Debian Bug report logs - #578271
expect: [kFreeBSD] spawn/wait waits forever if command has any output

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Andreas Metzler <ametzler@downhill.at.eu.org>

To: Debian Bug Tracking System <submit@bugs.debian.org>

Subject: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Sun, 18 Apr 2010 15:19:48 +0200

Package: expect
Version: 5.44.1.15-1
Severity: important

On kfreebsd-* spawn/wait does not work as expected:
----------------
ametzler@asdfasdf:$ cat /tmp/FIND/expect.bug
#!/usr/bin/expect -f

eval echo blubb
wait
ametzler@asdfasdf:$ /tmp/FIND/expect.bug
----------------
and the process just hangs and only reacts to kill -9. I have found
this when investigating why sv-bug-24169.exp in findutils testsuite
does not work on kfreebsd-*.

I suspect this should be an rc bug, please raise the severity if you agree.

http://git.savannah.gnu.org/cgit/findutils.git/tree/find/testsuite/find.gnu/sv-bug-24169.exp

cu andreas

-- System Information:
Debian Release: squeeze/sid
  APT prefers testing
  APT policy: (500, 'testing')
Architecture: kfreebsd-amd64 (x86_64)

Kernel: kFreeBSD 8.0-1-amd64
Locale: LANG=de_AT.UTF-8, LC_CTYPE=de_AT.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash

Versions of packages expect depends on:
ii  libc0.1                       2.10.2-6   Embedded GNU C Library: Shared lib
ii  tcl8.5                        8.5.8-2    Tcl (the Tool Command Language) v8

expect recommends no packages.

Versions of packages expect suggests:
ii  expectk                      5.44.1.15-1 A program that can automate intera

Message #10 received at 578271@bugs.debian.org (full text, mbox, reply):

From: Sergei Golovan <sgolovan@gmail.com>

To: Andreas Metzler <ametzler@downhill.at.eu.org>, 578271@bugs.debian.org

Cc: control@bugs.debian.org

Subject: Re: Bug#578271: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Sun, 18 Apr 2010 19:45:29 +0400

tags 578271 + kfreebsd upstream
severity 578271 grave
thanks

On Sun, Apr 18, 2010 at 5:19 PM, Andreas Metzler
<ametzler@downhill.at.eu.org> wrote:
>
> On kfreebsd-* spawn/wait does not work as expected:

You're right, the bug is reproducible not only on Debian kFreeBSD
but also on a plain FreeBSD.

> ----------------
> ametzler@asdfasdf:$ cat /tmp/FIND/expect.bug
> #!/usr/bin/expect -f
>
> eval echo blubb

Though I'd replace eval by spawn.

> wait
>
> I suspect this should be an rc bug, please raise the severity if you agree.

I agree, it's an RC bug for kfreebsd.

Cheers!
-- 
Sergei Golovan

Message #19 received at 578271@bugs.debian.org (full text, mbox, reply):

From: Sergei Golovan <sgolovan@gmail.com>

To: Andreas Metzler <ametzler@downhill.at.eu.org>, 578271@bugs.debian.org

Subject: Re: Bug#578271: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Sun, 18 Apr 2010 21:25:12 +0400

On Sun, Apr 18, 2010 at 5:19 PM, Andreas Metzler
<ametzler@downhill.at.eu.org> wrote:
> and the process just hangs and only reacts to kill -9. I have found
> this when investigating why sv-bug-24169.exp in findutils testsuite
> does not work on kfreebsd-*.

A quick look reveals that expect hangs on waitpid() call in
Exp_WaitObjCmd() function defined in exp_commands.c

(A few lines from exp_commands.c):
----------------------------------------------------------------------
		while (1) {
		    if (Tcl_AsyncReady()) {
			int rc = Tcl_AsyncInvoke(interp,TCL_OK);
			if (rc != TCL_OK) return(rc);
		    }

		    result = waitpid(esPtr->pid,(int *)&esPtr->wait,0);
		    if (result == esPtr->pid) break;
		    if (result == -1) {
			if (errno == EINTR) continue;
			else break;
		    }
		}
----------------------------------------------------------------------

It seems like FreeBSD kernel doesn't terminate the spawned child if it
did print something to stdout and nobody has read the printed chars.

Though I'm not sure if this behavior is wrong. It's just different.
So, may be the correct way is to fix expect script? The following
script works fine on Linux and FreeBSD:

spawn echo blubb
expect {
eof {puts [wait]}
}

BTW, I've reported this bug upstream. See
https://sourceforge.net/tracker/?func=detail&aid=2989008&group_id=13179&atid=113179

Cheers!
-- 
Sergei Golovan

Message #24 received at 578271@bugs.debian.org (full text, mbox, reply):

From: Sergei Golovan <sgolovan@gmail.com>

To: Andreas Metzler <ametzler@downhill.at.eu.org>, 578271@bugs.debian.org

Subject: Re: Bug#578271: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Mon, 19 Apr 2010 18:22:20 +0400

On Sun, Apr 18, 2010 at 9:25 PM, Sergei Golovan <sgolovan@gmail.com> wrote:
>
> Though I'm not sure if this behavior is wrong. It's just different.
> So, may be the correct way is to fix expect script? The following
> script works fine on Linux and FreeBSD:
>
> spawn echo blubb
> expect {
> eof {puts [wait]}
> }

And the following script hangs both on Linux and FreeBSD kernels:

spawn echo [string repeat a 10000]
puts [wait]

Cheers!
-- 
Sergei Golovan

Message #29 received at 578271@bugs.debian.org (full text, mbox, reply):

