Skip to content

browser: Upgrade the bundled CEF from 7871 (Chromium 150, end of life) to a supported Chromium #1864

Open

Description

@zackees

Problem. CMakePresets.json pins cef 7871 from obs-rust/cef (Chromium 150.0.7871.252), which no longer gets security fixes. #1092 tracks that, plus the no_sandbox = true work that @vpacher is doing for Linux. In #1092 the CEF upgrade itself was left to the maintainer, because it needs new archives in the obs-rust/cef mirror. This issue is that upgrade on its own. It also blocks the browser source in the AppImage.

We are a fork: upstream OBS's CEF bump can't simply be copied. obs-browser has to build and run against the new CEF API through our libobs.

Steps.

  1. Pick the CEF branch for the current Chromium stable/extended-stable, and record the version and where it comes from.
  2. Build or fetch archives for every platform in CMakePresets.json (macOS x86_64/arm64, Ubuntu x86_64/aarch64, Windows) into the obs-rust/cef mirror, with hashes.
  3. Bump the pin and fix any obs-browser API breaks (plugins/obs-browser, which is upstream code, so keep the patches minimal and list them).
  4. Smoke-test on each platform: a browser source renders a local page, a dock loads, obs-browser-page starts.

Testing without adding CI time. The existing builds with -DENABLE_BROWSER=ON already compile it. Add a version guard, a test in an existing binary that fails if the linked cef_version.h Chromium major is below a floor kept in one place, so an EOL pin can't sit unnoticed again. Render checks run in the release/app-build lanes that already exist, not on PRs.

Acceptance. The pin is on a supported Chromium, there are mirror archives with hashes for all platforms, obs-browser builds and renders on each, the version-floor guard exists, and there's a features.md line.

Part of #1092. Blocks the browser source in the AppImage (#1720).

Activity

  1. zackees commented on Oct 23, 2026

    @zackees
    MemberAuthor

    Findings before any CEF work starts:

    • Upstream OBS is still on 7871 too (obsproject/obs-studio CMakePresets.json, updated 2026-09-18 "Update CEF from 127/6533 to 150/7871"). So there's no newer upstream archive for the obs-rust/cef mirror to copy. That mirror holds byte-for-byte copies of the archives upstream OBS fetches from its CDN; it builds nothing.
    • Official CEF automated builds (cef-builds.spotifycdn.com) have a current stable for every platform we pin: 156.0.7+chromium-156.0.8078.17 for linux64, linuxarm64, macosarm64, macosx64 and windowsarm64, and 152.0.15+chromium-152.0.7977.169 for windows64. Each comes as minimal/standard distributions.
    • The catch: upstream OBS builds its own CEF with proprietary_codecs (H.264/AAC in pages). The official builds don't include them, so a Browser Source playing an H.264 MP4 alert or overlay video would stop playing it. VP8/VP9/AV1/Opus/WebM still work.

    Decision needed (maintainer):

    • A. Switch to the official CEF builds now (156, Windows x64 on 152), building libcef_dll_wrapper ourselves in the deps step. That's fast and security-current, but drops H.264/AAC in browser sources. Notice + features.md line.
    • B. Build CEF ourselves with proprietary codecs (Chromium source build per platform, hours each on large runners, plus a pipeline to keep it current). That keeps parity with upstream but is a lot of infrastructure and runner time.
    • C. A now, B later only if users report codec breakage.

    My recommendation is C: the CVEs are exploited in the wild and the sandbox is off, so getting off 150 matters more than H.264 inside pages. Nothing else in #1864 depends on this, but the archive source does.

  2. zackees commented on Oct 23, 2026

    @zackees
    MemberAuthor

    Decision (maintainer): B, build our own CEF with proprietary codecs, matching upstream OBS's H.264/AAC support.

    Claiming: #1864, Linux x86_64 first (AppImage, .deb; unblocks #1865).

    1. A CEF build recipe in obs-rust/obs-studio-libs: CEF branch 8078 (Chromium 156) via CEF's automate-git.py, --minimal-distrib --no-debug-build, GN_DEFINES="is_official_build=true proprietary_codecs=true ffmpeg_branding=Chrome use_sysroot=true", plus a prebuilt libcef_dll_wrapper like upstream's archive layout. Built locally under bosn in resumable steps (sync, then build; bosn caps a run at 4 h), not on CI runners.
    2. Publish to the obs-rust/cef mirror as release 8078 (cef_binary_8078_linux_x86_64.tar.xz + sha256).
    3. obs-studio PR: bump the cef pin for Linux, fix any obs-browser API breaks between 7871 and 8078, and add the version-floor guard.

    Not doable from this host: macOS (arm64/x86_64) and Windows (x64/arm64) CEF builds need those OSes and ~150 GB of disk per build, which GitHub-hosted runners don't have. Those platforms stay on 7871 in this pass, and I'll leave them as Decision needed: build hosts (self-hosted Mac/Windows builders or larger runners) rather than guess.

    Host budget: 16 cores, ~190 GB free. One Linux build is expected to take most of a day.

  3. vpacher commented on Oct 23, 2026

    @vpacher
    Member

    Some checks on the B recipe and on the findings comment, and how this ties to the Linux sandbox work in #1092 / obs-rust/obs-browser#3.

    Recipe (checked against CEF branch 8078, tools/automate/automate-git.py and tools/gn_args.py)

    1. PGO profiles: with is_official_build=true on Linux, Chromium sets chrome_pgo_phase = 2 (build/config/compiler/pgo/pgo.gni at 156.0.8078.17), so the build needs a profile. automate-git.py downloads it only with --with-pgo-profiles (default off). Add that flag, or set chrome_pgo_phase=0 in GN_DEFINES. Without one of them, I expect the build to stop when it does not find the profile.
    2. Pin the CEF commit: without --checkout, automate-git.py takes the newest commit of branch 8078. The official 156.0.7 build is 156.0.7+g47dbc1d+chromium-156.0.8078.17. With --checkout=47dbc1d, the archive is reproducible and matches a released CEF version (we can record it like our 7871 README: CEF version + Chromium version).
    3. Disk and time: --no-chromium-history (or --chromium-archive with --chromium-archive-sha256) makes the Chromium checkout much smaller than a full clone. That can matter with ~190 GB free and a 4 h cap per run.
    4. The flags match what our 7871 archive has: its libcef.so contains the FFmpeg H.264 and AAC decoders. So proprietary_codecs=true ffmpeg_branding=Chrome gives parity with upstream OBS.

    Correction to the findings comment

    • Windows x64 also has the official 156.0.7 build (minimal, standard, signed). 152.0.15 only looks newer in index.json because it was uploaded later on 10-22. All platforms have 156.0.7, the first build of the 156 branch. 152 is still updated (152.0.15 on 10-21/22).

    macOS and Windows in this pass

    They stay on 7871, which has no new CEF build since 150.0.22 on 10-09 (our archive is 150.0.19). Until there are Mac/Windows build hosts, there is one more choice for the "Decision needed: build hosts" item: the official 156.0.7 builds on those platforms (current Chromium, no H.264/AAC in pages), or 7871 with codecs and without security fixes.

    Ties to #1092 (Linux sandbox, mine)

  4. zackees commented on Oct 23, 2026

    @zackees
    MemberAuthor

    Linux x86_64 done: #1948 merged (CEF 8078 / Chromium 156 with proprietary codecs, from obs-rust/cef release 8078, recipe obs-rust/obs-studio-libs#11). obs-browser compiles unchanged against 8078. Still open here: macOS and Windows (need build hosts — decision pending), and linux aarch64. Runtime check of a browser source comes with #1865.

  5. zackees commented on Oct 25, 2026

    @zackees
    MemberAuthor

    Triage. Status: Linux x86_64 is done (#1948, CEF 8078 / Chromium 156). Left: linux aarch64, plus macOS and Windows, which wait on the maintainer's pending decision about build hosts for the obs-rust/cef archives. P1: Windows/macOS still ship an end-of-life Chromium in the browser source. size:L.
    /ready for linux aarch64, which needs no new decision. The macOS/Windows part stays with the maintainer's build-host decision noted above.

