1. Only output video starting from a keyframe
2. When calculating the timestamp offset to adjust live streams to start
at t=0, use the timestamp of the first tag from which a sample is actually
output, rather than just the first audio/video tag. The test streams in
the referenced GitHub issue start with a video tag whose packet type is
AVC_PACKET_TYPE_SEQUENCE_HEADER (i.e. does not contain a sample) and whose
timestamp is set to 0 (i.e. isn't set). The timestamp is set correctly on
tags that from which a sample is actually output.
Issue: #6111
PiperOrigin-RevId: 256147747
We currently don't display the last frame because the seek time is behind the
last frame's timestamps and it's thus marked as decodeOnly.
This case can be detected by checking whether all data sent to the codec is
marked as decodeOnly at the time we read the end of stream signal. If so, we
can re-enable the last frame. This should work for almost all cases because the
end-of-stream signal is read in the same feedInputBuffer loop as the last
frame and we therefore haven't released the last frame buffer yet.
Issue:#2568
PiperOrigin-RevId: 251425870
We are currently queuing periods in a way such that the new start position
lines up with the end of the previous period (to ensure continuous playback).
However, if the start position of the new period is larger than the total of
all previously played period durations, we may end up with negative renderer
timestamps when seeking back to the beginning of this new period. Negative
timestamps should be avoided as most decoders have problems handling them
correctly.
This change forces a renderer reset if we detect such a seek to a negative
renderer time and also resets the renderer offset to 0 every time all
renderers are disabled, as this is the only time where we can savely change
the offset of an existing media period.
Also, if playback starts with an ad, we choose the content position as
renderer offset to prevent the whole issue from occurring for the seek-behind-
midroll case.
Issue:#6009
Issue:#5323
PiperOrigin-RevId: 253790054
In some edge cases the renderer position may be slightly ahead of the
buffered position and the total buffered duration is thus negative. We already
filter that in ExoPlayerImpl for the publicly accessible value. However, we
forward the unfiltered value to other components like the LoadControl, which
may be confusing.
Issue:#6015
PiperOrigin-RevId: 253780460
This permission has normal access right and can't be revoked by the user.
However, an app can choose to revoke it when using ExoPlayer, e.g. if
no network is required and the app doesn't want to list this permission.
Support this use case by gracefully catching the exception in the relevant
places.
Issue:#6019
PiperOrigin-RevId: 253759332
It's only thrown in an edge case on API level 20 and below. If it
is thrown it causes playback failure when playback could succeed,
by throwing up through configureCodec.
It seems better just to catch the exception and have the codec be
configured using the format's own width and height.
PiperOrigin-RevId: 251745539
cache() opens all connections with unset length to avoid position errors.
This makes more data then needed to be downloading by the underlying
network stack.
This fix makes makes it open connections for only required length.
Issue:#5927
PiperOrigin-RevId: 250546175
We currently toggle the view in onTouchEvent ACTION_DOWN which is non-standard
and causes problems when used in a ViewGroup intercepting touch events.
Switch to standard Android click handling instead which is also what most
other player apps are doing.
Issue:#5784
PiperOrigin-RevId: 245219728
Using parallel adaptation for Formats without bitrate information currently
causes an exception. Handle this gracefully and also cases where all formats
have the same bitrate.
Issue:#5971
PiperOrigin-RevId: 250682127
According to Cue's constructor (for bitmaps) documentation:
+ cuePositionAnchor does horizontal anchoring.
+ cueLineAnchor does vertical anchoring.
Usage is currently inverted.
Issue:#5633
PiperOrigin-RevId: 250253002
This prevents further unexpected updates if the MediaSource happens to
finish its preparation at a later point.
Issue:#5915
PiperOrigin-RevId: 249439246
The SurfaceListener just sets the surface on the VideoComponent, but
SphericalSurfaceView already accesses the VideoComponent directly so it seems
simpler to update the surface directly.
PiperOrigin-RevId: 249242185
If a PlayerMessage throws an exception, it is currently not deleted from the
list of pending messages. This may be problematic as the list of pending
messages is kept when the player is retried without reset and the message is
sent again in such a case.
PiperOrigin-RevId: 247414494
Adding an explicit option to clear all downloads prevents repeated database
access in a loop when trying to delete all downloads.
However, we still create an arbitrary number of parallel Task threads for this
and seperate callbacks for each download.
PiperOrigin-RevId: 247234181
As justification for why we should not have this type of logging,
it would scale up to about 13K LOC, 1800 Strings, and 36K (after
pro-guarding - in the case of the demo app) if we did it through
the whole code base*. It makes the code messier to read, and in
most cases doesn't add significant value.
Note: I left the Scheduler logging because it logs interactions
with some awkward library components outside of ExoPlayer, so is
perhaps a bit more justified.
* This is a bit unfair since realistically we wouldn't ever add
lots of logging into trivial classes. But I think it is fair
to say that the deltas would be non-negligible.
PiperOrigin-RevId: 246181421
- Removed DownloadInternal and its sometimes-out-of-sync
duplicate state
- Fixed downloads being in STOPPED rather than QUEUED state
when the manager is paused
- Fixed setMaxParallelDownloads to start/stop downloads if
necessary when the value changes
- Fixed isWaitingForRequirements
PiperOrigin-RevId: 246164845
Can happen if the load position falls behind in every playlist and
when we try to load the next segment, the adaptive selection logic
decides to change variant.
Issue:#5816
PiperOrigin-RevId: 245923006
This resolves some naming confusion that previously existed
as a result of DownloadThread also being used for removals.
Some related variables (e.g. activeDownloadCount) would refer
to both download and removal tasks, whilst others
(e.g. maxParallelDownloads) would refer only to downloads.
This change renames those that refer to both to use "task"
terminology.
This change also includes minor test edits.
PiperOrigin-RevId: 245913671
Experiments show this is beneficial for rebuffers with only minor impact
on battery usage.
Configurations which explicitly set a minimum buffer duration are unaffected.
Issue:#2083
PiperOrigin-RevId: 244823642
There are no logic changes here. It's just moving code around and removing
the "internal" part of names where no longer required.
PiperOrigin-RevId: 245407238
This ensures we keep the loading period in sync with the the playing period in
PlybackInfo, when the latter changes to something new.
PiperOrigin-RevId: 244838123
- We had buildAddRequest and sendNewDownload. Converged to
buildAddDownload and sendAddDownload.
- Also fixed a few more inconsistencies, and brought the
action constants into line as well.
PiperOrigin-RevId: 244274041
- Listener based reporting of progress allows the content length
to be persisted into the download index (and notified via a
download state change) as soon as it's available.
- Moved contentLength back into Download proper. It should only
ever change once, so I'm not sure it belongs in the mutable part
of Download.
- Made a DownloadProgress class, for naming sanity.
PiperOrigin-RevId: 244242487
1. customCacheKey for DASH/HLS/SS is now asserted against
in DownloadRequest
2. Merging of event delivery in DownloadManager is very
tricky to get right and probably not a good idea
PiperOrigin-RevId: 244048392
Currently SimpleCache will touch cache spans whenever it reads
from them. With legacy SimpleCache setups this involves a potentially
expensive file rename. With new SimpleCache setups it involves
a more efficient but still non-free database write.
For offline use cases, and more generally any use case where the
eviction policy doesn't use last access timestamps, touching is
not useful. This change allows the evictor to specify whether it
needs cache spans to be touched or not. SimpleCache will only touch
spans if the evictor requires it.
Note: There is a potential change in behavior in cases where a
cache uses an evictor that doesn't need cache spans to be touched,
but then later switches to an evictor that does. The new evictor
may temporarily make sub-optimal eviction decisions as a result.
I think this is a very fair trade-off, since this scenario is
unlikely to occur much, if at all, in practice, and even if it
does occur the result isn't that bad.
PiperOrigin-RevId: 244005682
One goal we forgot about a little bit was to allow applications
to provide their own index implementation. This requires the
writable side to also be defined by an interface.
PiperOrigin-RevId: 243979660
ImaAdsLoader gets the player position after the app releases
the player to support resuming ads at their current position
if the same ads loader is reused.
PiperOrigin-RevId: 243969916
The removed condition only applies when prepare fails and no timeline was or
is known. We should keep a pending seek position in such a case. The internal
player does the same thing already.
PiperOrigin-RevId: 243819466
- Expose constructors that take DatabaseProvider. Deprecate those
that do not.
- Expose Cache.getUid. This will likely be used for naming of the
tables accessed by DefaultDownloadIndex .
PiperOrigin-RevId: 243637786
- Allows enforcing immutability, which in a future CL will allow
avoiding allocating a new array/list on every call.
- Also some left over doc cleanup from the DownloadState rename.
PiperOrigin-RevId: 243627352
The old domain automatically forwards to the new one. For consistency, change
all doc and code references regardless.
Also adds GitHub CNAME config file which configures our page for the custom
domain.
PiperOrigin-RevId: 243592110
handleSourceInfoRefreshed checks for various conditions which may trigger a
change in position. In all cases, we need to resolve the new position for ads,
and if the corresponding MediaPeriods already exist, perform the seek to the new
position.
This change simplified the code by
1. Moving all conditions in a single if-else structure which just determines
the new playback position. This makes it easier to read and to follow.
2. Doing the subsequent seek and position update in one final step to remove
duplicated code.
