## Description
Adds a "Prefer Dolby Vision over HDR10+" switch (off by default) under
Experimental settings. That screen only shows up once "Experimental
settings" is enabled under Advanced, and until then the switch is read
as off: it goes through `experimentalPreferences.get {}`, so nothing is
wrapped and the playback path is untouched.
Most current 4K web releases carry both Dolby Vision (profile 8.1) RPUs
and HDR10+ dynamic metadata in the same HEVC stream. Some MediaTek based
devices (Philips and TCL TVs, some Fire TV models) latch onto the HDR10+
metadata on those files and never engage Dolby Vision, even though a
Dolby Vision decoder is selected.
I isolated this on my Philips 55OLED810 by playing the same episode
three ways, direct play in all three cases:
| | Stream | TV shows |
|---|---|---|
| A | MKV with DV 8.1 + HDR10+ | HDR10+ |
| B | Same MKV with the HDR10+ metadata removed using
[hdr10plus_tool](https://github.com/quietvoid/hdr10plus_tool), DV
untouched | Dolby Vision |
| C | Same stream remuxed to MP4, both kinds of metadata present |
HDR10+ |
The container doesn't matter and the Dolby Vision signaling is picked up
correctly. The only thing the app can do is keep the HDR10+ metadata
away from the decoder.
When the switch is on, and only when a Dolby Vision decoder was actually
selected, the HDR10+ SEI messages (SMPTE ST 2094-40, recognized by their
ITU-T T.35 header `B5 003C 0001`) are masked in each access unit before
it is queued to MediaCodec. The SEI payload type byte is overwritten
with an unused value (254), so the decoder skips the message. That is
one byte rewritten in place per message: the masking filter returns the
size it was given, so the access unit is queued exactly as long as it
came out of the extractor, which is what makes it safe with both the
synchronous and asynchronous MediaCodec adapters. media3 does the same
thing for HAGC metadata in `HevcSeiUtil`; this filter only targets ST
2094-40, so other T.35 SEI such as CEA-708 captions are not affected.
The filter was written from the H.265 spec, no Kodi or media3 code was
copied.
The hook is generic rather than HDR10+ specific, as asked in the thread.
`BitstreamFilteringCodecAdapterFactory` wraps the codec adapter with a
`List<BitstreamFilter>`, and a filter is `(ByteBuffer, offset, size) ->
Int` where the returned `Int` is the new size of the access unit: the
adapter queues the size the chain reports, and hands each filter the
size the previous one reported. A filter which only overwrites bytes
returns the size unchanged, which is what the HDR10+ masking does; a
filter which drops NAL units compacts what it keeps and reports the
shorter size. A reported size which is negative or larger than the one
the filter was given is ignored rather than queued. That is what lets
the Dolby Vision profile 7 to 8.1 conversion of #244 arrive later as one
more filter instead of a second copy of the adapter plumbing.
When the switch is off, the playback path is unchanged. On a device
without a Dolby Vision decoder (or a display without DV support) the
wrapper is never installed, so HDR10+ only devices keep their metadata.
Encrypted samples, codec config buffers and empty buffers are passed
through untouched.
Kodi has an equivalent option for the same reason: xbmc/xbmc@032cbfeb
and xbmc/xbmc@e2c54e7f.
### Related issues
Implements what was requested in #1855. The filter list is what #244
(Dolby Vision profile 7 direct play) needs, so that conversion can be
added as another filter. #1895 implemented the same masking
independently and was closed in favour of this PR; the device profile
part of it was not carried over, see the note under Testing. Upstream:
androidx/media#1895, where the maintainers say the bug is in the device
codecs and that they don't plan a workaround in ExoPlayer itself.
## AI or LLM usage
The implementation (bitstream filter, MediaCodecAdapter wiring, unit
tests) were written with AI assistance (Claude). I reviewed the changes
and validated the behavior myself on a real MediaTek TV (Philips
55OLED810), with the setting on and off, on hybrid and DV-only content.
---------
Co-authored-by: Jycreyn <github.2ilp7@passmail.com>
## Description
Fixes the home page reloading from scratch when the app is resumed
This happened because the `userDto` is updated on app resume and since
it has a last accessed timestamp, it would always different and trigger
a reload.
### Related issues
Bug introduced by #1981
### Testing
Emulator
## Screenshots
N/A
## AI or LLM usage
None
## Description
When clicking on an episode, the app assumes if there is a season ID for
the episode, then there must be a series ID. However, on my JF 12.0
server, an episode (with no metadata) appeared on my Continue Watching
and it has a season ID (which in turn has a series ID) but no series ID.
I'm not sure if this really a valid situation or if it's a server bug,
but Wholphin shouldn't crash at least. So this PR checks both `seasonId`
and `seriesId` and if either is null, falls back to the generic "Episode
Page". Similarly, if a season has no `seriesId` (again, not sure if a
valid situation), a crash won't occur, however there is no generic
"Season Page", so an error page is shown.
### Related issues
TODO, check JF server issues, maybe fixed in 12.1?
### Testing
Emulator
## Screenshots
N/a
## AI or LLM usage
None
## Description
There is a race condition when switching users between
`HomeViewModel.init` and `HomeSetttingsService.loadCurrentSettings`.
This happens more often if the new user doesn't have a customized home
page. The result is that the previous user's home page settings are used
for the initial load of the home page.
This PR fixes this by collecting the flows for current user & home
settings, waiting until both are valid and refer to the same user, then
fetch the data.
### Related issues
Fixes#1976
Maybe related (and fixes) #1980, but need to investigate more
### Testing
Emulator. It's easy to reproduce by adding a manual `delay(2.seconds)`
at the start of `HomeSetttingsService.loadCurrentSettings`.
## Screenshots
N/A
## AI or LLM usage
None
## Description
Makes the `itemId` parameter optional allowing just passing
`serverId`/`userId` to switch the user and load the app at the home page
for that user.
Dev note: I added a new assert function `assertIs` which asserts that an
object `is` of a type. It also has a contract so the compiler will
smart-cast usages after the `assertIs`. Checking a result type is a very
common pattern in the unit tests, so this assert will simplify a lot of
current & future tests.
### Related issues
Closes https://github.com/damontecres/Wholphin/discussions/1824
### Testing
Emulator and added unit tests for `parseIntent`
## Screenshots
N/A
## AI or LLM usage
None
## Description
Fix two bugs:
- Favorites page didn't query for favorite people, introduced by #1864
- Fixes restoring focus on the search page after clicking a discover
item and going back
### Related issues
N/A
### Testing
Emulator
## Screenshots
N/A
## AI or LLM usage
None
