killing subprocess ==> reproducible core dump (fwd)

[p5pRT]

Migrated from rt.perl.org#935 (status was 'resolved')

Searchable as RT935$

[p5pRT]

From [email protected]

mips-sgi-irix6.2, perl 5.004_04.

The attached program will make perl dump core on Irix 6.2.
On Linux 2.0.36, it causes perl to exit with "out of memory" instead.

What I am trying to do is, run a subprocess, and if it doesn't complete
in a few seconds, kill it and move on.

In this example, the part of the non-terminating sub-process will be
played by ""perl -e '1 while 1;'" in order to make it nice and
self-contained. (In my actual program, the sub-process in question is
"giftopnm", which sometimes gets stuck in a loop when fed bad data.)

The call to "kill" apparently does nothing.

If you un-comment the call to waitpid, perl doesn't die -- but waitpid
never returns, because the pid hasn't actually died.

It dumps core the second time through the loop, as soon as it tries to
launch the second sub-process.

Any suggestions on how I can do what I'm trying to do?

Second question -- it seems that you can't use "use diagnostics" in
programs that use `eval' to do timeouts. It always says

  Uncaught exception from user code​:
  Uncaught exception from user code​:
  alarm
  main​::__ANON__('ALRM') called at /tmp/a.pl line 27
  eval {...} called at /tmp/a.pl line 18
  main​::foo() called at /tmp/a.pl line 54
  ...propagated at /tmp/a.pl line 35.
  main​::foo() called at /tmp/a.pl line 54

when, in fact, I *am* catching that exception (at least in the sense of,
the program does what I expect w.r.t. timeouts when I don't use
diagnostics.) Is there any way to get diagnostics and use alarms at the
same time?

--
Jamie Zawinski jwz@​jwz.org http​://www.jwz.org/

[p5pRT]

From [email protected]

dumper.pl