Weird SIGCHLD-related problem

Hello-

I'm experiencing an odd issue with a fork'd child process and I was
hoping somebody on this list might be able to help my find a solution.
Here are the conditions.

First things first:

OS: FreeBSD 4.11-RELEASE-p11
ruby: ruby 1.8.2 (2004-12-25) [i386-freebsd4]

I'm using a ruby script of my own devising to filter my email from
qmail. This works by just adding "|bin/mail-filter.rb" to the .qmail
file in my home directory. The script does a lot of things, but to
verify the problem is with signal handling, I created a really simple
script just for testing. It's contents are:

#!/usr/local/bin/ruby18
Signal.trap(0, lambda { puts "Terminating: #{$$}" })
Signal.trap('CLD') { puts 'Child died' }
fork && Process.wait

The output of which ought to be something like this:

Terminating: 62679
Child died
Terminating: 62678

That script was actually borrowed verbatim from Pickaxe II. When I run
this script from the shell, everything works as expected. However,
when it gets excuted by qmail (qmail-local, to be exact) the child
fork becomes a zombie. It's basically like the CLD signal is never
caught and therefore the Process.wait hangs indefinitely. Really odd.

While testing, I was having problems with all these zombied processing
hanging around forever and decided to alter the test script to kill
itself, using a Thread, after 30 seconds (obviously more than enough
time for the forked child to die), so the script became:

#!/usr/local/bin/ruby18
Thread.new { sleep(30); puts "Killing"; Process.kill(9) }
Signal.trap(0, lambda { puts "Terminating: #{$$}" })
Signal.trap('CLD') { puts 'Child died' }
fork && Process.wait

When run on the shell, this produces no noticable difference in output
as the race condition (30 seconds until the kill is fired) will pretty
much never happen before the child dies and the parent dies. However,
what happened when running this script under qmail really surprised
me. The output in my logs was this:

Terminating: 62679
Terminating: 62678

So the script, more or less, works, except that it seems like the CLD
signal is still not getting trapped. I haven't spent any time looking
into the internals of either ruby or qmail, so I'm not sure what goes
on with the signal handling in either of those, but it strikes me as
really strange that by simply introducing a Thread into the mix, the
Process.wait doesn't hang forever and the fork'd child doesn't become
a zombie.

Any ideas what might be going on?

To add more confusing to the mix, I should, also, note that this same
script (changing Signal to Kernel, of course) works fine under ruby
1.6 (ruby 1.6.8 (2003-10-15) [i386-freebsd4] to be exact. I have both
versions running side-by-side on my system.

If I get the chance (which I probably won't for a couple days), I'll
dig into ruby's signal handling and threading code and see if I can
figure anything out, but I thought I'd run it past the guru's first.

Cheers.

Ben