fix: re-enable output on backend mode request, force frame on fresh scene output

[perlowja]

Summary

A DP/HDMI replug leaves the compositor black on CIX Sky1 (trilin-dptx-cix,
kernel 7.2.3-sky1-ncz). On this driver an unplug destroys the wlr_output, so
the replug arrives as a new one through configure_new_output(). A
freshly-created wlr_scene_output has needs_frame unset, so
lab_wlr_scene_output_commit() short-circuits on
wlr_scene_output_needs_frame() and the atomic commit that binds a CRTC to the
connector never reaches the kernel. Scheduling a frame first makes the commit
run; the idle source it registers is a no-op once the commit sets
frame_pending, so there is no double commit.

Scope

Rescoped per review to the single hardware-confirmed change: two lines in
configure_new_output(). The earlier mode-only re-enable path and the
mode-test retry timer are both gone, since neither was verified on the affected
hardware. The retry is not being filed as a separate PR yet - it stays out of
tree until it has a hardware reproduction behind it.

Test plan

• Repeated unplug/replug of DP-3 on MS-R1 (4 outputs)
• Repeated unplug/replug on Radxa Orion O6N (1 output)
• Screen recovers without a compositor restart on both

AI assistance: disclosed

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-06T00:27:00.194813Z	2426867	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 242686761a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Commit the staged enable instead of scheduling a frame

When this branch handles a MODE-only request for a disabled output, enabled=true is only staged in output->pending; scheduling a frame cannot apply it. Any resulting callback reaches handle_output_frame(), which returns while the output is disabled because output_is_usable() requires wlr_output->enabled, so lab_wlr_scene_output_commit() is never called and the connector remains disabled. This path needs to commit the staged state directly, or otherwise permit the enabling frame callback to perform that commit.

Useful? React with 👍 / 👎.

[perlowja]

Confirmed — verified against actual source (output_is_usable() at src/output.c:1240 checks the live wlr_output->enabled, not output->pending). Fixed in 122b5d4: commits the pending state directly on the success path instead of relying on the frame callback to apply it. Hardware-retested on MS-R1 after the fix, still resolves the replug black screen.

[mirkobrombin]

Please split out the hardware-tested fresh-output frame fix and remove the unverified backend mode and power-state path.

[perlowja]

Split as requested in aba03df.

Removed entirely: the MODE-only re-enable fallback in handle_output_request_state() (src/output.c) that staged enabled=true and called wlr_output_commit_state() on a backend mode request, plus the power_off bookkeeping it depended on (the field in include/output.h and the two assignments in handle_output_power_manager_set_mode()). That path was the one the PR description itself flagged as reasoned-not-reverified -- wlroots' nested x11/wayland backends appear to be the only callers of wlr_output_send_request_state() in 0.20, not drm, so it was never confirmed to actually run on this hardware.

Kept: only the wlr_output_schedule_frame() call in configure_new_output() right before lab_wlr_scene_output_commit(). That's the hardware-confirmed fix (O6N + MS-R1, repeated unplug/replug) -- it forces wlr_scene_output_needs_frame() past its early-return on a freshly-created wlr_scene_output, which is what actually gets the replugged connector re-committed.

Rebuilt clean (meson + ninja pulling wlroots-0.20 via the subproject wrap) and meson test is 130/130 green, so the split didn't break anything else.

[mirkobrombin]

The fix is now correctly scoped, but please reduce the 27-line explanation around the single schedule call to a concise cause-and-effect comment.

[perlowja]

Trimmed in 133f910.

The 27-line block around the wlr_output_schedule_frame() call in configure_new_output() is now 7 lines: a fresh wlr_scene_output has needs_frame false, so lab_wlr_scene_output_commit() short-circuits and the atomic commit that binds a CRTC to the connector never reaches the kernel — black screen on replug; scheduling a frame sets needs_frame so the commit runs. Dropped the wlroots file/line trace, the field-by-field walk of wlr_scene_output_needs_frame(), and the double-commit aside.

Did the same to the four blocks around the mode-test retry (38 lines down to 8), keeping only the cause (some DP/HDMI transmitters report connected before link training settles, so the first mode test can fail for a display that is really there), the effect (without a retry the output stays disabled until a manual wlr-randr --on), the ~2s bound, and the one-line note on the use-after-free that the teardown in handle_output_destroy() guards against.

Comment-only, no functional change. Rebuilt clean on aarch64 and the full meson test suite is green at 130/130, unchanged.

[mirkobrombin]

The new mode-test retry is explicitly unverified on the affected hardware, so please move it to a separate PR and keep this one limited to the hardware-confirmed frame scheduling fix.

[perlowja]

Rescoped in bd580a4a. The branch is now one commit on current singularity, one file, +2 / -0.

Removed, per your review: the whole mode-test retry path - handle_output_enable_retry(), the OUTPUT_ENABLE_RETRY_MAX / _DELAY_MS bound, the enable_retry_timer and enable_retry_count fields in include/output.h, the timer teardown in handle_output_destroy(), and the configure_new_output() / finish_output_configuration() split it needed. You were right that it should not ride along: I had it on the branch with a reasoned argument and no hardware reproduction, which is not the same thing as a fix.

I have not filed it as a separate PR either. It stays out of tree until I can reproduce the mode-test failure on the affected board rather than infer it - it would be the same unverified change with a different number on it.

Kept: only wlr_output_schedule_frame(wlr_output) in configure_new_output(), immediately before lab_wlr_scene_output_commit(). That is the hardware-confirmed part - a fresh wlr_scene_output has needs_frame unset, so the commit short-circuits on wlr_scene_output_needs_frame() and the atomic commit that binds a CRTC never reaches the kernel.

Also trimmed the comment on it from 7 lines to 1, per your earlier aba03dfd note. The two lines this PR adds now are:

	/* A fresh scene_output has needs_frame false; the commit binds no CRTC. */
	wlr_output_schedule_frame(wlr_output);

I also had the branch based on main rather than singularity; it is on singularity now, which is why the file count dropped to one.

Still verified the same way: repeated unplug/replug of DP-3 on MS-R1 (4 outputs) and on Radxa Orion O6N (1 output), screen recovering without a compositor restart on both.

@mirkobrombin your 133f910f review is now stale - it asked for exactly the split that produced the new head, and gh api repos/singularityos-lab/labwc/compare/133f910f...bd580a4a reports status: diverged, ahead_by: 4, behind_by: 5. I cannot dismiss it from this account (403 on the dismissal endpoint), so flagging it here instead. Ready for re-review.