We were previously using the container format of the media being
played as the mimeType generating key requests, but this is not
always correct. As an example, where a manifest contains webm streams
but specifies initialization data using cenc:pssh elements in the
manifest, the media has a webm mimeType, but the DRM initialization
data has an mp4 mimeType.
It appears the spec calculation gives the h:w pixel ratio, where-as
we want w:h. It's pretty easy to convince oneself that this way round
is correct. Consider a video that's 100px by 100px, and setting
aspectRatioCode=3 to achieve this. The pixelWidthHeightRatio needs to
be 16/9 and not 9/16 :).
Issue: #965
Previously, when spectral band replication (SBR) or parametric
stereo (PS) was in use in an MPEG-4 stream, the channel configuration
chosen was likely incorrect. The channel configuration was *always*
incorrect for 7.1 audio (gave 7 channels instead of 8).
- NAME is optionial now in the Hls Manifest
- Use the id field in Format to store the NAME instead of
a field in Variant to mimic DASH's behaviour
(see the DASH Id PR, which is not merged yet at this time).
Added the property 'id' to the MediaFormat class
which serves as an identifier for the track.
DASH Representations will have the "id's" from their
Media Presentation Description mapped to the id property
in the MediaFormat class that will represent the track.
We needed this for an use case where we wanted to read the 'id'
value from the DASH representation and present it to the user
in order for the user to select the right track.
The sample data position is the sum of the data offset and the base data offset.
The base data offset is either specified in the stream, or defaults to the first
byte position in the moof box. (We only support one traf per moof currently, so
the offset does not need to be assigned for later track fragments.) The data
position can optionally be offset by a data position read from the trun.
The auxiliary information offset is calculated in the same way, but using an
offset read from the saio box.
Issue: #837
Issue: #861
Context:
- Currently, playback is significantly more juddery with it disabled,
particularly on AndroidTV.
- We should be able to do the "best" job of this internally, so injection
doesn't buy anything useful. If someone has a better implementation for
adjusting the frame release, they should improve the core library.
The only Samsung devices with names starting "d2" that we're aware of
are Galaxy S3 variants, and also one Samsung Galaxy Pocket Neo d2aio
SAMSUNG-SGH-I747Z. This change speculatively includes that device too because
its name is very similar to SAMSUNG-SGH-I747 which is known to be affected.
Issue: #548
Try re-sync'ing to the next level 1 element when invalid data is found. This
corrects the behavior for test case 4 in the mkv test suite.
Partially Fixes Issue #631
This change keeps the proportion offset * 256 as a floating point value rather
than rounding it before linear interpolation, which will increase precision
slightly when seeking in streams with XING headers.
In practice, this won't make much of a difference because precise seeking in VBR
MP3s with XING headers seems not to be possible without reading the entire file,
due to the fact that the (uneven) distribution of bits is represented by a fixed
number of table of contents entries.
- The old approach was technically incorrect, because the checks
were "capacity < sampleSize" and hence neglected the fact that
the buffer position may be greater than 0 (e.g. if the caller
wants to prefix the sample with some additional data).
- Also proactively throw an exception if the buffer is too small,
rather than wait for the failure when we actually do the write.
- Line parameter
- Added support for value and line alignment attributes.
- Support negative numbers when line is an absolute number (not a
percentage).
- Position parameter
- Added support for value and position alignment attributes
- Added support for WebVTT comment blocks
- Percentage values now accept decimal numbers (as webvtt spec states)
- Added new WebVTT tests for testing all new implemented features
- do not denormalize styles at parsing time but only put normalized style info
into TtmlNode tree. Resolve styles on demand when Cues are requested for a
given timeUs.
- create TtmlRenderUtil to have static render functions separate
- added unit test for TtmlRenderUtil
- adjusted testing strategy for unit test to check resolved style on Spannables after rendering
When switching format in HLS, we instantiate a new extractor, which
adjusts TS presentation timestamps so that they align properly with
the start of the first segment in the new format. Some HLS streams
appear to have slightly misalignment that causes a glitch when using
this approach.
It's better to re-use the same timestamp adjustment across formats,
and only reset it when seeking or when there's an actual discontinuity.
This is because the HLS spec guarantees PTS timestamp alignment across
different formats.
We'll also need something like PtsTimestampAdjuster to share between
separated audio and WebVTT tracks, which also contain PTS timestamps
that are aligned, and will need to share a common adjustment.
Issue: #692
- Don't allow "nothing has changed" optimization in the case
that only styling has changed (TextUtils.equals will return
true in this case, but we shouldn't optimize).
- Add functionality to suppress embedded styling; seems useful
to have.
- Added "this." for clarity.
