Fix AV1 videos being uploaded without transcoding, and two AV1 playback defects

[ThatHunky]

Fixes the AV1 send-side defect reported in #487, plus two related AV1 playback defects.

Not built or device-tested. I have no Android SDK/NDK set up, and gifvideo.cpp needs an NDK rebuild. The diagnosis below is evidence-backed and the change applies cleanly, but please build and test before merging. Flagging this up front rather than letting a reviewer discover it.

1. AV1 videos are uploaded without transcoding — gifvideo.cpp

nGetVideoInfo reports PARAM_NUM_SUPPORTED_VIDEO_CODEC from a hardcoded allowlist:

codec_id == AV_CODEC_ID_H264 || H263 || MPEG4 || VP8 || VP9 ||
(sdkVersion > 21 && codec_id == AV_CODEC_ID_HEVC);

AV_CODEC_ID_AV1 is absent. The consequence chain:

gifvideo.cpp returns 0 → PhotoViewer sets videoConvertSupported = false → the entire compression-setup block is skipped, so resultWidth/compressionsCount/selectedCompression are never populated → compressItem.setState(videoConvertSupported && compressionsCount > 1, …) disables the compression UI → no videoEditedInfo carrying convert params → the SendMessagesHelper gate videoEditedInfo != null && needConvert() fails → the original file is uploaded verbatim.

It fails silently because it isn't an error path — it's a disabled feature. No error, no compression option, no indication the file went out unprocessed.

The fix adds AV1 to the allowlist, gated on API 29 where AOSP ships c2.android.av1-dec, mirroring the existing HEVC guard. Note the convert pipeline decodes via platform MediaCodec, not the bundled FFmpeg — so no build_ffmpeg.sh change is required.

Evidence

Device captures on a Pixel 9 Pro (Telegram 12.9.1, Android 17), logging enabled:

Source	Res	Route	Converted?	Codecs instantiated	Uploaded from
AV1	4K	share sheet	❌	—	cache/sharing/<original>
HEVC	1080p	picker	✅	exynos.hevc.decoder + exynos.h264.encoder	generated cache file
AV1	4K	picker	❌	c2.google.av1.decoder only — no encoder	/Download/… verbatim
HEVC	4K	picker	✅	exynos.hevc.decoder + exynos.h264.encoder	generated cache file

Rows 3 and 4 are the controlled pair — identical resolution, duration, HDR characteristics and route; only the codec differs:

begin convert …/TEST_4K_HEVC_control.mp4 rWidth = 1080 rHeight = 1920 oWidth = 2160 oHeight = 3840
compression completed time=27387 needCompress=true w=1080 h=1920 file size=23,8 MB encoder_name=c2.exynos.h264.encoder

For AV1, neither line appears, no encoder is created, and upload starts at file_part=0 straight from the on-disk path. The AV1 file was freshly re-encoded (different SHA256 and different decoded-frame MD5), ruling out dedup.

2. One transient decoder error permanently disabled AV1 — VideoPlayer.java

onPlayerError wrote unsupport_video/av01 on any MediaCodecDecoderException mentioning av1, and nothing ever cleared it — no expiry, no retry, no reset on upgrade. supportsHardwareDecoder() then returned false for the life of the install, which gates thumbnails, playback, and strips AV1 URIs from the quality list. Codec-instance exhaustion from opening several videos at once is enough to trigger it, after which a device with working hardware AV1 behaves as if it had none. Only recovery was clearing app data.

Replaced with a failure counter (threshold 3) that expires after 30 days and self-resets on app upgrade or when the verdict goes stale, and CodecException.isTransient() failures aren't counted at all.

3. Software decoders were never considered — VideoPlayer.java

supportsHardwareDecoder() skipped every non-hardware-accelerated codec. Android 10+ ships a software AV1 decoder, but the method was used as a blanket gate, so on devices without hardware AV1 the video track was dropped entirely while audio kept playing — the widely reported "AV1 plays sound with a black screen". The video wasn't failing to decode; it was never being decoded.

Added supportsSoftwareDecoder() and supportsDecoder(codec, w, h) (software permitted up to 1080p), used for the three hard gates that discarded the track. supportsHardwareDecoder() is retained for quality-selection heuristics, where preferring hardware is correct.

Review notes

Things a reviewer should push on, stated rather than hidden:

• Retry termination. The AV1 error path now retries once per player instance (triedAv1CodecFallback), reset on first-frame render and in releasePlayer — neither is on the retry path. Below the blacklist threshold filterByCodec may strip nothing, so an unbounded retry would loop forever.
• isCodecBlacklisted() writes SharedPreferences from what reads as a predicate (the self-healing reset). Self-limiting: only reached on a cache miss.
• Two unsynchronized static caches. Upstream already had one; this adds a second following the same pattern. Synchronizing would change behavior beyond this fix's scope.
• reportCodecFailure's read-modify-write isn't atomic — concurrent failures can lose an increment. That only makes blacklisting slower, never wrong.
• Residual: three genuinely corrupt AV1 files within 30 days can still blacklist a working codec until the TTL lapses. Wiring a success-path reset needs per-URI codec tracking in onRenderedFirstFrame, which felt too invasive here.

Also reported upstream at https://bugs.telegram.org/c/64017 (DrKLO/Telegram has issues disabled).