Skip to content

test: original fixtures fail in both libobs builds under the Tier 2 image (deinterlace 2x field order, screencast resume, pulse stream restart, pulse default-follow PipeWire) #4064

Open
0 of 1 issue completed

Description

@zackees

Refs #1390

The unchanged original fixtures reproduce failures in both hybrid and C-reference libobs under the Tier 2 image. This records original-fixture failures separately from Rust port regressions. Deinterlace is already tracked in #456; its row is supporting evidence for that issue, not a new implementation claim.

Source: 3973cc9. GCC/G++ 15.2.0, PipeWire 1.6.2, Xvfb/software OpenGL, image obs-rust-linux-validate:74e06f6c010409ba (sha256:0d49edfb939933f67f208ca8396d1dbbe806f97399c590fce14356ce0b41b554). Ten original CTest invocations per lane. Cells are pass/fail/skip.

Original CTest Hybrid C-reference Evidence
test_deinterlace_2x_field_order_60i_into_60 1/9/0 2/8/0 Field count/cadence at C:493/516/522; #456
test_deinterlace_2x_field_order_50i_into_50 1/9/0 0/10/0 Same assertions
test_deinterlace_2x_field_order_50i_into_60 1/9/0 0/10/0 Same assertions
test_deinterlace_2x_field_order_50i_into_30 1/9/0 1/9/0 Same assertions
test_deinterlace_2x_field_order_60i_into_50 2/8/0 0/10/0 Same assertions
test_pipewire_screencast_resume 8/2/0 9/1/0 showing_a_source_starts_its_stream C:1261 slowest < 1000; reference also visibility C:930
test_pulse_stream_restart 7/3/0 2/8/0 failing_stream_is_paced_then_given_up C:832: 100 restarts versus 99; reference repetition 10 times out at 90 seconds
test_pulse_default_follow_pipewire 9/1/0 8/2/0 C:450 audio expectation; both lanes fail default_output_unplugged and new_default_output

The private daemon readiness checks succeeded. Hybrid screencast repetition 9 measured 1,334 ms first-frame latency. Native PulseAudio default-follow is a separate selector: the original image lacks the server and skips before any C cases; adding the server is an environment correction, not evidence that the PipeWire selector passes. Camera registry race is likewise a separate environment fix: GNU ld wrapping is lost when callers and definitions share one unity object.

Host load was high and variable; sequential lanes are not a controlled performance comparison. Shared failures establish reproduction in both implementations, not a root-cause attribution. Keep these fixtures out of new native-graph registrations until their complete original cases pass in both lanes; do not weaken the C tests or public headers.

Activity

  1. zackees commented on Nov 8, 2026

    @zackees
    MemberAuthor

    Refs #1390. Batch 43 rebuilt the Tier 2 image with the actual PulseAudio server and reran the original native selector five times per core, without retries or C test edits.

    GCC 15.2.0; image obs-rust-linux-validate:e718f7a0f723d48e, immutable ID sha256:be98ad6a3bb704a618308ac3e2f7061327b2c8d041cebbb47c934f9f42105fda; exact 163-package runtime manifest includes pulseaudio=1:17.0+dfsg1-2ubuntu4. The test and real linux-pulseaudio module were linked to their freshly built lane's libobs.

    Original native PulseAudio selector Successful invocations Case passes Case failures Skips
    Hybrid 3/5 27/30 3 0
    C-reference 0/5 16/30 14 0

    Hybrid failed default_output_switch, default_output_unplugged, and new_default_output. Across the C-reference repetitions, every one of the six case names failed at least once. All six cases executed in every invocation; no timeouts or not-run slots. These observations extend the original report's native-server setup confirmation: installing the server removes the wrapper skip, but the unchanged fixture still fails in both cores. The separate PipeWire selector remains unchanged, and the native selector is not being added to the native graph.

    The separate camera setup correction passed all four original cases in all five repetitions per lane in both CMake and the native graph (20 invocations, 80 case passes, zero skips). Both consumers preserve two unity/PCH groups, and disassembly confirms calls to all three __wrap_obs_pipewire_* hooks. This establishes the camera fixture's instrumentation correction without attributing shared PulseAudio failures to Rust.

  2. vpacher commented on Nov 13, 2026

    @vpacher
    Member

    Coordinator triage: this is a tracker for the original-fixture failures that occur in both cores. Deinterlace is #456, now a sub-issue. The screencast resume, PulseAudio stream restart and PulseAudio default-follow failures stay on this checklist until each gets its own issue.

    Added as a sub-issue of #1390.

  3. added
    trackerTracks work in sub-issues or a checklist; not claimable itself
    on Nov 13, 2026
  4. zackees commented on Nov 18, 2026

    @zackees
    MemberAuthor

    Another occurrence: test_pulse_default_follow_pipewire failed case default_output_unplugged (test_pulse_default_follow.c:450) in Tier 2 Linux tests ON on Dell runner dell-fsx3mn2-4, merge-group run https://github.com/obs-rust/obs-studio/actions/runs/44224065309 (job 134536982767), for PR #5026, which only touches hotkey enumeration. The other 973 tests passed. The PR was requeued once.

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

    trackerTracks work in sub-issues or a checklist; not claimable itself

    Type

    No type

    Projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions