If a Webvtt HlsChunkSource got to schedule its chunk load before the
master HlsChunkSource (the one that downloads the TS or the fMP4
chunks), the player would never get past the buffering state.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132985792
- Also make some of the naming more concise + misc style cleanup.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132899979
- Also fix an issue that allowed blacklisting of all tracks,
due to incorrect index being used.
- Also fix an issue with track deselection for HLS.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132882151
This allows the adjustment of timestamps in microseconds along with
TS timestamps. This is useful for containers that include the
timestamps in microseconds format, like fMP4 and WebVTT.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132547521
- This change fixes seeking before the prepare (or more
accurately, before the timeline is set). The fix a minimal
one to fix the behavior. It's inefficient compared to
posting the pending seek onto the playback thread, which
will be the long term solution.
- As of this change, I think we can call V2 "done". There are
some loose ends to tie up, but the API is effectively
finalized and the implementation is in a state where you
can take it, use it and expect it to work.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132468107
Adds a few unused fields to HlsUrl and moves things towards the Hls
reimplementation we are looking for. Also fixes a bug related to
asuming every getNextChunk().loadable == null being related to
reaching the live edge.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132305206
If the live window has a small duration, we currently
end up setting the default start position to be right
at the start of the window. This increases the chance
of a BehindLiveWindowException.
With this change we impose a minimum 5s gap between
the start of the window and the default start position.
If the window is *really* small (<10s) then doing this
would push the default start position too close to the
end of the window. We don't have much time to play with
in either direction in this case, so we put the default
start position in the middle of the window and hope for
the best.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132054802
Now you can do cool things (if you really want to!) like
play a video twice, then play a second video, then loop
the whole thing, all seamlessly.
new LoopingMediaSource(
new LoopingMediaSource(firstVideoSource, 2),
secondVideoSource));
You can also just loop, which is probably more useful :).
Issue: #490
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132049599
People will inevitably try and do it, and it's pretty
easy to handle properly, so why not...
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=132047019
- Don't send a timeline to the listener until all children
have reported their timelines.
- Propagate a proper merge error if merging fails.
- The PlayerActivity hack is necessary due to the way Andorid's
MediaController widget attaches to the window :(. It'll go
away once we get our own player controls.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131958169
This could cause us to "lose" allocations backed by an
initial block, meaning they became unavailable for use
despite still being allocated.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131931465
- The need to pass a Clock is pretty much only for testing, so
make the constructor that takes one package private + use
the system clock for public constructors.
- Add default timeout values.
- Also make sure we set Content-Type in all license requests,
since when using Cronet the stack requires it to be set.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131404298
This is done through the bitstream id field and allows removing
the isEac3 parameter from the constructor.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131393477
If they're omitted, it's reasonable to assume it's because
they were uninteresting (i.e. sample data always tightly
packed at the start of the mdat). This is an issue for some
SmoothStreaming streams.
We actually already play such streams successfully, but
that's only due to another bug to be fixed in a following CL.
The same is true for V1, but given the low impact nature,
the fix will be V2 only.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131191975
If currentTrackBundle is updated at the start of the
block and then something goes wrong in the middle (e.g.
one of the skipFully calls) then the extractor wont
resume from the correct place.
This would be caught by our extractor tests if we had
a test sample that requires skipping to the sample data.
I'll try and construct one of those.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131191174
- The -1 needs to be a 0. My bad.
- Create AAC CSD if not defined in manifest, like in V1.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131190995
Also tweak how the null checks happen in a few DataSource
implementations (should be no-op changes, but allow you
to look at close() and be happy it does the right thing
without having to loop at the open() implementations).
Issue: #1759
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131172427
There's still some internal to clean up to do, and in particular
it remains a TODO to be able to handle seek calls before the
timeline is set (for this CL, such calls are dropped). This change
does however finalize the API.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131171318
Fix the calculation of the seek window for multi-period DASH.
Snap the default initial position back to the start of its segment, to ensure
that the first sample provided when transitioning to a DASH live source is a
key-frame.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131052912
- Remove playingPeriodEndPositionUs. It doesn't look like it's
required.
- Rename time variables to make it clearer what they are.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131027164
Window is potentially confusing with Android's Window class.
Once Window is renamed, it makes sense to rename Timeline too.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130938392
Also inline a few methods/classes where they can be made
private and therefore be removed from the public API.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130935090
Mainly, this allows the extractor to expose multiple audio tracks.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130928152
- Use a single constant for unset/unknown times across
all time bases. Note also that this moves away from
use of -1 for unset/unknown times in ms, which was a
bad choice (it might conflict with real times, such
as a time representing being just behind the start
of a live window).
- Add a few other unset constants, and use them.
- Fix some hardcoding of -1 where existing constants
should have been used.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130775594
- Remove getNewPlayingPeriodIndex from MediaSource
- Remove all absolute references to period indices from
Window. Their existence prevents ConcatenatingMediaSource
from being able to efficiently handle children with
large numbers of entries (e.g. a cursor backed source),
since it would copy all windows into its merged timeline.
- Fix ExoPlayerImplInternal to attempt a restart in the
case that the loading (but not yet prepared) period is
removed, in absence of a playing period.
- Implement logic for finding the "next" period in the old
timeline when attempting a retry.
Removing some of the nasty US<->MS conversions left as a
virtual TODO.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130659973
Period transitions can either be to new windows (in which case the default
position for the new window should be loaded) or to the next period of the
current window (in which case the the new period should be played from zero).
Fix the logic for calculating the new period index to load to implement this.
In processManifest, periodsById may contain periods that have been removed from
the manifest, which are still being used by the player (it releases periods on
receiving the source info refresh after processManifest returns). Ignore
periods that have been removed from the manifest when calling updateManifest.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130626441
- This change marks the methods that will be removed, and
renames the Window methods to remove "Window" from their
names.
- The Window methods need to be made to work (rather than
throw exceptions) when the timeline isn't set. Once that
gets done in a subsequent CL, the deprecated methods will
be removed.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130612979
- This change also enables seeking in live windows in the
ExoPlayer demo app.
- The added playlist doesn't transition properly by itself,
but for manual transitions it works correctly, and
demonstrates seeking into a default position.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130515880
- Add missing callback call.
- Allow injection of live edge offset.
- Refine calculation of live window size to correctly
handle just-started streams where the DVR window
hasn't yet grown to full size.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130412465
Also allow use of suggestedPresentationDelay taken from the
manifest, and enable this by default.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130409924
SinglePeriodTimelines can all use the same identifier, as their identifier is
wrapped in a pair with the (fixed) source index when they are concatenated with
other sources' timelines.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130386297
This is a 1:1 mapping. This change formalises the fact,
and makes it possible to easily query the mapping.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130375111
This was always a bit of a hack; for Play Movies. It may well
no longer be necessary, and if not I'd like to think of a nicer
or more general way of doing it. We can always bring it back
if needed.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130373433
- This avoids the need to have to use the timeline to
calculate a window duration, which can be awkward.
- Window now represents a window of availability with
an isSeekable flag, rather than a window of
seekability.
- Promoted Timeline and Window to top package; they're
pretty important :).
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130278509
It's no longer safe to convert END_OF_SOURCE -> duration on
the main therad, since the Timeline from which the duration
is obtained is posted to the main thread, where-as the
buffered position is passed by updating a volatile. Hence
an update to the latter might become visible to the main
thread before the corresponding Timeline.
This change moves the conversion to the playback thread.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130265893
This is in preparation for making it so that periods aren't reused.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130113382
- For audio/video, we should report the minimum largest
queued timestamp. This ensures that buffering continues
rather than stops in the case of bad sample interleaving.
- If there are "sparse" tracks then they should be ignored
for this calculation, otherwise the buffered position can
end up getting stuck at some small value (e.g. if there's
a text track containing a single sample with a small
timestamp).
- In the edge case that there are only "sparse" tracks, we
probably want to take the maximum largest queued time
instead.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130094266
*** Reason for rollback ***
Referential equality was probably the right thing to do, since using .equals
breaks track selection in the case that a source exposes two or more tracks
whose formats are equal. We should fix the way overrides work instead.
*** Original change description ***
Fix multi-period transitions with track selection overrides
An override applies across periods provided they expose the
same track groups according to .equals, but the formats in
the override are then compared against the period's formats
according to ==. Use .equals consistently to fix.
***
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=130083840
This change allows MediaPeriod instances to replace
SampleStream instances when the selection isn't changing.
It also allows MediaPeriod instances to retain a
SampleStream but indicate that the renderer consuming
from it needs to be reset.
The change is used to fix the ref'd bug, and is used to
do the same thing in HLS without the need for the source
to report a discontinuity. Note that reporting discontinuity
could cause unnecessary failure when used as a child of
MergingMediaSource.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129971782
If a renderer is reading ahead of the playing period, seeking within the
playing period would fail, because renderers would not be disabled but then try
to read from SampleStreams that have a released period.
Also, in the same circumstances, seeking within the reading period would fail,
because all renderers would be disabled, but their sample streams have already
been read. When they are reenabled they expect to see a format but don't
receive one.
In both cases, seeking can just clear the timeline. This only occurs in rare
circumstances when the player is reading ahead, so the cost of re-preparing
will not be incurred often, and allows the seeking logic to be simpler.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129891060
Referential equality is going to become important for detecting
whether a SampleStream has been replaced, so we need to create
new instances as we do elsewhere.
This also enables multiple SampleStreams to be provided for a
single TrackGroup, as is also true for DASH and SmoothStreaming.
It's forbidden to ask for multiple SampleStreams from a single
TrackGroup currently, but we may choose to relax that at some
point (and indicate whether it's allowed as a flag on each
TrackGroup).
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129842336
- Support specifying a preferred text language.
- Score based selection for text/audio/other tracks.
- Prefer default tracks to non-default ones.
- Allow overriding of base select*Track methods.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129626994
When seekToPeriodPosition found that the seek destination period was already
prepared, it would not disable/re-enable renderers. This was fine if the
playing period wasn't changing, but in other cases the renderers would be left
reading the incorrect streams (and the underlying periods may have been
released).
Also, transition to the buffering state before re-enabling renderers, so that
the renderers are not started until leaving the buffering state.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129625632
- Enforce viewport constraints for fixed video track selection.
- Select best fixed video track, not the first one.
- Better handling of video tracks with unknown dimensions.
- Mini bug fix.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129226593
- Only setup a codec to allow adaptation to other compatible
formats in the stream. If something like the mimeType is
changing adaptation will never be possible, so there's no
point.
- Incorporate maxInputSize into the reconfiguration logic.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129088464
On an edge case, player may get stuck when the renderers are ready but the buffer doesn't get full enough because of a fatal error in data source. An example state can be created by starting a live DASH stream and switching between normal and slow network connections.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129084824
An override applies across periods provided they expose the
same track groups according to .equals, but the formats in
the override are then compared against the period's formats
according to ==. Use .equals consistently to fix.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129081591
For self-initializing segments, DefaultTrackOutput receives
a null format and then the actual format. This broke the
deduplication logic in InfoQueue.format, since there were
two transitions: FormatX->Null and Null->FormatX. This CL
allows deduplication to succeed in this case.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129081583
The codec cannot be reconfigured if the format rotation
changes, or if the format dimensions exceed the current
decoder's configured maximum dimensions.
Issue #1707
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=129004187
When seeking to the default position in a period, the containing source may
actually return a position in another period. Multi-period live sources can do
this to seek the player to the live edge.
ExoPlayerImplInternal uses the same functionality when the playback position
reaches the end of a period to determine what period/position to play next.
This means that when playback transitions to a multi-period live source from
some other source (playing a concatenation of those two sources), the player
will play the live edge rather than the beginning of the earliest period.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=128984355
Useful functionality promoted to core library:
1. Management of SurfaceHolder.Callback lifecycle
promoted to SimpleExoPlayer
2. Ability to determine whether audio/video tracks
exist but are all unsupported promoted to
MappingTrackSelector.TrackInfo
3. Read external storage permissions check promoted
to Util
4. SubtitleView given ability to act directly as a
TextRenderer.Output to remove layer of indirection
5. SubtitleView given ability to configure itself to
user's platform wide caption styling
6. KeyCompatibleMediaController promoted to library's
UI package.
Relocation of boring stuff:
1. ID3 frame logging moved to EventLogger.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=128714230
This change merges the duties of FormatEvaluator into
TrackSelection classes, so that both the static and
dynamic parts of track selection are implemented in a
single place.
New feature: Demo app now allows you to enable random
adaptation in the track selection dialog.
Notes:
- It should be quite easy to allow application side
track blacklisting in addition to source side, as an
extension to this. That would effectively allow
applications to do seamless/deferred track selection
by creating a TrackSelection with all tracks enabled,
and then toggling the blacklist flags to select the
ones they want to be active.
- It should be trivial to implement format blacklisting
for DASH and SS as an extension to this. Will do in a
follow up CL.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=128707517
