TextEngine was storing closed captions and releasing them only on
teardown. This might be a potential memory leak for long running
sessions.
This PR addresses it by removing cached captions along with removed
segments. Furthermore, it adds some more tweaks and improvements, such
as:
- simplify CEA cache structure, from `Map<string, Map<string,
Array<Cue>>>` to `Map<string, Array<Cue>>`
- adjust TextEngine buffered ranges with cached cues values
Close https://github.com/shaka-project/shaka-player/issues/8519
Fixes https://github.com/shaka-project/shaka-player/issues/8475
Introduce `NativeTextDisplayer` as a replacement of
`SimpleTextDisplayer`. But keep them both work.
Is MSE mode, `NativeTextDisplayer` creates `<track>` elements for text
streams. And listens to change events on both ends to keep them in sync.
In SRC mode, `NativeTextDisplayer` would do nothing, the player uses
original TextTracks instead.
Advantages of `NativeTextDisplayer`:
- Allow text track selection using the browser built-in UI
- Allow text track manipulation using native APIs
- Avoid transferring and processing cues in SRC mode
---------
Co-authored-by: Álvaro Velad Galván <ladvan91@hotmail.com>
Changes:
- The look has been changed to make it more similar to YouTube:
- The main background color is now black, and the font is white.
- Presentation time has been moved to the bottom.
- Cast and airplay buttons are now more accessible.
- Tooltips have been enabled except on mobile platforms.
- The ad information has been moved to appear in the same position as
the presentation time when the ad is present.
- A mark indicating the current quality has been added (e.g.: HD, 2K,
4K, 8K)
- The spinner has been replaced with one that works well on Smart TVs
and is very similar to the current one. The animation is included in the
SVG element itself rather than through CSS.
- More LESS variables have been added to make customization easier in
forks.
- The maximum size of the menus is dynamically calculated so that they
never extend outside the video container.
- The size of the subtitle container when the UI appears is now
calculated dynamically.
- The Demo has been updated to show the seekbar when trickplays are
enabled.
- UI performance on Smart TVs has been improved (Tested on Tizen 5.0)
- Many offsets that were hardcoded have been removed, but not all (in
CSS).
Close https://github.com/shaka-project/shaka-player/issues/8406
Fixes#8278
Migrates deprecated rules from `eslint-config-google` and our rules as
well to stylistic.
Additionally removes broken `eslint-disable` python check and replaces
with eslint `reportUnusedDisableDirectives` option.
With reference to #8025, this is a partial proposal to see if this style
of extensible customisation could be favourable to the shaka-player
community:
Proposal for some accessibility options whereby an App builder can
customise subtitle/caption size, with the potential of an App offering
accessibility options for text size.
This style could be used for other text styling attributes in a similar
way, but this is just the first step.
---------
Co-authored-by: Álvaro Velad Galván <ladvan91@hotmail.com>
Fixes https://github.com/shaka-project/shaka-player/issues/8087
Implements handling of multiple samples in a MP4/ISOBMFF/DASH TTML
segment/fragment. Such segments are allowed by ISO14496-12 and
ISO23000-19. gpac creates such segments. The prior code just treated the
full MDAT as one TTML XML document and tried to parse it in whole
without accounting for sample(s). A testcase is included which was
created by taking the testdata from ttml-segment.mp4 and splitting the
subtitles into two independent TTML-XML documents, which then were put
as individual samples.
The testdata for the prior existing multiple MDAT testcase was invalid.
It was created by taking the same ttml-segment.mp4 as a source and just
duplicating the MDAT box, but without then also fixing the TRUN box. The
duplicated data was thus not referenced. The test case still worked,
because the prior code did not look at the TRUN box and the sample
specification at all and just handled any full MDAT box = 1 sample. The
testdata was replaced with a new file, which is basically the same as
for the multiple samples case, but with the two samples split into two
MDAT boxes.
This moves VTT sequence mode offset calculations into a method.
It also makes all X-TIMESTAMP-MAP usage dependent on HLS specifically,
rather than sequence mode, simplifying the conditions. Sequence mode is
typically only used with HLS, and X-TIMESTAMP-MAP is explicitly only for
HLS. So excluding X-TIMESTAMP-MAP for DASH makes sense, instead of
conflating HLS and sequence mode.
This required updating some tests to explicitly set both the manifest
type and sequence mode flag.
This does *not* change the offset calculations themselves. Changes will
be made in follow-up PRs.
Issue #6320
Álvaro Velad Galván
Gary Katsevman
Wojciech Tyczyński
xiao
David HM Morgan
Juliane Holzt
Wojciech Tyczyński
MichaelSweden
Joey Parrish
theodab
Dave Nicholas
Antonio Díaz Correa