refactor(client-core/W8): split video.rs into flat decoder-backend siblings

Break the 1974-line pf-client-core/src/video.rs into flat sibling modules
(matching the crate's video_d3d11.rs / video_pyrowave.rs convention), leaving
video.rs as the contract + Decoder dispatch facade:
  - video_color.rs   : ColorDesc + csc_rows (the Y'CbCr->RGB matrix)
  - video_software.rs : the libavcodec/swscale SoftwareDecoder
  - video_vaapi.rs   : the Linux-only VAAPI/DRM-PRIME backend (mod is cfg(linux))
  - video_vulkan.rs  : the FFmpeg Vulkan Video backend
Every crate::video::X / video::X path stays byte-stable (ColorDesc + csc_rows
re-exported from video.rs; frame POD, VulkanDecodeDevice, QueueLock, Decoder,
decodable_codecs*, ffmpeg_codec_id, fourcc/drm_fourcc_for all stay in video.rs).
Code-driven placements: averr, AVERROR_EAGAIN, frame_is_keyframe stay in video.rs
(shared by all three decoders); DrmFrameGuard's field + drm_fourcc_for +
Software/Vaapi/VulkanDecoder ctors/decode became pub(crate) (sibling access);
the test module split three ways (software tests need private decoder internals).
Pure move; no behavior change.

Verified on Linux (home-worker-5): cargo clippy -p pf-client-core (default
[pyrowave] + --no-default-features, --all-targets -D warnings) + cargo test.
Windows verify BLOCKED environmentally: pf-client-core -> sdl3 build-from-source
-> CMake/CL.exe fails on winbox's non-ASCII home path (fails the baseline too,
independent of this split); the split's Windows surface (facade cfg(windows) bits
+ video_d3d11) is verbatim-preserved.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

This commit is contained in:

Enrico Bühler

2026-07-17 14:06:57 +02:00

parent ffa63a74f2

commit 570ff504ad

6 changed files with 1154 additions and 1101 deletions

crates/pf-client-core/src/video.rs

+10 -1101

View File

File diff suppressed because it is too large Load Diff