  6. zackees commented on Oct 25, 2026

    @zackees
    MemberAuthor

    /ready

  7. zackees commented on Oct 25, 2026

    @zackees
    MemberAuthor

    Decided (build hosts for the macOS and Windows CEF archives):

    • Proprietary codecs stay a requirement. No switch to the codec-less official CEF builds: H.264/AAC in browser sources (alerts and overlays) would regress on Windows and macOS.
    • No CI runners. A Chromium build doesn't fit GitHub's 6-hour job limit on standard runners, and larger runners are paid. The archives are built like the Linux x86_64 one: from the recipes/cef/ recipe in obs-rust/obs-studio-libs, under bosn on a project machine, with the recipe and hashes recorded in the obs-rust/cef release.
    • Windows (x64, arm64): cross-compiled from Linux on the same build host, using Chromium's supported Linux→Windows cross build (packaged Windows SDK toolchain). A recipes/cef/windows recipe in obs-studio-libs, with the same codec checks as the Linux one.
    • macOS (arm64, x86_64): on a Mac. Chromium has no Linux→macOS cross build. One Apple Silicon Mac builds both architectures. Until one is available, macOS stays on 7871, which is also upstream OBS's current pin, so it's no worse than upstream.
    • Linux aarch64: the Linux recipe with target_cpu=arm64 (already Ready).

    Next: separate child issues for the Windows recipe and the macOS build, each sized, with the existing acceptance criteria (archive and hash in obs-rust/cef, obs-browser builds, a browser source renders, the version-floor guard).

    (Decided by the coordinator under the maintainer's go-ahead of 2026-10-25. The maintainer can reverse it.)

  8. zackees commented on Oct 27, 2026

    @zackees
    MemberAuthor

    Claiming: Linux aarch64 CEF 8078 slice only. Plan: (1) check the existing Linux recipe and archive consumers; (2) add a failing arm64 recipe or version test, then build and publish the codec-enabled arm64 archive with its SHA-256; (3) update the obs-studio pin and validate the browser build and runtime on arm64. First PR expected after the archive and focused test are ready. Windows and macOS remain separate.

  9. zackees commented on Oct 27, 2026

    @zackees
    MemberAuthor

    Releasing the Linux aarch64 CEF slice. The workspace automatic write review rejected the first test edit in the obs-studio-libs sister checkout as a protected external write. No code or archive was changed or pushed. A worker with write access to that checkout can continue from the existing CEF Linux recipe; the claim is free.

  10. zackees commented on Oct 27, 2026

    @zackees
    MemberAuthor

    Provenance caveat for the macOS 8078 archives, recorded here because this issue holds the build-host decision: the two macOS archives are now on the obs-rust/cef 8078 release, but they were built on a contributor's own Mac and are verified only against hashes that contributor reported, so they carry no reproducible build and no attested run. They are pinned only at the builder's own risk until identity validation completes (see #2418) and the recipe is built on a runner obs-rust controls.

    The build-host decision recorded above also assumed no CI runners, which is right for GitHub-hosted macOS (6 h job cap, 14 GB SSD against a recipe that needs 70 GiB free) but not for a self-hosted one (5-day job cap, persistent disk). That is the change worth making: a self-hosted Mac runner executing recipes/cef/macos from a workflow in obs-studio-libs, publishing the archive with a build attestation, so the hash consumers pin has a verifiable root rather than a person. That work is the durable answer to the provenance gap, not another manual rebuild.

  11. zackees commented on Oct 27, 2026

    @zackees
    MemberAuthor

    The source-record half of the provenance gap now has its own issue: #2674 (mirror releases identify no source; tags point at an unrelated commit). It covers the three existing releases — manifest, honest tags, the source artefact, and whether the fork's OBS patches are in the 8078 archives — and is independent of the self-hosted-runner/attestation work described above.

  12. zackees commented on Oct 30, 2026

    @zackees
    MemberAuthor

    Status after the macOS archives landed:

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

    P1Next upreadyMeets the definition of Ready; claimablesize:LSeveral days

    Type

    No type

    Projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions