RE: [xml] Windows threading issues

[Jesse Pelton <jsp PKC com>]

Does anyone have a copy of Eugene's patch?  He hasn't responded to a message
I sent several days ago.

> -----Original Message-----
> From: Stéphane Bidoul [mailto:stephane bidoul softwareag com] 
> Sent: Friday, July 11, 2003 5:11 AM
> To: 'Jesse Pelton'
> Cc: xml gnome org
> Subject: RE: [xml] Windows threading issues
>
>
> I'm CC'ing the list, because other can help too.
>
> > -----Original Message-----
> > From: Jesse Pelton [mailto:jsp PKC com]
> > Sent: 10 July, 2003 21:56
> > To: 'Stéphane Bidoul'
> > Subject: RE: [xml] Windows threading issues
> >
> >
> > Thanks for pointing out Eugene's message; I wish I'd seen it
> > earlier.  (I
> > only recently realized that I was subscribed to libxslt but
> > not libxml.)
> > His summary of the situation pretty much covers the bases.
> > My situation is
> > slightly different - I recently figured out that DllMain()
> > never gets called
> > with DLL_THREAD_DETACH, presumably because IIS terminates its
> > threads with
> > TerminateThread() - but it's the same basic problem.  
> Neither approach
> > (watchdog thread or DllMain()) solves the problem, despite 
> my earlier
> > optimism about the second one.
>
> If IIS is really doing TerminateThread, you're in trouble, indeed:
> you'll get memory leaks each time a thread terminates.
>
> > Perhaps it would make more sense for xmlCleanupThreads() to
> > release the
> > thread-local data.  I'm already calling xlmCleanupParser(),
> > which then calls
> > xmlCleanupThreads(), so this would allow a clean exit.  The
> > trouble is, I
> > don't see a good architecture.  I think there'd have to be an
> > application-global list of thread data.  Access to this list
> > would have to
> > be synchronized, with possibly significant performance
> > consequences.
>
> Eugene's patch is doing just that. I don't think there
> are performance issue (access to the synchronized list is
> only during thread creation and termination).
> It's probably a good starting point.
>
> > I'm
> > just muttering out loud here, and hoping that something
> > workable will come
> > to me in my dreams tonight.  If not, I can probably use
> > Eugene's approach,
> > but I'd like to come up with a general solution.  Right now
> > my head hurts
> > and I'm going home.
>
> -sbi