Re: [PATCH] getdtablesize: Fix returning hard instead of soft rlimit

[Rich Felker <dalias@libc.org>]

On Sat, Aug 13, 2016 at 09:36:13PM +0200, Olivier Brunel wrote:
> On Sat, 13 Aug 2016 15:25:02 -0400
> Rich Felker <dalias@libc.org> wrote:
> 
> > On Sat, Aug 13, 2016 at 08:31:49PM +0200, Olivier Brunel wrote:
> > > ---
> > >  src/legacy/getdtablesize.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/src/legacy/getdtablesize.c b/src/legacy/getdtablesize.c
> > > index 682da6d..b30c193 100644
> > > --- a/src/legacy/getdtablesize.c
> > > +++ b/src/legacy/getdtablesize.c
> > > @@ -7,5 +7,5 @@ int getdtablesize(void)
> > >  {
> > >  	struct rlimit rl;
> > >  	getrlimit(RLIMIT_NOFILE, &rl);
> > > -	return rl.rlim_max < INT_MAX ? rl.rlim_max : INT_MAX;
> > > +	return rl.rlim_cur < INT_MAX ? rl.rlim_cur : INT_MAX;  
> > 
> > Is there a motivation for this?
> > 
> > Rich
> 
> Well, I found this running tests for findutils, and having a couple
> failing because of it. I'm not sure about possible
> implications/issues it could cause in the actual tools, was only trying
> to get the tests to pass (and things to work as expected), but I
> wasn't facing an "actual" issue/bug, if that was the question.

Given that the sysconf analog of this function uses rlim_cur rather
than rlim_max, it's probably better for consistency if nothing else.

It's possible we should consider dropping the implementation of
getdtablesize and just having it call sysconf. This would somewhat
enlarge static binaries calling getdtablesize, but the idea would be
to remove the motivation to use an unspecified, deprecated, legacy
function for the sake of size.

Rich