[dallo28]

Fixes https://bugs.telegram.org/c/67051 and https://bugs.telegram.org/c/67049: some iPhone videos fail to send from the attachment menu. The message gets a red error mark at once while the "Processing" ring keeps spinning.

Cause

With "Record Sound: Spatial Audio", the Camera app writes an APAC track (Apple Positional Audio, 4-channel first-order ambisonics) next to an AAC stereo track. In the original file the AAC track comes first. Trimming the video in Photos rewrites it with the APAC track first and enabled, and the AAC track disabled:

	Original (sends)	Trimmed in Photos (fails)
1st audio track	AAC stereo, enabled	APAC 4 ch, enabled
2nd audio track	APAC 4 ch, disabled	AAC stereo, disabled
metadata tracks	6	1

TGMediaVideoConverter (unedited sends) and MediaEditorVideoExport (edited sends, stories) both take the first audio track and decode it with no channel constraint. The AAC writer input refuses the first buffer whose channel count differs from its own (AVFoundationErrorDomain -11800, OSStatus -12780), so the conversion fails before any upload starts. HDR isn't involved: the video track is the same HLG HEVC in both files.

The same failure hits more than Spatial Audio. Testing every layout in the app showed it for 4- and 9-channel ambisonic PCM (ProRes Spatial Audio), B-format, quad and 5.1 tracks, and for plain stereo PCM whenever the quality preset encodes mono audio (AAC sources happen to be downmixed by the decoder, PCM ones are not).

Change

• Track: take an AAC track when the file has one, otherwise the first audio track as before. For Spatial Audio videos this is the stereo mix the camera recorded, and it skips rendering the APAC track (about 60 ms of CPU per second of audio). _preferredAudioTrackForAVAsset: in TGMediaVideoConverter, preferredAudioTrack(in:) in MediaEditorUtils.swift (the main video and collage/dual-video inputs; the added-music track is unchanged).
• Decoding: the audio reader decodes straight to the encoder's channel count and layout (from audioSettingsForPreset: / configuration.audioSettings), so Core Audio downmixes or renders whatever track is used: APAC and PCM ambisonics (ACN SN3D/N3D, B-format) are rendered as a sound field (a source on the left stays on the left), 5.1 and quad are downmixed, stereo is folded to mono when the preset is mono. Mono and stereo AAC sources come out identical to before (same channels, length and level). The channel count stays the preset's, as for every other video: stereo at Medium and above (so a camera video's APAC is rendered to stereo), mono at Low and Very Low, none for animations and profile videos.
• Logging: FetchVideoMediaResource logs conversion and export errors (library and local-file paths, both pipelines) under FetchVideoResource; the legacy converter reports the writer's error when the reader has none (it reported nil before) and logs an exception raised by appendSampleBuffer.

"Send as file" still uploads the original file untouched, including the APAC track.

Logic

Both converters now handle audio the same way:

1. Pick the track. The first AAC track in the file, even if it is disabled (Photos disables it in trimmed Spatial Audio videos). With no AAC track, the first audio track, whatever its format.

2. Decode to the encoder's format. The reader outputs linear PCM with the channel count and layout of the AAC encoder, which come from the quality preset:

   Preset	Encoder	What happens to the source audio
   Medium, High, Very High	AAC stereo, 44.1 kHz	AAC/PCM stereo passes through; mono is duplicated to both channels; APAC and PCM ambisonics are rendered to stereo as a sound field; 5.1 and quad are downmixed
   Low, Very Low	AAC mono, 44.1 kHz	everything is folded to mono (ambisonics rendered first)
   Animation, profile	no audio	audio is dropped, as before

3. Encode with the same AAC settings as before. Sample-rate conversion stays with the encoder, as before.

In MediaEditorVideoExport the same settings go to the AVAssetReaderAudioMixOutput, so the editor's volume mix and an added music track are mixed into that same layout. When the encoder settings have no channel count (stickers, avatars), the reader settings stay nil, as before.

Testing

Built with Make.py build --configuration=debug_sim_arm64 and run in the iOS 27.0 simulator (iPhone 17e), logged in, sending each file on its own to Saved Messages from the attachment menu, then checking the app log and the converted file in the media cache:

File	Audio tracks	Before	After
Reporter's file trimmed in Photos	APAC first, AAC (disabled)	fails	sent, from the AAC track
Reporter's original file	AAC first, APAC	sent	sent, unchanged
t14	APAC first, AAC stereo	fails	sent, from the AAC track
t01	APAC only	fails	sent, APAC rendered
t05	APAC 9 ch (2nd order)	fails	sent
t02b	PCM 4 ch ambisonics, PCM stereo (ProRes layout)	fails	sent
t10	PCM 4 ch ambisonics (N3D)	fails	sent
t03	PCM B-format	fails	sent
t04	PCM 9 ch ambisonics	fails	sent
t11	PCM quad	fails	sent
t06	AAC 5.1	fails	sent
t09	PCM stereo (mono preset)	fails	sent
t07	AAC mono	sent	sent
t08	no audio	sent	sent

Edited sends (trimmed in Telegram's editor, MediaEditorVideoExport): the reporter's trimmed file (AAC track used, L/R correlation 0.99 like the camera's AAC) and t10 (left-panned tone rendered left: L −10.0 dBFS, R −27.3 dBFS) both sent.

Which track was used was checked on the output: the reporter's file now matches its AAC track (L/R correlation 0.99; the rendered APAC is 0.95), and t14, whose AAC track carries the tone in the left channel only, comes out like that AAC track rather than like the rendered APAC. Every output keeps the source length (minus the 0.1 s the legacy converter always trims) and level.

Not tested on a device. Please send a Photos-trimmed Spatial Audio video from the attachment menu on an iPhone 16 or 17.

Limitations

• If Photos stores an "Audio Mix" edit (Cinematic/Studio) only as metadata for playback, neither the AAC track nor the rendered APAC includes it. Not verified.
• When a conversion does fail, the message still shows the "Processing" ring until the chat is closed (FetchManagerImpl has no error handler); an album with one failing item waits for the rest, then sends without it. The new log lines at least show why.