Re: Pretest next week

[Adrian Robert]

On Jan 28, 2009, at 7:40 PM, Stefan Monnier wrote:

> > As it turns out, this isn't needed.  SIGALRM is triggering
> > alarm_signal_handler() in atimer.c correctly, but run_timers()  
> > never  gets
> > called under SYNC_INPUT.  I don't fully understand why or where   
> > the problem
> > is here.  But undefining SYNC_INPUT gets poll_for_input()  called  
> > (and hence
> > Ctrl-g detected) even in tight loops, with no ill  effects so far.
>
> I wrote SYNC_INPUT specifically for X11 input handling.  From what you
> say above, there is apparently a bug in the way it handles SIGALRM.
>
> Apparently the QUIT macro should check pending_atimers somehow.

Yes, adding

    else if (pending_atimers)                           \
      do_pending_atimers();                             \

at the end of QUIT allows poll_timer() to fire under SYNC_INPUT and  
Ctrl-g to be detected, with no apparent other ill effects (in very  
limited testing).


>   If so, we should probably create a new
> var `pending_signals', which should always reflect
> "pending_timers || interrupt_input_pending"

I'm not sure if the extra 0-comparison would significantly add to  
overhead but I guess code size could take a hit.  Though maybe the  
part of QUIT above (when there is a quit_flag) could also be reduced  
to a function call to slim things down?