Re: [PATCH v2 0/4] MR10421: winegstreamer: Improve quartz MPEG-1 video seeking.

[Akihiro Sagawa (＠sgwaki)]

On Wed Mar 25 14:49:28 2026 +0000, Elizabeth Figura wrote:
> This looks good, but 4/4 gives me pause. It's not wrong, and actually
> testing shows that native seems to go higher, allocating 100 samples.
> But why would it stick after seeking with a more limited pool?
> As you may know, MPEG video uses inter‑frame compression. I'm still checking the details, but it appears that the MPEG video decoder receives P‑frames and B‑frames and keeps the required reference frames until all dependencies are resolved. Because of this, when frames appear in a sequence like PBBPBBPBB after seeking, a buffer of eight samples was not sufficient.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/10421#note_133736