As a side effect, it also improves some edge cases. For example when skipping
an ad to an already prebuffered period, we now properly update the existing
prebuffered period.
PiperOrigin-RevId: 243588014
- Make a best effort to avoid posting MSG_DOWNLOAD_THREAD_STOPPED
if the internal thread has already quit. Doing so is harmless,
but causes an error in Logcat. We used to generate this warning
quite a bit in ExoPlayerImplInternal as well, and we got quite
a few issues filed about it, so best to avoid it if possible.
- Null out the back reference DownloadThread holds to its manager
on release. This avoids a potential issue where a download thread
can prevent GC of its manager if it doesn't cancel quickly.
There's similar logic (with a similar comment) in Loader.LoadTask.
PiperOrigin-RevId: 243365143
- Rename getAllDownloads to getCurrentDownloads to make it clear
that it doesn't include completed and failed downloads
- Change getDownloadCount to isWaitingForRequirements, which is
what it's used for. Added TODO to make it returns the right thing
PiperOrigin-RevId: 243257856
If the returned retry delay is unset, we should use a DONT_RETRY_FATAL action
instead and dispatch the event with the "canceled" flag.
PiperOrigin-RevId: 243251554
Non-trivial download state transitions are currently split across
DownloadManager and Download. These transitions are part of the
same state machine, so it's clearer if they're all in the same place
(i.e. DownloadManager, since this is the component that transitions
downloads between states).
PiperOrigin-RevId: 243249915
The only state DownloadInternal holds is duplicate of state contained
in Download, but can confusingly be temporarily out of sync. This is
error prone because it's easy to use the wrong variable (e.g.
download.state vs state). DownloadInternals methods are called into
and call out into DownloadManager, which makes some code paths that
are quite hard to follow.
It's possible to simplify DownloadManager quite a lot by removing the
duplicated state in DownloadInternal, at which point DownloadInternal's
methods flatten into DownloadManager, which can just hold an internal
list of Downloads directly.
This is a first step, which makes it clear that DownloadThready only
needs its immutable DownloadAction + an isRemove flag.
PiperOrigin-RevId: 243245288
Put methods into a more sensible order (the same order as
in the switch statement, which is a more natural order with
e.g. initialize coming first and release coming last).
PiperOrigin-RevId: 242891742
- Changed startDownloads/stopDownloads back to their previous behavior
of starting and stopping all downloads at the manager level.
- Made setManualStopReason methods for the new case of setting a manual
stop reason or some or all downloads.
- Added plumbing to specify an initial manual stop reason when adding a
new download, without also overwriting the manual stop reasons of all
other preexisting downloads. Using the value is left as a TODO pending
a bit of further cleanup that'll make it easier.
PiperOrigin-RevId: 242891688
- Remove completed downloads from progress notification.
- Treat restarting as a downloading, When there are both
remove and download tasks ongoing, we prefer to show a
download notification to a remove one. A restarting task
encapsulates is basically a (remove+download) task. Given
our preference, we should treat it as a download rather
than as a removal.
PiperOrigin-RevId: 242867228
Both manifest loads currently use the default retry delay although a
LoadErrorHandlingPolicy is provided to the MediaSource.
PiperOrigin-RevId: 242852672
- Remove NETWORK_TYPE_NOT_ROAMING and NETWORK_TYPE_METERED
because JobScheduler doesn't support them, and they're
probably not useful to many people (data when roaming is
normally enabled/disabled at the OS level, and restricting
to *only* metered networks seems niche)
- Convert network requirements proper flags
- Stop persisting requirements in DownloadIndex. The direction
we're headed to solve the manager start/stop problem is
going to involve state in DownloadManager determining whether
downloads actually start, and if we're doing that then it's
no worse to do it for this as well
PiperOrigin-RevId: 242713196
Robolectric calls messages posted to the main thread immediately inline. This
will change in the upcoming release and will break tests implicitly relying
on that behavior.
Update ConcatenatingMediaSourceTest to actually wait for the posted message.
PiperOrigin-RevId: 242631531
HashMap.values does not guarantee a stable order, but our tests
assert that the order is stable. It's probably a nice property
that the order is the same as that in which downloads are added,
so this change makes the order stable, with that order.
A future change will likely improve the way this works, by
changing the internal thread "downloads" variable to have type
ArrayList<Download>, at which point the internal thread can just
pass a read-only copy of its list to the main thread whenever a
state changes.
PiperOrigin-RevId: 242460725
- Fix ID being dropped when DownloadAction is serialized and
de-serialized as it's passed to DownloadService
- Properly set DownloadAction.id when building an action from
a DownloadState
- Make ID a required constructor argument. Else it's too easy
to not propagate it by accident.
PiperOrigin-RevId: 242457831
