(ours is patched to not do this) leave in utmp. PR 26168.
I'm doing this by testing for ^:[0-9] in the line field of utmp(x),
rather than by attempting to stat the device name corresponding to the
entry as in Martin's patch, because (1) no valid tty should ever have
a name beginning with a colon, and (2) this way we don't silently skip
over real ttys that should be there but for some reason have disappeared.
(I suppose this might conceivably fail to catch entries for displays
connected via XDCMP; however, ~nobody does that any more and XDCMP is
a security hole anyhow. If anyone's really concerned about this, let
me know.)
In the long run we should look into ways of getting "tty" messages to
users logged in with an X session and no terminal windows open, as
that's probably fairly common for the desktop.
606b73cdb5