From: BG - Ben Armstrong <BArmstrong@...>
Date: 2005-05-27T01:52:40+09:00
Subject: Re: shim for select() for VMS & general porting questions

Nobu,

> Moved all to vmsruby_private.[ch].  Does this work?

First, I think our problem is bigger than this pair of files.  I know
Masamichi was making *all* of his commits to the stable tree, which I
agree is not the usual thing to do, and should be fixed.  Shouldn't we
take care to ensure all of the changes he committed to this branch are
merged into HEAD before proceeding with further patches?

Second, before I try your new patch I think I should mention we
encountered two problems with your earlier patch when we tried to
exercise code that actually does select() on sockets.  In our earlier
test it appeared that the threads deadlocked and the deadlock condition
was not detected.  I think this is more than just bad luck because the
problem was completely reproducible.  Also, just a short while ago when
we exercised this code again, instead of a deadlock, it threw a "Bus
Error":

A:002000> ruby newfiles.new DYMA
newfiles.new:128: [BUG] Bus Error
ruby 1.8.2 (2004-12-16) [alpha-vms]

The stack dump (edited) includes these relevant lines:

 RUBYSHR  VMS  rb_vms_select            13135 0000000000001064
000000000013EF54
 RUBYSHR  EVAL  rb_thread_schedule      23455 00000000000198C4
00000000000BE5C4
 RUBYSHR  IO  appendline                16807 0000000000002838
00000000000E0D48

The code exerpt this happens in is as follows:

126    File.open(dirFilename) do |input| 
127        $stderr.puts "open"
128        input.each do |line| 
129            $stderr.puts "line: #{line}"
130            yield line
131        end
132    end

Since your new patch didn't substantially change, I expect it will have
the same bug.  I unfortunately do not yet have a minimal failing test
case, but I'll try to make do with what we have so far to figure out
what's happening here.

Ben