csd-power: display blank timer is not re-armed after a screensaver inhibit is released, leaving the monitor powered on indefinitely

[malte-timmermann]

Summary

When an application takes an org.freedesktop.ScreenSaver inhibit while the
display is already blanked, csd-power powers the display back on. After the
inhibit is released, csd-power does not re-evaluate the current idle time,
so the display is never blanked again for the remainder of that idle period --
even though the X idle counter keeps increasing far beyond the configured
sleep-display-ac timeout.

The result is a monitor that stays on for hours after a single, ~11 second
inhibit. From the user's point of view the "Turn off the screen" setting looks
completely broken, while it actually works correctly the first time.

Environment

• Linux Mint 22.2 (Zara), Cinnamon 6.4.14
• cinnamon-settings-daemon 6.4.3+xia
• X11 session (not Wayland)
• NVIDIA proprietary driver 580.173.02, external DisplayPort monitor,
  laptop lid closed
• org.cinnamon.settings-daemon.plugins.power sleep-display-ac = 1800
• org.cinnamon.desktop.session idle-delay = 600

Observed behaviour

Logged with a script polling XScreenSaverQueryInfo and xset q every 5 s:

19:57:53  idle = 30 min   Monitor=On
19:58:03  MONITOR: On -> OFF   (idle = 30.2 min)     <- correct
20:27:49  idle = 60 min   Monitor=OFF
23:46:36  MONITOR: OFF -> On   (idle = 258.8 min)    <- spontaneous

The wake-up at 23:46:36 happened with no user input at all -- the X idle
counter was at 258.8 minutes and kept increasing across the event.

A parallel dbus-monitor --session "interface='org.freedesktop.ScreenSaver'"
shows the cause, two seconds earlier:

23:46:34  method call  interface=org.freedesktop.ScreenSaver; member=Inhibit
             string "thunderbird-esr"
             string "Playing audio"
23:46:44  method call  interface=org.freedesktop.ScreenSaver; member=UnInhibit

After the UnInhibit the display was never blanked again. It was still on the
next morning.

Minimal reproduction

1. Set "Turn off the screen when inactive for" to 1 minute.

2. Leave the machine untouched until the display powers off.

3. Without generating any input, from another machine over SSH (or a script
   started beforehand) call:

   dbus-send --session --print-reply \
     --dest=org.freedesktop.ScreenSaver /ScreenSaver \
     org.freedesktop.ScreenSaver.Inhibit string:'test' string:'test'
   

   -> the display powers back on immediately.

4. Release it again with the returned cookie:

   dbus-send --session --dest=org.freedesktop.ScreenSaver /ScreenSaver \
     org.freedesktop.ScreenSaver.UnInhibit uint32:<cookie>
   

5. Keep the machine idle. The display never powers off again, although
   XScreenSaverQueryInfo reports an idle time far above the configured
   timeout.

I verified step 3 locally: with the monitor off and no input, the idle counter
read 192 s, 196 s and 199 s across the Inhibit call while xset q flipped from
Monitor is Off to Monitor is On.

Expected behaviour

When the last inhibitor is released, csd-power should re-evaluate the current
idle time and blank the display immediately if it already exceeds
sleep-display-ac, instead of waiting for an idle alarm that has already fired
and will not fire again until the user generates input.

Additional note

Arguably a screensaver inhibit whose stated reason is audio playback
("Playing audio") should not power a display back on in the first place --
it should at most prevent it from turning off. But the core problem here is the
missing re-arm: without it, any single short inhibit permanently defeats the
display timeout for the rest of the idle period.