pthread_kill: signals remain pending after target thread exits

[Corinna Vinschen]

On Nov  2 23:54, John Carey wrote:
> > From: Corinna Vinschen [corinna-cygwin@cygwin.com]
> > Sent: Monday, November 02, 2015 9:58 AM
> > On Nov  2 15:18, Corinna Vinschen wrote:
> > > On Oct 29 16:16, John Carey wrote:
> > > > > From: Corinna Vinschen [corinna-cygwin@cygwin.com]
> > > > > Sent: Thursday, October 29, 2015 1:02 AM
> > > > > On Oct 28 10:14, Corinna Vinschen wrote:
> > > > ...
> > > > > > > Thanks; that was fast!  I tried replacing cygwin1.dll with cygwin1-20151023.dll .
> > > > > > >
> > > > > > > The original test case now works.  I checked some of my other tests,
> > > > > > > and unfortunately some of them failed, so I extracted out a new test
> > > > > > > case, which is attached.  My guess is that something is subtly different
> > > > > > > about the timing on this test.
> > > > > >
> > > > > > Is this a regression?  Did it work with 2.2.1?
> > > > >
> > > > > Answering myself here, this didn't work at all in 2.2.1.  I can
> > > > > reproduce the problem and I'm going to take a look.  Not sure if
> > > > > there's a quick solution, though.  This looks like a deadlock
> > > > > situation.  The signal definitely got send, I just don't see yet
> > > > > why it's not handled in wait_sig.
> > > >
> > > > Yes, test_pending_signal2.c fails in 2.2.1, though the symptoms
> > > > differ: there is no Cygwin process left hanging that cannot be
> > > > killed by a Cygwin signal.
> > > >
> > > > But both tests appear to pass in Cygwin 1.7.9, so in that sense
> > > > it is a regression, just not a recent one.
> > >
> > > The regression here was my original fix.  It didn't clear pending
> > > signals for an exiting thread thorougly enough.  I applied a patch now
> > > which should fix this issue.  I'll upload a new developer snapshot to
> > > https://cygwin.com/snapshots/ later today or tomorrow (still looking
> > > into the PrlFS issue).  Please give it a try.
> > 
> > Snapshot is up.  The fix is in the 2.3.0-0.5 test release as well.
> > Please give any one of them a try.
> 
> I tested snapshot cygwin1-20151102.dll, and it works in all of my tests.  Nice!

One good news, thanks!

> I'm glad you figured out how to delete those pending signals.

Given that I introduced the regression in the first place, it's a bit
embarrassing that it took me so long to understand it 8-}


Corinna

-- 
Corinna Vinschen                  Please, send mails regarding Cygwin to
Cygwin Maintainer                 cygwin AT cygwin DOT com
Red Hat
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://cygwin.com/pipermail/cygwin/attachments/20151103/5ecaa737/attachment.sig>