Skip to content

FFmpeg update to 9.0 #2

Open
Open

Description

@andrefmsilva

Maybe is time to start planing FFmpeg update to recent newer release.

Activity

  1. superuser404notfound commented on Aug 14, 2026

    @superuser404notfound
    Owner

    Thanks for the nudge. This build is pinned to n8.1.2, and 9.0.1 was checked against it. The short version: the update is mechanically cheap, but it currently buys this build nothing, so it is deferred rather than dropped.

    What was measured against n9.0.1:

    • All four FFmpeg source patches in build.sh apply unchanged (yadif_videotoolbox autorelease balance, pgssubdec missing-palette recovery, the visionOS OpenGL ES buffer attribute, and the matroska TrackTimestampScale warning). None of them has landed upstream, so all four are still needed.
    • The public API delta is small. 35 symbols were removed between n8.1.2 and n9.0.1, and the consumer (AetherEngine) uses none of them. Struct field removals are not covered by a symbol diff, so a real compile remains the actual proof.
    • Every component enabled here (decoders, demuxers, parsers, bitstream filters, filters, muxers, protocols) still exists in 9.0.1.
    • All sonames bump: lavc and lavf 62 to 63, lavu 60 to 61, lavfi 11 to 12, lsws 9 to 10, lswr 6 to 7. So the move would be a major version of this package.

    Why not right now:

    • Nothing in 9.0 changes decode behaviour for the component set this build ships. The one genuinely interesting addition is the new dovi_split bitstream filter for multi-layer Dolby Vision HEVC, and that is a feature of its own rather than a reason to move the pin.
    • release/8.1 is still maintained upstream (40 commits since n8.1.2, most recent on 2026-08-03), so staying there is not a security or bugfix backlog.
    • A batch of reporter retests is currently in flight against the AetherEngine releases that consume this package. Moving to a new FFmpeg major underneath them would give every failed retest a second possible cause.

    Plan: take n8.1.3 as a cheap intermediate step when it appears, then move to a 9.0.x once that retest batch has drained. This issue stays open as the tracker and will get an update when the pin moves.

  2. superuser404notfound commented on Sep 22, 2026

    @superuser404notfound
    Owner

    The intermediate step from the plan above is done: 3.5.0 builds on FFmpeg n8.1.3.

    Six of the 268 upstream commits touch code this build ships, among them a heap out-of-bounds read and write in VP9 after avcodec_flush_buffers() under frame threading, and a Dolby Vision RPU that lost its dynamic metadata when it carried trailing zero padding. Soname majors are unchanged, and all six local patches still apply.

    9.0 stays on this tracker. The measurement from August still holds: the only thing in it for a player is ff_dovi_split_bsf, and every soname major moves, so it wants its own window rather than riding along with a security and correctness step.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions