Re: kern/44950: fcntl(F_MAXFD) returns bogus values

[Julio Merino <jmmv%julipedia.org@localhost>]

The following reply was made to PR kern/44950; it has been noted by GNATS.

From: Julio Merino <jmmv%julipedia.org@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: martin%NetBSD.org@localhost, kern-bug-people%netbsd.org@localhost, 
 gnats-admin%netbsd.org@localhost, netbsd-bugs%netbsd.org@localhost
Subject: Re: kern/44950: fcntl(F_MAXFD) returns bogus values
Date: Tue, 10 May 2011 09:30:34 -0400

 On 5/10/11 9:20 AM, martin%NetBSD.org@localhost wrote:
 >> Description:
 >
 > The test case in /usr/tests/lib/libc/gen/t_closefrom fails when run when
 > invoked from a user with /bin/sh or /bin/ksh as shell, but works when invoked
 > from a (t)csch user.
 >
 >> How-To-Repeat:
 >
 >      cd /tmp
 >      /usr/tests/lib/libc/gen/t_closefrom closefrom_basic
 >
 > Ktrace shows that fcntl(F_MAXFD) returns 3 in the csh case, but 9 in the
 > (k)sh one.
 
 I see several possibilities here:
 
 * The test case is bogus because it's relying on the shell to leave a 
 predeterminate amount of file descriptors open.
 
 * ksh/sh are leaking file descriptors.
 
 Note that a test misbehaving when run by hand (not through atf-run) 
 should not be seen as a problem though.
 
 Another thing is that atf-run should be closing the file descriptors in 
 a consistent manner, but I can't see it doing so at the moment.  (If we 
 conclude that there is nothing else to do but adjust atf-run, assign the 
 PR to me please.)