Repository navigation
Conversation
Automated AI review
Verified: the new The main gap: Wakes that bypass
|
Automated AI review
Follow-up on head Verified: the new commit fixes the per-event re-arm, re-arms the reader and the blank timer on screen changes, and the description now documents the healthy-reader trade-off. It also breaks an existing repository test (new finding below). The wake route through locked key binds is still open.
Verified: the opening points of the previous review still hold at this head: New:
|
|
Reviewed by Claude Opus 5.5, with a second opinion from Codex Medium. I pushed one fix, and the head is now Pushed: What ran, on a worker built from the Omarchy ISO and linked onto the branch:
That shows the logic, not the hardware. No fingerprint reader was run, so nothing here confirms the thermal behaviour on a CV3, Goodix or Elan reader. The PR's author can't give that confirmation either; it has to come from someone else seeing the overheating, against Second opinion: Codex Medium found nothing further at
Still open, for the maintainer:
Related: several open pull requests fix the same overheating differently: #11918, #11273, #7179, #12461, #12544, #13819, #10653 and #10393. This one stops the reader from being armed while the display is dark. #11918, which is already marked ready, backs off retries that fail instantly, which covers the awake lock this one leaves alone. They look complementary rather than alternatives. Which to take, and whether to accept the healthy-reader trade-off, is the maintainer's decision, so this is waiting on them. |
|
|
I took Nothing else has changed since my last comment. The review, the worker results and the open points listed there still stand at |
|
Merged #7158 covers the immediate-error backoff alternative mentioned here. Pausing new scans while the display is blank remains unmerged: the landed code keeps healthy readers listening through normal scan timeouts. Keeping this open for that distinct arming-policy change. GPT-6 in Codex and Claude Opus 5.5 Medium checked the coverage against the landed source; Claude's agreement does not guarantee independence. This closure review was read-only and did not rerun runtime tests. |
A scan window that ends without a finger counts as a usable attempt, so with the panel blanked the reader re-arms every 250 ms for as long as the screen is off, self-heating thermal-limited readers into a cutoff. Stop the retry loop in runBlank() and refuse new scans in startFingerprint() while displaysBlank is set. Re-arm on every route the panel can come back: runWake() and onScreensChanged() already clear the flag, and a resume that keeps the same outputs (and so never reaches runWake()) now clears it from restartFingerprintAfterSleep(). A falling displaysBlank re-arms scanning from the change handler rather than on every input to an awake lock.
de40367 to
d82f5c1
Compare
| // restartFingerprintAfterSleep() clears it for the resume route, which lights | ||
| // the panel without either. | ||
| onDisplaysBlankChanged: { | ||
| if (!displaysBlank && lockRequested && fingerprintConfigured && !authenticatingPassword) startFingerprint() |
There was a problem hiding this comment.
Screen changes restart dark scans
If the screen set changes while the lock is blanked and the remaining panels are still dark, onScreensChanged() clears displaysBlank, and this handler starts fingerprint scanning. The blank timer has already fired and is not restarted on this route, so retries can continue on the dark lock instead of stopping after a fresh blank interval.
| // keypress. Defer the clear so this function has settled any orphaned | ||
| // attempt first; the falling flag re-arms scanning through the | ||
| // displaysBlank change handler (runBlank stopped the retry timer). | ||
| Qt.callLater(function() { root.displaysBlank = false }) |
There was a problem hiding this comment.
When a blanked lock resumes with the same outputs, this deferred clear restarts fingerprint scanning without restarting the blank timer. That single-shot timer was consumed when the lock blanked, so if the display stays awake without further input, scans continue indefinitely.
|
Addressed the resume gap (from the omarchybot read on #12202): read against current Rebased onto current |
|
Pausing new fingerprint scans while Same feature as CONFLICTING #11273 — prefer this MERGEABLE head. Related cluster: lid-closed gate #14280/#14251, retry-storm reports #13944/#13748/#12461. #7158's merged backoff is complementary, not a substitute for the blank gate. |
|
Since this head ( Read against
There is one literal difference. Here, any screen change that clears the blank re-arms the reader. On This was checked by reading the source only. Nothing ran on a worker, because there is nothing left to verify on this branch. Codex Medium reviewed the same question separately and reached the same answer with the same line references. I checked those references against The classification is still unsettled. Claude Opus 5.5 reads this as a compatibility bug, because it targets readers that thermally self-disable. Codex Medium reads it as an enhancement, because it changes arming for every reader on purpose. So this pull request carries no class label. This now waits on the maintainer. Since |
|
Superseded by #15294 ( Closing in favour of #15294. Thanks to omarchybot and jandrusk for the reads that pinned this down. |
What and why
While a display is blanked, the lock screen keeps the fingerprint reader running scan cycles against an empty pad. On fingerprint readers that thermally self-disable (Broadcom CV3 /
0a5c:5843, Goodix27c6:533c, Elan04f3:0c4b, alllibfprint-tod), that cycles the driver intoDevice disabled to prevent overheatingrefusals: 31,534PAM_AUTHINFO_UNAVAIL/ code 9 returns in under four hours on a CV3, in a 205 s-refusing / 122 s-available cycle. Each refusal re-arms the 250 ms retry timer, so the reader stays in the storm.This PR stops new scans while the display is blanked and re-arms on wake. With it, a CV3 logged 0 refusals across a 15-minute lock and read on the first touch after wake (measurement thread).
Changes (gating, vs
quattro)runBlank(): stopfingerprintRetryTimerso a blanked display is no longer re-arming scan cyclesstartFingerprint(): return early whiledisplaysBlank(single call site; every trigger goes through it)onDisplaysBlankChanged: re-arm a scan only on the falling edge ofdisplaysBlank, so fingerprints do not restart on every pointer move or key press on an already-awake lockrestartFingerprintAfterSleep(): cleardisplaysBlankon resume, so a lid-open that keeps the same outputs (a resume that never reachesrunWake()) re-arms the reader instead of leaving it off until a key pressdisplaysBlankis already cleared byrunWake()andonScreensChanged()on the base, so the change handler turns either of those routes into a re-arm; the resume clear covers the one route that lights the panel without going throughrunWake().No
abort(): an in-flight scan is allowed to finish; on readers wherepower/wakeupis disabled, the reader cannot wake the compositor, so aborting on blank can strand it with nothing to re-arm it.Trade-off on healthy readers (documented)
After a lock blanks, a reader whose session is still within its verify timeout keeps working; once the in-flight session ends, no new scan starts until a wake event (pointer, unbound key press, password, panel change). On readers that never overheat this means touch-to-unlock from a fully dark lock screen stops a short time after blank where the base kept re-arming. This is the cost of the gating; the alternatives on the table for follow-up:
PamErrorresults (the refusal signature), so healthy readers keep the base behavior while refusing readers still stop cyclinghyprctl monitors -jfor video wallpapers)Both are intentionally left out of this PR to keep the validated core small; happy to take either as a follow-up.
Verified by stock A/B capture (2026-09-25)
Swapped the running lock plugin to a pristine stock build and held a blank-idle lock for 25 minutes: stock arms the reader continuously through the blank period, producing repeated
Device disabled to prevent overheatingstorms, a realfprintdSIGSEGV in its libusb/gusb thread (auto-restarted by systemd), and a re-arm thrash where every 250 ms attempt errors instantly — which reads as a "dead reader" to the user. The gating mechanism prevents exactly this: the reader is not armed while the display is blank, so the overnight self-heating cycle that trips the thermal gate never starts. Full details: /issue/12202#issuecomment-5841664212.Notes
quattro; behavior changes only while the display is blanked (see trade-off above). A session already running when the lock blanks is allowed to finish; only new scans pause.runWake(),onScreensChanged()and the sleep/resume hookrestartFingerprintAfterSleep().test/shell.d/lock-fingerprint-blank-gating-test.shpins these.