bug#50043: 28.0.50; USABLE_SIGOI undef code paths do not work correctly

[Eli Zaretskii]

> Date: Mon, 15 Nov 2021 14:26:08 -0500
> Cc: larsi@gnus.org, 50043@debbugs.gnu.org
> From: Ken Brown <kbrown@cornell.edu>
> 
> >> I think wait_reading_process_output gets stuck for 2 seconds in a call to 
> >> select
> >> (actually xg_select because I'm testing a gtk build).  This is independent 
> >> of
> >> the fact that x-selection-timeout is 2 seconds; it happens even if
> >> x-selection-timeout is 0.  select returns after 2 seconds because the 
> >> poll_timer
> >> fires.
> > 
> > Sorry, I don't understand: select waits for up to 2 seconds because
> > that's what we ask it to do, and those 2 sec do come from
> > x-selection-timeout.  If x-selection-timeout is zero, select is not
> > supposed to wait at all, so why does it?  What am I missing?
> 
> Setting x-selection-timeout to zero actually makes the timeout infinite:

Ah, I thought you meant a literally zero timeout, not the "zero means
infinite" one.

> > Anyway, AFAIU, the wait is supposed to end because XTread_socket reads
> > a SelectionNotify event, and that modifies the cell for which we
> > wait.  What I'm not sure I understand is how are we supposed to call
> > XTread_socket when we are stuck inside select all the time?
> 
> We're never stuck for more than 2 seconds [when there's no SIGIO] because 
> poll_timer fires and either sends SIGALRM or makes timerfd read ready.  
> Either 
> way, select returns, and the next iteration of the main loop checks for input 
> and checks for a cell change.
> 
> >> We certainly don't want to always skip the select call, but would it make 
> >> sense
> >> to use a very short timeout for select in that case?  Or maybe someone has 
> >> a
> >> better idea.
> > 
> > Making timeout shorter might be the solution, but I'd like to
> > understand the problem better first.

If the code is based on the premise that we check for selection when
we exit select, then I think on systems without USABLE_SIGIO we should
call wait_reading_process_output with a short timeout but in a loop,
so that the summary wait is still 2 sec, but we exit the loop as soon
as selection arrives because each call to wait_reading_process_output
has a much shorter timeout, say, 25 msec.  WDYT?