible timeline
when a seek is performed. We now resolve the seek position
using the externally visible timeline, then adjust the
period index as required for the internal timeline. If the
period is missing we follow similar logic for the existing
case where the playing period is removed.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138402076
This CL adds a SpliceInfoDecoder for the most common SCTE35 commands for splcing.
So far, it only includes TransportStreams, but porting it to HLS and DASH should be
fairly easy.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138389807
This is the first step towards allowing discontinuities in the
playlist tracking. Also changed durationSecs for durationUs in
MediaPlaylist.Segment.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138207732
handlePeriodPrepared
->setPlayingPeriodHolder
->enableRenderers
passes rendererPositionUs to renderer.enable(), but
this value is not set correctly.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138176114
A VOD-style period in a dynamic manifest would result in a NullPointerException.
Also fix another issue in which an unrecognized mime type would also result in
NullPointerException.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138075137
In order to expose the live window, it is necessary (unlike before) to refresh
the live playlists being played periodically so as to know where the user can
seek to. For this, the HlsPlaylistTracker is added, which is basically a map
from HlsUrl's to playlist. One of the playlists involved in the playback will
be chosen to define the live window. The playlist tracker it periodically.
The rest of the playilst will be loaded lazily. N.B: This means that for VOD,
playlists are not refreshed at all. There are three important features missing
in this CL(that will be added in later CLs):
* Blacklisting HlsUrls that point to resources that return 4xx response codes.
As per [Internal: b/18948961].
* Allow loaded chunks to feed timestamps back to the tracker, to fix any
drifting in live playlists.
* Dinamically choose the HlsUrl that points to the playlist that defines the
live window.
Other features:
--------------
The tracker can also be used for keeping track of discontinuities. In the case
of single variant playlists, this is particularly useful. Might also work if
there is a that the live playlists are aligned (but this is more like working
around the issue, than actually solving it). For this, see [Internal: b/32166568]
and [Internal: b/28985320].
Issue:#87
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=138054302
- Use A/V tracks only for buffering position when available
in ExtractorMediaPeriod.
- Fix layering of exo_simple_player_view so that subtitles
are above album art. The test stream provided has both.
- Make album art in SimpleExoPlayer view respect the resize
mode, like we do for video.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=137698473
TrackSelector no longer has a listener. Instead, tracks
change events are reported through ExoPlayer.EventListener.
Applications interested in retrieving the selection info
should retrieve it directly from the TrackSelector by
calling an exposed getter.
Pretty sure the ref'd issue is fixed as a side effect of
this change.
Issue: #1942
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=137183073
Unlike with PesReaders, sections don't have a standard way
of providing timestmaps or even generating tracks. We need
to pass this information so that readers decide what to do
with it.
Issue:#726
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=137162494
This allows the user to create section readers(usually) for reserved pids,
like SDT, EIT, CAT, etc.
Issue:#726
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=137150853
