Repository navigation
Conversation
69fd511 to
a3598a0
Compare
|
a3598a0 to
56d3feb
Compare
|
Appreciate the hardware testing notes (M1 Touch ID / fprintd). Two triage notes:
Leaving as a product/UX call pending maintainer taste on the fade behavior. |
Add a shared FingerprintReader that watches fprintd's D-Bus signals, so the lock shows when the reader is waiting, reading a finger, or rejected it: a status pill, an accent glow and border sweep while reading, and a red shake with a cross on a miss. Display only; PAM still decides the unlock. A rejection counts only after the reader reported a finger on the sensor: the Apple SEP driver answers pam_fprintd's 30 s verify timeout with verify-no-match, which would otherwise flash a blanked lock.
Reuse FingerprintReader so the fingerprint card says when the reader is waiting or reading a finger, swaps to a check or cross for the verdict, and shakes with a red border on a miss. A rejection that lands after pam_fprintd falls through to the password is still reported under the field. Display only; PAM still decides the authorization.
0ee6d53 to
d2247f1
Compare
|
Thanks @jandrusk, both points taken.
|
After pam_fprintd falls through to the password, another flow may use the reader, and its match made the password prompt say Authorized. Fingerprint mode is when this request's pam_fprintd holds the reader, so every signal is ours.
pam_fprintd falls through to the password as soon as it gets the final VerifyStatus, and the monitor reads the same signal through a pipe, so the last rejection can arrive after fingerprint mode ends. Stopping the monitor then dropped it, and the dialog never said the finger wasn't recognized. A finger landing while this request holds the reader now keeps the monitor listening until that read's verdict, for at most 500 ms after the reader is let go. A new request or a closed dialog ends the wait.
|
Tested
No QML errors after the plugin reload. |
|
The enrolled-entry matcher is incorporated in merged #7158. Passive reader feedback, animations and polkit status remain separate UI work. Keeping this open; the landed error recovery is not a substitute for those features. 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. |
Summary
The lock screen and the polkit dialog show a static fingerprint glyph. You can't tell whether the reader is waiting, has seen your finger, or has rejected it. This PR adds live reader feedback to both.
pam_fprintd only reports the final verdict, so a new shared
FingerprintReader(shell/Ui/FingerprintReader.qml) watches fprintd's D-Bus signals (finger-needed,finger-present,VerifyStatus) with a passivegdbus monitor. It is display-only: authentication is still decided by each surface's PAM context.Scope: this is UI feedback only. It does not change the fingerprint retry behaviour, and it does not fix the 250 ms retry loop or the thermal-guard lockout in #13944, #13906 or #9905.
The fade-over-desktop part that used to be in this PR is now split out into #13962.
1. Lock screen
accent/urgentcolors instead of the lock tokens. Some themes set every lock token to one color (Tokyo Night setslock.text-errorto the foreground), so with the lock tokens a rejected read would look identical to a scan.verify-no-match, which would otherwise flash a rejection on an idle lock.#N:lines. The old check matched any "finger", which includes "no fingers enrolled".2. Polkit dialog
The square fingerprint card gets the same states, with a status pill under the card styled like the action label above it:
When pam_fprintd runs out of tries it falls through to the password. If the last read was a rejection, the pill says "Fingerprint not recognized, enter password" above the password field.
This builds on the ideas in #11162 (flash on a rejected read) and #11952 (fingerprint status text in the lock and polkit dialogs, and the fprintd-list check). Merging this would overlap with both.
Testing
pkexec truewith this commit'sPolkitAgent.qmlandFingerprintReader.qmldropped into acb.polkitclone. Screenshots confirmed the waiting, identifying and rejected states, including the return to waiting afterwards. The match check was confirmed by eye.quattro: all files compile in Quickshell, and the lock, polkit, sleep and runtime-smoke suites undertest/shell.dpass at each commit.fingerprint-feedback-test.shruns the reader's D-Bus line parser againstgdbus monitoroutput captured from real hardware, and checks the lock and polkit wiring. Thelock-fingerprint-indicatorfixture covers the lock's visual states.