Re: [PATCH.HURD] Limit the minimum timeout delay to 1us for hurdselect.c to avoid problems

[Samuel Thibault]

Svante Signell, le Thu 01 Nov 2012 18:04:35 +0100, a écrit :
> On Thu, 2012-11-01 at 17:44 +0100, Samuel Thibault wrote:
> > Svante Signell, le Thu 01 Nov 2012 17:32:53 +0100, a écrit :
> > > --- ../hurdselect_orig.c  2012-10-21 22:55:26.000000000 +0200
> > > +++ ../hurdselect_orig_timeout.c  2012-11-01 12:58:00.000000000 +0100
> > > @@ -84,9 +84,13 @@
> > >  
> > >        to = (timeout->tv_sec * 1000 +
> > >              (timeout->tv_nsec + 999999) / 1000000);
> > > -      if (strcmp(program_invocation_short_name, "vi") && 
> > > strcmp(program_invocation_short_name, "vim") && 
> > > strcmp(program_invocation_short_name, "vimdiff") && !to)
> > > - to = 1;
> > >      }
> > > +  /* XXX: A timeout of 0 returns immediately, even if no file
> > > +     descriptors are ready. This is correct according to
> > > +     POSIX.1-2001. As many programs rely on file descriptors being
> > > +     ready for a timeout of zero use 1 msec as the minimum delay */
> > > +  if (to == 0)
> > > +    to = 1;
> > 
> > Nack.  We've already tried that in the past, and and using even just 1
> > makes some applications sluggish like hell.  See the archive list for
> > the current work on actually fixing the issue properly, i.e. adding a
> > file server interface for probing.
> 
> Which applications have problems, I would like to verify that.

I don't remember, but it was posing enough problems that we had to use
0.

> For vim you have added 1us (or is it 1ms?) as a workaround.

I have added 1, just like you tried to in your patch.

      to = (timeout->tv_sec * 1000

so it's ms.

> Where in the archives to find the correct implementation ideas, both
> bug-hurd and debian-hurd has a long history?

I don't remember.  Most probably mails with "select" in the title, of
course.

Samuel