You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Lock screen password field fills with stray keystrokes (lock key auto-repeat / wake keys) #12398
When the screen is locked (Ctrl+Super+L) and I return to unlock it, the password field already contains a row of masked characters (● dots) that I never typed while looking at the screen.
What was expected
The password field should be empty when I start typing my password.
Steps to reproduce
Press Ctrl+Super+L to lock the screen.
Hold the lock combo slightly longer than a single press (or wait until the display blanks, then press a few keys to wake it).
Wake/unlock the screen.
The password field shows several masked characters that came from those keystrokes.
Why this happens (from reading the plugin source)
shell/plugins/lock/Service.qml sets inputEnabled: root.lockRequested immediately, and LockView.qml focuses the TextInput right away (Component.onCompleted / onInputEnabledChanged → forcePasswordFocus). There is no debounce/grace period after locking, and no handling for keystrokes that merely wake the blanked display, so:
the auto-repeat of the held lock-combo keys lands in the field, and
any keys pressed to wake the blanked screen land in the field.
Suggested fix
Ignore input for a short grace period (~300–500 ms) after the lock surface gains focus.
When the display is blanked and a key press triggers wake, consume that key event (do not insert it) and/or clear the field.
Following the merge of #7806, Repeated wake/held keys are suppressed by the merged change, but original wake keystrokes still reach the password field. This issue also asks for wake-only input and post-lock input handling, so it remains open for that remainder.
Scope checked by GPT-6 in Codex and Codex Medium against the discussion and landed change; review independence is not guaranteed. No tests were rerun for this cleanup.
I went through the lock screen's key handling on current quattro (defbf90) and Hyprland 0.56.2's keyboard path to see which of the paths you describe still put characters in the field.
The row of dots is fixed on quattro, not yet released.#7806 (merged 4 October, after 4.0.4) makes the password field drop every auto-repeated key except Backspace and Delete (shell/plugins/lock/LockView.qml:69 and :212). That covers both a held key and the flood that happens when waking the panel stalls the compositor long enough for client-side repeat to kick in before the key's release arrives. Holding Ctrl+Super+L can no longer type anything either: the first L press is the lock binding and beginLock clears the password, and every repeat after that is dropped. It will reach you with the next release.
A single key pressed to wake a blanked screen still types a character. Hyprland hands the key to the focused surface first and only then turns DPMS back on (InputManager::setupKeyboard, onKeyboardKey before the key_press_enables_dpms check), and Omarchy turns that option on in default/hypr/input.lua:62. The lock keeps input enabled while blanked (Service.qml:494) and its key handler wakes the display without consuming an ordinary key (LockView.qml:210), so a printable wake key lands in the field as one dot, and so does each separate key pressed while the panel is still coming back. Shift, Ctrl or Enter on an empty field leave nothing. I did not reproduce this on a worker; the mechanism is in the source on both sides.
What to do about that remainder is a behaviour choice rather than a contained bug fix: swallowing the wake key keeps stray characters out, but it also eats the first character for anyone who types their password straight into a dark screen, and runWake clears the blank flag the moment a key arrives, so "wait until the panel is visible" has no reliable signal to hang on today. That decision is with the maintainer. In the meantime, Escape or Ctrl+U clears the field before you type.
Second opinion: Codex Medium reviewed the same code independently and reached the same conclusions on all three paths, adding the point that the blank flag is cleared before the panel is actually lit. Its independence from my reasoning is not guaranteed.
Activity
omarchybot commentedon Oct 4, 2026
Following the merge of #7806, Repeated wake/held keys are suppressed by the merged change, but original wake keystrokes still reach the password field. This issue also asks for wake-only input and post-lock input handling, so it remains open for that remainder.
Scope checked by GPT-6 in Codex and Codex Medium against the discussion and landed change; review independence is not guaranteed. No tests were rerun for this cleanup.
Omabot on behalf of DHH
omarchybot commentedon Oct 12, 2026
I went through the lock screen's key handling on current
quattro(defbf90) and Hyprland 0.56.2's keyboard path to see which of the paths you describe still put characters in the field.The row of dots is fixed on
quattro, not yet released. #7806 (merged 4 October, after 4.0.4) makes the password field drop every auto-repeated key except Backspace and Delete (shell/plugins/lock/LockView.qml:69and:212). That covers both a held key and the flood that happens when waking the panel stalls the compositor long enough for client-side repeat to kick in before the key's release arrives. Holding Ctrl+Super+L can no longer type anything either: the first L press is the lock binding andbeginLockclears the password, and every repeat after that is dropped. It will reach you with the next release.A single key pressed to wake a blanked screen still types a character. Hyprland hands the key to the focused surface first and only then turns DPMS back on (
InputManager::setupKeyboard,onKeyboardKeybefore thekey_press_enables_dpmscheck), and Omarchy turns that option on indefault/hypr/input.lua:62. The lock keeps input enabled while blanked (Service.qml:494) and its key handler wakes the display without consuming an ordinary key (LockView.qml:210), so a printable wake key lands in the field as one dot, and so does each separate key pressed while the panel is still coming back. Shift, Ctrl or Enter on an empty field leave nothing. I did not reproduce this on a worker; the mechanism is in the source on both sides.What to do about that remainder is a behaviour choice rather than a contained bug fix: swallowing the wake key keeps stray characters out, but it also eats the first character for anyone who types their password straight into a dark screen, and
runWakeclears the blank flag the moment a key arrives, so "wait until the panel is visible" has no reliable signal to hang on today. That decision is with the maintainer. In the meantime, Escape or Ctrl+U clears the field before you type.Second opinion: Codex Medium reviewed the same code independently and reached the same conclusions on all three paths, adding the point that the blank flag is cleared before the panel is actually lit. Its independence from my reasoning is not guaranteed.