When a live stream ends, what typically happens is that the manifest
is refreshed and the refreshed version is not marked as live/dynamic.
When this happens we:
1. Don't want the duration of the track to change.
2. Still want to consider the possibility that we may have fallen behind
the live window.
3. Don't want to allow futher manifest refreshes.
This change uses the right thing in the right place.
- Admit we don't know the mime type (using unknown mime types) rather
than passing the container mime type.
- Pass the correct mime type for opus, vp9 and vp8, and remove the incorrect
container checks in the corresponding extensions.
- With this change, you can select from the individual video formats in
the demo app, as well as the regular "auto" (adaptive) track.
- DashRendererBuilder no longer needs to create MultiTrackChunkSource
instances for the multiple tracks to be exposed.
I think such manifests are invalid, and I haven't seen any examples,
but given it's trivial to fill in the duration if the periods define
durations, it seems worth being robust.
- It's not possible to determine a period's duration just from the
corresponding Period element in a DASH manifest. It's necessary
to look at the start time of the next period (or the duration of the
manifest for the last period) to determine how long a period is. We
don't currently do this in the parser, and hence set duration incorrectly.
We also set period start times incorrectly because we don't set it to
equal to sum of the durations of prior periods in the case where it's
not explicitly defined.
- We're currently propagating these (incorrect) values all over the place
through data-structures that we build when parsing the Period element.
- This CL removes this redundancy, storing only the start time of each
period in Period elements, and not propagating it elsewhere. It's then
used when required in DashChunkSource.
The following sequence was problematic:
1. See start of a cluster having not output a seek map. Decide
to seek for the cues. Enter CUES_STATE_BUILDING state.
2. Error occurs before seek map is output.
3. ExtractorSampleSource isn't prepared yet, so restarts from the
start of the stream.
4. See start of the same cluster having not output a seek map.
This time cuesState is CUES_STATE_BUILDING, so we just carry
on. We then fill the buffer with sample data, despite the
source not being prepared, at which point we get stuck.
It's unclear to me why cuesState needed three states, so I've rm'd
the BUILDING state. Step (4) above will now do the same thing as
in step (1). If the failure repeats, we'll eventually fail, which
is WAI.
These MP3s are unseekable but allow calculating the VBR duration correctly.
Treat streams as live only if they are unseekable and lack a duration.
Issue: #713
- Remove unused method in DashChunkSource.
- Remove inputEncoding parameter for subtitle parsers. We're
ignoring it in all but one of the parsers, and for the one
that does use it, it'll only ever receive null, since that's
all we're passing.
- Make TextTrackRenderer advance to the next subtitle even if
the current one hasn't finished, in the case that they overlap.
This shouldn't ever really happen, but it seems best to trust
the start time of the new sample rather than the last event
time of the previous one.
- Video track is always marked as adaptive, the resolution is
stripped out (since it's otherwise just set to whatever the
resolution of the first selected variant is), and the max
dimensions are set.
Issue #514
This is needed for several use cases:
- ExtractorSampleSource with option to play both embedded and out-of-band
subtitles.
- HLS multi-audio and out-of-band-webvtt.
- Migrate demo app to use new APIs.
- Add multi-track support for ExtractorSampleSource case.
- Add multi-track support for SmoothStreaming use case.
The final step is to add support back for the DASH use case and
delete MultiTrackChunkSource. This is blocked on multi-period support
landing, in order to prevent a horrendous merge conflict. We also
need to update HLS to expose sensible track information.
Issue: #514
When ChunkSource implementations implement multi-track for DASH and SS,
format selection will move inside of ChunkSource. If we, for example, fail
to query the decoder to determine which tracks are playable, we need an
opportunity to fail (i.e. say we're not prepared, so that maybeThrowError
is called, from which we can throw).
This may go away in the future if we remove the distinct preparation step
and treat tracks/formats as things that can change dynamically, but for now
this is what we have.
Issue #514.
Fix reading the first slice flag, which before could cause a read out of bounds
if the NAL unit started at the end of the buffer.
Handle non-VCL NAL units by flushing a pending sample when starting to read one.
multi-track support upstream to the ChunkSource interface.
This change does not yet make use of the newly exposed APIs. This
will come in a subsequent CL.
Issue #514.
- Currently all subtitles we parse contain timestamps relative to the sample
timestamp, however we add the sample timestamp in inconsistent ways (sometimes
in the Subtitle, sometimes in the SubtitleParser). This change converges on
a single approach. It also paves the way for passing absolute offsets to use
instead, and being able to apply them in a consistent way in a single place
(PlayableSubtitle). This functionality will be required for ISO 14496-30 TTML
embedded subtitles.
Issue: #689
- Parse all attributes that may exist in either the AdaptationSet or
in the child Representation elements at both levels.
- Correctly infer TYPE_TEXT for Representation elements whose mimeType
is application/mp4 and whose codecs attribute indicates a known text
codec type.
Issue: #689
Remove MPEG TS stream filtering based on AudioCapabilities.
Pass AudioCapabilities to MediaCodecAudioTrackRenderer so it can choose between
passthrough/raw and decoding for AC-3 tracks.
- Generalize rendererEnabledFlags to be selected track indices through
ExoPlayerImpl/ExoPlayerImplInternal.
- Selecting an out-of-bound track index (e.g. -1) is equivalent to
disabling a renderer prior to the generalization.
- A prepared TrackRenderer that exposes 0 tracks is equivalent to a
TrackRenderer in the STATE_IGNORE state prior to the generalization.
Issue #514.
On NVIDIA Shield, recreated passthrough AudioTracks have incorrect playback head
positions, due to shared state with the previously destroyed passthrough
AudioTrack. Calling AudioTrack.flush() before AudioTrack.release() ensures this
state is cleared.
This is a safe change so I have not made it a device-specific workaround. The
combined time for flush and release is less than 10 ms (with flush normally
taking less than two ms).
Everything I've seen that uses ExoPlayer sets it to true, and
setting it to false is poorly supported / likely to result in
bad initial A/V sync after each seek.
Both FragmentedMp4Extractor and WebmExtractor now invoke seekMap() with
SeekMap.UNSEEKABLE if there isn't an index in the stream, so DASH playbacks
will end up printing this warning every 5 seconds or so. This is obviously
not desirable, so this CL just removes the warning! The alternative would
have been to print a warning if the value is anything other than UNSEEKABLE,
but it doesn't really seem worth it.
This is useful to allow playback of individual segments from a
DASH stream as regular fmp4 files. These segments don't typically
contain a segment index. For playback to start, we need to invoke
seekMap with the UNSEEKABLE index. We do this if we haven't seen
a segment index when we encounter an mdat box (if one were present,
it would have been located earlier than this point).
1. [Cleanup] Remove unused Track types, including TYPE_TIME_CODE.
2. Add subtitle track type, which is different to the existing text type.
3. Set duration on the media formats for text and subtitle tracks. This
was causing the player to report unknown media duration for mp4 files
containing such tracks.
4. Make TextTrackRenderer do the right thing when not started.
Issue: #635
- Make UnrecognizedInputFormatException public so the app can more easily handle
it when it is propagated to onPlayerError.
- Format the description using the simple class names for the extractors.
- Don't retry loading after it's thrown, but just throw immediately.
- ExtractorSampleSource takes an array of extractors to test for suitability.
- Extractors now implement a sniff() method that returns whether they can
extract samples in the input stream's format.
- Switch demo app samples to use format detection.
Issue: #438
1. Fix seeking in test2.mkv by handling non-default timescale
after duration.
2. Fix handling of missing cues in test6.mkv by allowing playback
to continue (but all seeks will reset to t=0).
Issue #631
The ID_SEGMENT can only be read once, as seeing the element a second time is
assumed to indicate that the file contains multiple segment elements (which is
not supported).
This change allows the element to be read twice if it is at the same position,
so that retrying loading from the start can succeed.
1. Workaround for decoders that fail to handle the END_OF_STREAM flag.
2. Revert processing of final output buffer if it's non-empty. This
introduced another bug (#596)
Reverts: b88012f51f
Issue: #417
Issue: #596
SubtitleLayout no longer trigger re-layouts of the view hierarchy.
Instead, the SubtitleLayout just invalidates itself. This is made
possible by making SubtitleLayout a regular View that draws each Cue
directly onto the canvas, rather than having SubtitleLayout be a
ViewGroup with a child View for each Cue.
The SubtitleLayout is now properly aligned with the surface.
This means the subtitles remain on top of the video in portrait
mode, rather than being huge and below it.
Prior to this change, there was a bug where playback would
fail with the following steps:
1. Start playback.
2. Pause playback.
3. Disable all renderers.
4. Enable at least one renderer.
5. Resume playback.
The new logic assumes that an input format change will be
followed by an output format change, but I think this is
pretty much guaranteed. If this weren't to happen then the
new pixel aspect ratio wont be picked up, but I think it's
extremely unlikely (it would require the format to stay
exactly the same except for the pixel aspect ratio, which
would be bizarre).
1. An optional U+FEFF BYTE ORDER MARK (BOM) character.
2. The string "WEBVTT".
3. Optionally, either a U+0020 SPACE character or a U+0009 CHARACTER
TABULATION (tab) character followed by any number of characters that
are not U+000A LINE FEED (LF) or U+000D CARRIAGE RETURN (CR) characters.
4. Exactly one WebVTT line terminators to terminate the line with the file
magic and separate it from the rest of the body.
Issue: #580
Improve Mp4Extractor test.
Add support for Xiph lacing in Matroska files.
Add support for EBML lacing in Matroska files.
Handle the initial sticky intent for HDMI audio plug.
An accumulation of several fixes:
1. Change to HlsExtractorWrapper is just a move + documentating
things that were already true + adding a precondition in the
configureSpliceTo method.
2. Change in HlsSampleSource.readData ensures that configureSpliceTo
and hasSamples aren't called on an extractor that isn't prepared.
3. The other change in HlsSampleSource ensures the correct "previous"
TsChunk is used. If a TsChunk fails to load and is replaced, the
previous chunk should be the one before that whose load completed
successfully.
4. Determine switchingVariantSpliced based on the actual format of the
previous chunk, so it's set correctly in the case of a TsChunk load
failure and subsequent replacement.
This fixes the case where we need to switch from one variant to
another during preparation, having loaded zero samples, because
the first chunk for that variant gave us a 404 error. In this
case the first extractor will have no samples, and there will
be a second extractor that does.
There was a mess where we were indexing into both a list of variants
and a (differently ordered and possibly of differing length) list of
formats. This sanitises everything.
1. Refine the way TtmlNode handles whitespace collapsing when constructing
the output text. Can of worms.
2. Start using SpannableStringBuilder. This will allow attaching of spans
in getText as nodes are encountered, which is how we'll be incorporating
styling information into the result.
1. Remove requirement for TrackRenderer implementations to report
current position, unless they are time sources.
2. Expose whether renderers have media to play. The immediate benefit
of this is to solve the referenced GitHub issue, and also to only
display the appropriate Audio/Video/Text buttons in the demo app
for the media being played. This is also a natural step toward
multi-track support.
Github issue: #541
- Fix bug where duration of initially disabled tracks wasn't correctly
incorporated into the overall duration reported by the player.
- Don't transition to STATE_ENDED unless the position has reached the
duration, if the duration is known. This allows for "playback" to
continue when all renderers are disabled, rather than jumping straight
to STATE_ENDED.
- contentType can be defined on an AdaptationSet.
- language can be defined either in AdaptationSet or in a contained
ContentComponent.
- The id from the AdaptationSet should come from the AdaptationSet.
to if the playlist load fails.
NB - I'm aware the casting is a bit messy, but I don't want a common
interface because I'm hopeful that TsChunk will go away at some point.
Issue: #537
* fix channel number encoding using the AUDIO_SPECIFIC_CONFIG_CHANNEL_COUNT_TABLE
* allocate the correct number of bits (4) int the CodecSpecificDataUtil struct
This makes it so that it's no longer necessary to specify the number
of downstream renderers to HlsSampleSource, FrameworkSampleSource and
ExtractorSampleSource, by forcing the downstream renderers to register
with the SampleSource instances in their constructors. This eliminates
a common source of subtle client bugs where the passed value is incorrect.
ChunkSampleSource had a null check solely for the VTT case,
where DashChunkSource wasn't setting a MediaFormat on VTT
chunks. It's trivial to do so, and is more consistent, so I've
done that and removed the null check. Also done some small
tidying.
This OMX component is listed but can't be instantiated on this device.
According to the GitHub issue, some other devices are also affected, so these
will have to be added too.
Issue: #377
They're current location is annoying, because it creates a dependency
from the dash package to the webvtt package. For apps like Play Movies
where WEBVTT isn't used, it's nice just to delete the whole package at
import time, which requires that this dependency be removed.
- There's definitely potential for more code sharing in these
classes, but deferring for now.
- Also made no-scheme default to file://, and allowed smoothstreaming
URLs to be specified with or without the /Manifest suffix.
Before preparation, and when seeking, Mp3Extractor did not handle retrying:
- synchronizedHeader was set before the header was known to be valid, which
means that after seeing one valid frame header and then failing to read, the
synchronization would be treated as complete.
- The input buffer would keep data loaded during synchronization but on the next
call to synchronize when retrying it was not returned to the mark position to
re-parse the data.
This change fixes these issues.
- Make HlsPlaylistParser treat non-positive dimensions as unknown.
- Make HlsPlaylistParser parse floating point resolutions, because
technically that's how they're spec'd.
- Make VideoFormatSelectorUtil treat non-position dimensions as unknown.
Issue: #461
When I moved the id3Reader instantiation out of the while
loop (below where it is now) it was no longer guarded by
the streamTypes.get(streamType) deduping check. This brings
back an equivalent check.
When a XING header is present but not usable (due to missing fields), CBR
seeking can be used instead. It relies on the bitrate. The bitrate from the
unusable XING header is not correct, which leads to incorrect seeking.
Also fix VBRI seeking by setting the correct offset on the frame to parse. Few
people seem to use that format, but I have found two very short truncated
samples which were falling back to the CBR case before but are using VBRI with
this change.
When a passthrough AudioTrack is replaced (due to seeking) the new one behaves
as if it is still emptying data from the old one, with its playback position
advancing until it runs out of data.
Data written while the 'old' AudioTrack was emptying would be discarded, so
avoid writing to the new AudioTrack while the old one is still emptying.
Also avoid using AudioTrack.getTimestamp with passthrough tracks, as this causes
the playback position to jump to a position that breaks audio/video
synchronization.
This also fixes a technical mistake where HlsChunkSource is fed
seekPositionUs=-1 when obtaining the first chunk. This is wrong,
but the usage of this variable within HlsChunkSource enforces that
the seek must stay within bounds, so we get away with it.
Issue: #385
1. prepare() needs to load a TsChunk to actually prepare the source.
2. Source is prepared, but no tracks are enabled (this is why it's
an edge case - no-one is likely to be doing this!).
3. The TsChunk load completes.
We should not load additional chunks in this case.
For Live SmoothStreaming, referential equality checking
isn't enough (it breaks once the manifest is updated).
Updated other instances too just for consistency.
Play movies has an Allocator that attempts to allocate a single
huge byte[] up front to minimize the risk of GC pauses. This abstraction
will be required to keep that when updating them to the new Exo.
VP8 can be decoded by MediaCodec (since very early versions of android). Now that we want WebmExtractor to be general purpose, adding VP8 makes sense as it is a common use case.
This CL adds support for WebM files which have Cues present at the end of the file (i.e.) after Clusters. The file referenced in the bug can now be played back using the demo app. It adds a new flag to WebmExtractor which should be set to true only when being used through ExtractorSampleSource. All others (e.g. DASH) should not set it.
Reference file: http://demos.webmproject.org/exoplayer/glass_vp9_vorbis_cues_at_end.webm
I've found myself doing this a couple of times during local
debugging. It's harmless to have it public, and seems pretty
useful for debugging inside of the mp4 package.
Using a file:// URL for loading an HLS stream (for testing) would fail due to
casting the connection to an HttpURLConnection in DefaultHttpDataSource.
This change makes UriDataSource an interface for objects that are DataSources
with URIs. That allows for reading manifests for HLS using just a UriDataSource
rather than an HttpDataSource (URLs in the playlist are relative to the data
source's URL so the getUri method is needed).
On retrying loading a chunk, the state of the extractor was reset due to a call
to seek(). Prevent this call by only calling init() if no bytes were loaded.
Also make the DefaultExtractorInput use the loading position not the original
stream position so that its getPosition() method returns offsets relative to
the start of the stream, which fixes a bug where the chunk index offsets would
be relative to the wrong position if there was a retry while loading the chunk.
Vorbis decoder in android MediaCodec [1] expects the last 4 bytes of the sample to be the number of samples in the current page. This definition holds good only for Ogg and is irrelevant for WebM. So we always set this to -1 (the decoder will ignore this value if we set it to -1). The android platform media extractor [2] does the same.
This works around an issue the audio track continues to play audio data written
after calling AudioTrack.pause(), which breaks rebuffering behavior (as video
can never catch up if audio continues to be consumed, in some circumstances).
Also don't increment the timestamp wrap count if the track is paused, to work
around an issue where the playback head position jumps back to zero after the
AudioTrack position jumps back to zero.
The OMX component needs to be configured with a format that has a
MIME type of audio/raw. Remove Ac3PassthroughAudioTrackRenderer,
which is no longer used.
The limit on the ts packet buffer can be reduced during processing
if it's discovered to have padding. Hence we need to reset it back
to the ts packet size before processing each packet.
- Keys should not expire during normal playbacks of correctly configured content.
- Attempting to refresh on expiration causes a race condition, that may result
in either failure or brief re-buffer, for the 30s license test video.
- This change provides deterministic behavior, which is to always fail.
Some servers, probably edge cache machines that exclusively serve
chunked media, don't support partial requests. Which is kind of
vaguely reasonable for that particular case. This change modifies
DefaultHttpDataSource to correctly handle this case, by manually
skipping data up to the requested position (and making sure not to
read more data than the requested length).
- Have extractors read from an ExtractorInput. Benefits of this are
(a) The ability to do a "full" read or skip of a specified number
of bytes, (b) The ability to do multiple reads in your extractor's
read method. The ExtractorInput will throw an InterruptedException
if the read has been canceled.
- Provides the extractor with the ability to query the absolute
position of the data being read in the stream. This is needed for
things like parsing a segment index in fragmented mp4, where the
position of the end of the box in the stream is required because
the index offsets are all specified relative to that position.
- Move to builder naming.
- Propagate formats to the TrackOutput instances, rather than having
them be read through the Extractor. There was actually some weird
indexing going on here before (which happened to work, but wasn't
well defined).
As per http://tools.ietf.org/html/draft-pantos-http-live-streaming-04#section-5.2,
the initializaton vector (IV) of the AES decryption algorithm should be set to:
- the IV attribute value if present,
- the sequence number otherwise.
Currently, the IV is set once and use over all next media sequences
where the IV attribute is not set. The fix is to use the provided IV if
given or use the current media sequence number.
- It's probably easiest to think of this as a standalone HLS change, which is splitting out the "loading" and "consuming" sides of HlsExtractor and a good structural change in its own right. To do this, HlsExtractorWrapper becomes a final class implementing the consuming side. HlsExtractor becomes an interface defining the loading side.
- The bigger picture is that, hopefully, HlsExtractor will become a lightweight extractor interface that can be used throughout the library. Because it doesn't need to implement the consuming side, we'll save on having to re-implement the consuming side for every extractor (we'll probably need one consuming side implementation for HLS/DASH/SmoothStreaming, and a second one for everything else, both of which will use SampleQueue). It's expected that the HlsExtractor interface will need to change to accommodate all use cases.
- The next step in unification will be to try and have FragmentedMp4Extractor implement HlsExtractor (which will need renaming). Once this is done, I'll try and move the chunk package over to use the HlsExtractor interface.
- This is a step toward hopefully converging HLS and CHUNK packages.
- Add support for encrypted samples.
- Add support for appending from a DataSource.
- The fmp4 extractor was reading from sampleEncryptionData even
for not-encrypted samples, which I'm pretty sure isn't right.
Fixed this. For all encrypted content I've seen, this change
will be a no-op because isEncrypted is always true if there's
an encryptionBox present.
- Made webm extractor only set cryptoInfo if isEncrypted is true.
- Align variable naming in the two extractors, for clarity.
- Rather than returning a map, return a DrmInitData object,
with mapped and non-mapped implementations.
- Include a suitable mimeType to pass to the MediaDrm. Previously
we were incorrectly passing the mimeType of the samples,
where-as MediaDrm expects the container mimeType. Note that
it doesn't matter whether the mimeType starts with "video" or
"audio", hence using video mimeTypes everywhere.
It was possible for a codec input buffer to be filled with two frames' worth of
data, if seekTo was called after populating a buffer, if waitingForKeys was true
and seeking did not trigger a flush. This caused the CryptoInfo to be configured
as if the input buffer contained a large amount of reconfiguration data as
cleartext.
Move resetting waitingForKeys to flushCodec, so that we don't try to read the
next sample from the source until the first one has been consumed or discarded.
Also remove uriIsFullStream. It's not doing anything particularly
useful, so I think it makes sense to remove it from the public API;
it's unlikely anyone is using it.
Issue: #329
For fields encoded using UTF-16 or UTF-16BE charsets when looking for
termination character we have to look for two zero consecutive bytes.
Otherwise, as many characters encoded with UTF-16 or UTF-16BE has one
of their 2 bytes set with the value zero, we will be truncating text
fields.
- Added support for ID3 frames of types GEOB and PRIV.
- GEOB type is commonly used by dynamic ads provider to include in the
stream information about the ads to be played.
- PRIV type is commonly used for time synchronization (example:
synchronizing playback of a live stream and its webvtt captions) and
also by analytics companies to include tracking information in the
stream.
- Added a sample stream from Apple that contains ID3 metadata.
Note: I'm fairly confident that NetworkLoadable.Parser implementations
can live without the inputEncoding being specified. But not completely
100%...
Issue: #311
Issue: #56
The return value here assumed that the time being searched for
was beyond the start time of the last segment. This fix also
handles the case where the time is prior to the start of the
first segment.
Clear stale blacklist in getChunkOperation before getting next variant.
This ensures:
1.- Player resilience to failures, always trying to look for a working
playlist that allows player to non stop playback.
2.- High quality blacklisted playlists can be reused in case they go up
after a failure. Player always trying to provide the best user
experience.
Added an expiration time field to playlists blacklisted to allow
Exoplayer to continue playback when playlists that failed were
recovered from a bad state.
In live environments, some times occur that primary encoder stop
working for a while. In that cases, HLS failover mechanism in the
player should detect the situation and “switch” to playlists served by
the backup encoder (in case a backup encoder exists). This was well
managed before these changes.
However, and to ensure a playback experience that can recover itself
from temporary issues, we cannot blacklist a playlist forever. When
streaming live events using HLS, it is quite typical that the player
needs to switch from primary to backup playlists, and from backup to
primary ones, from time to time to have playback working when temporary
issues in the network/encoder are happening. Most of the issues are
recoverable, so what I have implemented is a mechanism that makes
blacklisted playlist to be available again after a while (60 seconds).
Evaluation of this algorithm should happen just when something fails.
If player is working with a backup playlist, it shouldn’t switch to the
primary one at least something fail.
Support is provided for the following schemes:
urn:mpeg:dash:utc:direct:2012
urn:mpeg:dash:utc:http-iso:2014
urn:mpeg:dash:utc:http-xsdate:2012
urn:mpeg:dash:utc:http-xsdate:2014
- Data needs to be unescaped before it's passed to SeiReader.
- SeiReader should loop over potentially multiple child messages.
- I also changed the sample passed to the EIA-608 renderer so that
it's the entire sei message payload. The first 8 bytes are
unnecessary, but it seems nicer conceptually to do it this way.
Issue: #295
Previous regular expression for extracting codec information was wrong,
given a line that defines a variant it added information from “CODEC=“
text to the end of the line (including also information about
RESOLUTION or alternate rendition groups as part of the CODEC field).
This is not causing a functional problem (at least known by me)
although is making codecs field storing information that is not related
with the codec.
Some extractor implementations underneath MediaExtractor require a seekTo
call after tracks are selected to ensure samples are read from the correct
position. De-duplicating logic was preventing this from happening in some
cases, causing issues like:
https://github.com/google/ExoPlayer/issues/301
Note that seeking all tracks a side effect of track selection sucks if
you already have one or more tracks selected, because it introduces
discontinuities to the already selected tracks. However, in general, it
*is* necessary to specify the position for the track being selected,
because the underlying extractor doesn't have enough information to know
where to start reading from. It can't determine this based on the read
positions of the already selected tracks, because the samples in these
tracks might be very sparse with respect to time.
I think a more optimal fix would be to change the SampleExtractor
interface to receive the current position as an argument to selectTrack.
For our own extractors, we'd seek the newly selected track to that
position, whilst the already enabled tracks would be left in their
current positions (if possible). For FrameworkSampleExtractor we'd
still have no choice but to call seekTo on the extractor to seek all
of the tracks. This solution ends up being more complex though, because:
- The SampleExtractor then needs a way of telling DefaultSampleSource
which tracks were actually seeked, so that the pendingDiscontinuities
flags can be set correctly.
- It's a weird API that requires the "current playback position to seek
only the track being enabled"
So it may not be worth it! I think this fix is definitely good for now,
in any case.
Issue: #301
- This change:
1. Extracts HlsExtractor interface from TsExtractor.
2. Adds AdtsExtractor for AAC/ADTS streams, which turned out to be
really easy.
Selection of the ADTS extractor relies on seeing the .aac extension.
This is at least guaranteed not to break anything that works already
(since no-one is going to be using .aac as the extension for something
that's not elementary AAC/ADTS).
Issue: #209
I think this is the limit of how far we should be pushing complexity
v.s. efficiency. It's a little complicated to understand, but probably
worth it since the H264 bitstream is the majority of the data.
Issue: #278
Use of Sample objects was inefficient for several reasons:
- Lots of objects (1 per sample, obviously).
- When switching up bitrates, there was a tendency for all Sample
instances to need to expand, which effectively led to our whole
media buffer being GC'd as each Sample discarded its byte[] to
obtain a larger one.
- When a keyframe was encountered, the Sample would typically need
to expand to accommodate it. Over time, this would lead to a
gradual increase in the population of Samples that were sized to
accommodate keyframes. These Sample instances were then typically
underutilized whenever recycled to hold a non-keyframe, leading
to inefficient memory usage.
This CL introduces RollingBuffer, which tightly packs pending sample
data into a byte[]s obtained from an underlying BufferPool. Which
fixes all of the above. There is still an issue where the total
memory allocation may grow when switching up bitrate, but we can
easily fix that from this point, if we choose to restrict the buffer
based on allocation size rather than time.
Issue: #278
- Remove TsExtractor's knowledge of Sample.
- Push handling of Sample objects into SampleQueue as much
as possible. This is a precursor to replacing Sample objects
with a different type of backing memory. Ideally, the
individual readers shouldn't know how the sample data is
stored. This is true after this CL, with the except of the
TODO in H264Reader.
- Avoid double-scanning every H264 sample for NAL units, by
moving the scan for SEI units from SeiReader into H264Reader.
Issue: #278
The complexity around not enabling the video renderer before it
has a valid surface is because MediaCodecTrackRenderer supports
a "discard" mode where it pulls through and discards samples
without a decoder. This mode means that if the demo app were to
enable the renderer before supplying the surface, the renderer
could discard the first few frames prior to getting the surface,
meaning video rendering wouldn't happen until the following sync
frame.
To get a handle on complexity, I think we're better off just removing
support for this mode, which nicely decouples how the demo app
handles surfaces v.s. how it handles enabling/disabling renderers.
Reordering in the extractor isn't going to work well with the
optimizations I'm making there. This change moves sorting back
to the renderer, although keeps all of the renderer
simplifications. It's basically just moving where the sort
happens from one place to another.
I'm not really a fan of micro-optimizations, but given this method
scans through every H264 frame in the HLS case, it seems worthwhile.
The trick here is to examine the first 7 bits of the third byte
first. If they're not all 0s, then we know that we haven't found a
NAL unit, and also that we wont find one at the next two positions.
This allows the loop to increment 3 bytes at a time.
Speedup is around 60% on Art according to some ad-hoc benchmarking.
1. AdtsReader would previously copy all data through an intermediate
adtsBuffer. This change eliminates the additional copy step, and
instead copies directly into Sample objects.
2. PesReader would previously accumulate a whole packet by copying
multiple TS packets into an intermediate buffer. This change
eliminates this copy step. After the change, TS packet buffers
are propagated directly to PesPayloadReaders, which are required
to handle partial payload data correctly. The copy steps in the
extractor are simplified from:
DataSource->Ts_BitArray->Pes_BitArray->Sample->SampleHolder
To:
DataSource->Ts_BitArray->Sample->SampleHolder
Issue: #278
- TsExtractor is now based on ParsableByteArray rather than BitArray.
This makes is much clearer that, for the most part, data is byte
aligned. It will allow us to optimize TsExtractor without worrying
about arbitrary bit offsets.
- BitArray is renamed ParsableBitArray for consistency, and is now
exclusively for bit-stream level reading.
- There are some temporary methods in ParsableByteArray that should be
cleared up once the optimizations are in place.
Issue: #278
This is the start of a sequence of changes to fix the ref'd
github issue. Currently TsExtractor involves multiple memory
copy steps:
DataSource->Ts_BitArray->Pes_BitArray->Sample->SampleHolder
This is inefficient, but more importantly, the copy into
Sample is problematic, because Samples are of dynamically
varying size. The way we end up expanding Sample objects to
be large enough to hold the data being written means that we
end up gradually expanding all Sample objects in the pool
(which wastes memory), and that we generate a lot of GC churn,
particularly when switching to a higher quality which can
trigger all Sample objects to expand.
The fix will be to reduce the copy steps to:
DataSource->TsPacket->SampleHolder
We will track Pes and Sample data with lists of pointers into
TsPackets, rather than actually copying the data. We will
recycle these pointers.
The following steps are approximately how the refactor will
progress:
1. Start reducing use of BitArray. It's going to be way too
complicated to track bit-granularity offsets into multiple packets,
and allow reading across packet boundaries. In practice reads
from Ts packets are all byte aligned except for small sections,
so we'll move over to using ParsableByteArray instead, so we
only need to track byte offsets.
2. Move TsExtractor to use ParsableByteArray except for small
sections where we really need bit-granularity offsets.
3. Do the actual optimization.
Issue: #278
If the timesource track renderer ends, but other track renderers
haven't finished, the player would get stuck in a pending state.
This change enables automatic switching to the media clock in the
case that the timesource renderer has ended, which allows other
renderers to continue to play.
These may occur in VOD streams where a representation's data
is small enough not to require segmentation or an index. For
example subtitle files.
Issue: #268
SampleExtractor will initially only be implemented by FrameworkSampleExtractor
which delegates to a MediaExtractor, but eventually it will also be implemented
by additional extractors.
The sample extractor can be used as a source of samples via DefaultSampleSource.
Also added clamping to getSegmentNum in one case where
it was not already implemented, and defined this behavior
property in the getSegmentNum javadoc.
Issue: #262
Without this, the byte is cast as follows (in bits) if the top
byte is set:
10000010 -> 1000000000000000000000000000010
This works because we then always shift at least one bit left,
and only look at the bottom 8 bits of the result. It's confusing
though. It's clearer if the cast to int gives just adds zeros to
the front, like:
10000010 -> 0000000000000000000000010000010
