Y test failures (zpty)

Y test failures (zpty)

[Clint Adams]

Y01completion hangs for me.

+comptestinit:7> zmodload -i zsh/zpty
+comptestinit:9> comptest_zsh=zsh
+comptestinit:11> getopts z: opt
+comptestinit:12> case z (z)
+comptestinit:13> comptest_zsh=/home/clint/tmp/zsh/Test/../Src/zsh
+comptestinit:11> getopts z: opt
+comptestinit:16> ((  OPTIND > 1  ))
+comptestinit:16> shift 2
+comptestinit:18> export PS1=<PROMPT>
+comptestinit:19> zpty zsh /home/clint/tmp/zsh/Test/../Src/zsh -f +Z
+comptestinit:21> zpty -r zsh log1 *<PROMPT>*

The child zsh process shows as defunct.  Linux kernel is 2.4.17, glibc
is 2.2.5, zsh is current CVS.

As another datapoint, someone (whose complaints led me to run make check
in the first place), running Linux 2.4.19pre1, glibc 2.2.4, zsh 4.0.4,
gets the following, in addition to such errors for Y02 and Y03.

./Y01completion.ztst: starting.
comptestinit:zpty:19: can't open pseudo terminal: bad file descriptor
comptestinit:zpty:21: no such pty command: zsh
Test ./Y01completion.ztst failed: non-zero status from preparation code:
  comptestinit -z $ZTST_testdir/../Src/zsh &&
{
    mkdir dir1
    mkdir dir2
    touch file1
    touch file2
  }
./Y01completion.ztst: test failed.

Any ideas what's going on?

Re: Y test failures (zpty)

[Bart Schaefer]

On Mar 11, 12:32am, Clint Adams wrote:
}
} Any ideas what's going on?

It would appear that the get_pty() function is failing.  Why that's
happening, I couldn't say.  There are two different versions of that
function in zpty.c -- is the wrong one getting used for some reason?

Has the naming convention for pty devices changed?

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net   

Re: Y test failures (zpty)

[Clint Adams]

> It would appear that the get_pty() function is failing.  Why that's
> happening, I couldn't say.  There are two different versions of that
> function in zpty.c -- is the wrong one getting used for some reason?

The one defined is one with /dev/ptyxx.

I'm not entirely certain that this is the wrong one; I have
Unix98 ptys in /dev/pts (this is what nearly everything uses),
and I have /dev/pty?? as compatibility symlinks to /dev/pty/m[0-9]+.
The nodes in /dev/pty are char devices (major 2), which are the
traditional BSD-style pty's.

> Has the naming convention for pty devices changed?

Unix98 ptys showed up as an option quite some time ago, but
you used to be able to still get BSD-style pty's if you asked
for one.

I imagine that something could be broken with the kernel or devfsd.
Irrespective of that, I'd think that zsh should do the /dev/ptmx
decision at runtime.  Would this hurt anything?

Re: Y test failures (zpty)

[Bart Schaefer]

On Mar 11,  2:02pm, Clint Adams wrote:
}
} I have Unix98 ptys in /dev/pts (this is what nearly everything uses),
} and I have /dev/pty?? as compatibility symlinks to /dev/pty/m[0-9]+.

Is it possible that all the /dev/pty?? terminals are simply used up?

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net   

Re: Y test failures (zpty)

[Sven Wischnowsky]

Bart Schaefer wrote:

> On Mar 11,  2:02pm, Clint Adams wrote:
> }
> } I have Unix98 ptys in /dev/pts (this is what nearly everything uses),
> } and I have /dev/pty?? as compatibility symlinks to /dev/pty/m[0-9]+.
> 
> Is it possible that all the /dev/pty?? terminals are simply used up?

First: I have 2.4.19pre1 with glibc 2.2.4, too, on my laptop and it
works fine.

I once had a similar problem on that Digital Unix box.  It turned out
that for some reason a /dev/ptyXX was free but the corresponding
/dev/ttyXX was not writable by anyone else than root.  That made
get_pty() fail, of course.

Maybe we should change that function to re-try /dev/pty's in that
case?  Haven't seen that problem anywhere else, though.


Bye
  Sven

-- 
Sven Wischnowsky                          wischnow@berkom.de

Re: Y test failures (zpty)

[Clint Adams]

> Is it possible that all the /dev/pty?? terminals are simply used up?

Well, there are under 15 /dev/pts pty's in use out of a possible 256.
There are no /dev/pty?? in use at all.

Re: Y test failures (zpty)

[Clint Adams]

> I once had a similar problem on that Digital Unix box.  It turned out
> that for some reason a /dev/ptyXX was free but the corresponding
> /dev/ttyXX was not writable by anyone else than root.  That made
> get_pty() fail, of course.
> 
> Maybe we should change that function to re-try /dev/pty's in that
> case?  Haven't seen that problem anywhere else, though.

I don't know about Digital Unix, but looking at the sources for
OpenSSH, modemu, and rxvt-ml, I see that OSF/1 has something called
openpty().

Also, there is grantpt() and unlockpt(), getpty(), /dev/ptmx, /dev/ptc
(AIX), /dev/ptym/clone, and various sets of static /dev/pty.+ device
nodes.

All three programs mentioned above use build-time decisions, so I am
confused as to how binaries are portable between systems that do and do
not support Unix98 pty's.