fix: no longer input locks when TTY is switched before full compositor

[Atan-D-RP4]

This fixes the case of starting Niri from a separate TTY (via Tmux or a TTY switch before full startup) and then switching to that Niri session causing an input lock.

I hack on Niri mostly in separate TTY and this behavior was a super annoying blocker. Resolving it required either restarting my computer (since all input was blocked) or sshing into my machine and killing that Niri session.

[YaLTeR]

Do other compositors do something similar?

[Atan-D-RP4]

I can only confirm that Hyprland has the same problem, from my own experience and testing for this.

From what I can tell, both Mutter and Kwin use systemd-logind to manage inactive VT and defer input processing until it sends active session signal to them.

[YaLTeR]

From what I can tell, both Mutter and Kwin use systemd-logind to manage inactive VT and defer input processing until it sends active session signal to them.

That's also what niri does

[YaLTeR]

Maybe it doesn't do so at startup? Does mutter also call libinput suspend at startup?

[Atan-D-RP4]

Maybe it doesn't do so at startup? Does mutter also call libinput suspend at startup?

I checked and no it does not. Which might be why it also has this same issue, from my testing.
I can't even fully switch to Kwin's session before my entire screen is locked, with only a frozen TTY visible to me.

[YaLTeR]

Huh. I want to be careful here then if this is a bit of an uncharted ground. Perhaps a better solution is, when the session is inactive at startup, call on_session_event() with session suspended, and let it handle libinput suspend and others?

[Atan-D-RP4]

I want to be careful here then if this is a bit of an uncharted ground.

Seems to be the case, though I don't know why. Do devs not use separate TTY during development? I ran into this right away when I started hacking on Niri.

Anyway, now using event_loop.insert_idle to insert PauseSession event.

[YaLTeR]

Any reason for using an idle rather than calling it in init()?

[Atan-D-RP4]

I didn't know better. I've played around with the insert_idle for something else, saw it was there reached for it.
Doing it in Tty::init() is actually much cleaner.