From: Andreas Metzler <ametzler@downhill.at.eu.org>

To: 578271@bugs.debian.org

Subject: Re: Bug#578271: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Sat, 1 May 2010 12:07:04 +0200

On 2010-04-18 Sergei Golovan <sgolovan@gmail.com> wrote:
> On Sun, Apr 18, 2010 at 5:19 PM, Andreas Metzler
> <ametzler@downhill.at.eu.org> wrote:
> > and the process just hangs and only reacts to kill -9. I have found
> > this when investigating why sv-bug-24169.exp in findutils testsuite
> > does not work on kfreebsd-*.
[...]
> The following
> script works fine on Linux and FreeBSD:

> spawn echo blubb
> expect {
> eof {puts [wait]}
> }

Hello,

thak you for your help. Due to my (non existing) tcl knowledge I fail
to apply this to my avctual problem in the findutils testsuite
http://git.savannah.gnu.org/cgit/findutils.git/tree/find/testsuite/find.gnu/sv-bug-24169.exp
since I need the exit code.

The aim of this test is to run "find -neweraa" and verify that it
runs and exits cleanly with an error (exit code 1) instead of
segfaulting. (Bug #496853).

cu andreas

Message #34 received at 578271@bugs.debian.org (full text, mbox, reply):

From: Sergei Golovan <sgolovan@gmail.com>

To: Andreas Metzler <ametzler@downhill.at.eu.org>, 578271@bugs.debian.org

Subject: Re: Bug#578271: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Sat, 1 May 2010 16:38:58 +0400

[Message part 1 (text/plain, inline)]

On Sat, May 1, 2010 at 2:07 PM, Andreas Metzler
<ametzler@downhill.at.eu.org> wrote:
>
> The aim of this test is to run "find -neweraa" and verify that it
> runs and exits cleanly with an error (exit code 1) instead of
> segfaulting. (Bug #496853).

The attached patch to sv-bug-24169.exp should make test working on
kfreebsd also.

BTW, now I'm not sure which behavior is correct - FreeBSD which
doesn't kill a child process before the parent reads it's output, or
Linux which kills the child if it's output is sufficiently short.

Cheers!
-- 
Sergei Golovan

[sv-bug-24169.exp.diff (application/octet-stream, attachment)]

Message #39 received at 578271@bugs.debian.org (full text, mbox, reply):

From: Ben Hutchings <ben@decadent.org.uk>

To: Sergei Golovan <sgolovan@gmail.com>

Cc: Andreas Metzler <ametzler@downhill.at.eu.org>, 578271@bugs.debian.org

Subject: Re: Bug#578271: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Thu, 29 Jul 2010 04:52:20 +0100

[Message part 1 (text/plain, inline)]

On Sat, 2010-05-01 at 16:38 +0400, Sergei Golovan wrote:
> On Sat, May 1, 2010 at 2:07 PM, Andreas Metzler
> <ametzler@downhill.at.eu.org> wrote:
> >
> > The aim of this test is to run "find -neweraa" and verify that it
> > runs and exits cleanly with an error (exit code 1) instead of
> > segfaulting. (Bug #496853).
> 
> The attached patch to sv-bug-24169.exp should make test working on
> kfreebsd also.
> 
> BTW, now I'm not sure which behavior is correct - FreeBSD which
> doesn't kill a child process before the parent reads it's output, or
> Linux which kills the child if it's output is sufficiently short.

A pipe normally has a buffer of size 1 page, so a process that writes up
to 4K to a pipe will not block while writing, but a process that writes
more than 4K will block if nothing reads from the other end of the pipe.

Neither kernel kills the child process, but Linux allows it to exit
immediately whereas FreeBSD appears to make it block on exit if it has
output buffered in a pipe.  This behaviour is useful because it
guarantees that the parent will receive the child's output and its exit
code in the 'right' order.  However, I think that either behaviour is
acceptable.

Ben.

-- 
Ben Hutchings
Once a job is fouled up, anything done to improve it makes it worse.

[signature.asc (application/pgp-signature, inline)]

Message #44 received at 578271@bugs.debian.org (full text, mbox, reply):

From: Sergei Golovan <sgolovan@gmail.com>

To: Ben Hutchings <ben@decadent.org.uk>

Cc: Andreas Metzler <ametzler@downhill.at.eu.org>, 578271@bugs.debian.org

Subject: Re: Bug#578271: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Thu, 29 Jul 2010 08:37:19 +0400

On Thu, Jul 29, 2010 at 7:52 AM, Ben Hutchings <ben@decadent.org.uk> wrote:
>
> Neither kernel kills the child process, but Linux allows it to exit
> immediately whereas FreeBSD appears to make it block on exit if it has
> output buffered in a pipe.  This behaviour is useful because it
> guarantees that the parent will receive the child's output and its exit
> code in the 'right' order.  However, I think that either behaviour is
> acceptable.

So, it isn't a bug (and certainly not a bug in Expect which just get what kernel
send to it)? Then I'll close the bugreport and will send your explanation to
the upstream. Thanks for clarification!

Cheers!
-- 
Sergei Golovan

Message #49 received at 578271-close@bugs.debian.org (full text, mbox, reply):

From: Sergei Golovan <sgolovan@gmail.com>

To: 578271-close@bugs.debian.org

Subject: Re: Bug#578271: expect: [kFreeBSD] spawn/wait waits forever if command has any output

Date: Thu, 29 Jul 2010 13:47:04 +0400

Hi!

I'm closing this bugreport as invalid. [wait] behavior in Expect
reflects the underlying Linux or FreeBSD kernel differences.

Cheers!
-- 
Sergei Golovan

Send a report that this bug log contains spam.