[Patch]: Setting the winpid in pinfo

Pierre A. Humblet pierre@phumblet.no-ip.org
Sat Sep 11 03:17:00 GMT 2004

At 10:49 PM 9/10/2004 -0400, Christopher Faylor wrote:
>On Fri, Sep 10, 2004 at 09:29:35PM -0400, Pierre A. Humblet wrote:

>>Here are 2 other small changes.
>>When killing the (system stressing) program I used to track the
>>fork bug, I noticed that it occasionally leaves stray processes
>>behind, as in
>>821 49315459       1  557307 4245651837    0  740 20:50:10 /c/HOME/PIERRE/A
>>What's happening is that there is a race between the winpids
>>scanning in kill_pgrp and the creation of new processes.
>>The patch in fork.cc improves the situation by exiting
>>children immediately when the parent dies during a fork.
>>Also a ^C immediately terminates processes that are initializing.
>>That is not desirable for processes created by Cygwin processes. 
>>The patch in exceptions.cc changes that.
>I don't understand.  You're removing an exit on a CTRL-C in
>exceptions.cc and essentially adding it to fork.cc.
The change in fork.cc affects all signals, not only ^C
Without the change, you get those stray orphan processes
on "kill -9 parent". And 5 minutes (FORK_WAIT_TIMEOUT) later
you get a fatal_api message on your screen, for no apparent reasons.

> How is that an improvement?  Why aren't both parent and child going away 
> when they see a CTRL-C already?
The parent may have chosen to ignore the ^C, in which case it will 
stay alive. Its fork will then fail unexpectedly. Few programs are
designed to retry on a failed fork. With the change, all goes well.

>If the child process has not become fully awake yet it should just exit now.

Not if it is destined to ignore ^C. The group leader will wait until the 
[grand]child has grown up and its sendsig is != NULL, and then send the
The child will then handle it properly.

The motivation for looking at this was
<http://cygwin.com/ml/cygwin/2004-07/msg01120.html>, as well as personal
observations. I have run thousands of tests and eveything has gone well.


More information about the Cygwin-patches mailing list