This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Spurious testsuite failures
On Fri, Apr 30, 2004 at 09:54:20AM -0700, Dan Kegel wrote:
> Dan Kegel wrote:
> >>This turned out to be Sun BUG 4927647. Which is triggered by expect.
> >>No patch availiable as of today :-(
> >
> >I beg to differ. Try building expect from sources yourself
> >and applying the patch
Sorry for beeing terse in my previous mail. I did try the patches and built
tcl, expect (with your patch thanks for the pointer btw.) and dejagnu from
source and that didn't fix it.
> > http://kegel.com/crosstool/current/patches/expect-5.39/pr12096.patch
> >That fixes it for me.
>
> ... on Linux, anyway. Did you try the patch I mention,
> and it still didn't work on Solaris?
Yes exactly. This is not PR12096, it is a different bug in the solaris
kernel. The following program sometimes exits with exitcode 1 without
reading anything from the pty (read returns EINVAL):
int main ()
{
int i;
char * name;
char buf[100];
int fd, master = open ("/dev/ptmx", O_RDWR);
grantpt (master);
unlockpt (master);
name = ptsname (master);
if (fork ()) {
wait (0);
errno = 0;
while (read (master, buf, 1) > 0);
if (errno == EINVAL)
return 1;
return 0;
}
fd = open (name, O_RDWR);
ioctl (fd, I_PUSH, "ptem");
ioctl (fd, I_PUSH, "ldterm");
write (fd, "blahblub", 8);
if (fork ())
wait (0);
close (fd);
return 0;
}
This is similar to what expect does but there's probably not much
that expect can do to work around this problem.
> Can you give a link to the description for that solaris bug?
Unfortunately not. You'll need a sunsolve account, then just search for
the bug number. However, this needs Solaris 9 (maybe 10) and a multi
processor machine to trigger.
regards Christian
--
THAT'S ALL FOLKS!