Re: [xml] Race condition in xsltInit (thread safety)

[Daniel Veillard <veillard redhat com>]

On Wed, Feb 22, 2006 at 03:06:51PM -0500, Christopher R. Palmer wrote:
> Daniel Veillard wrote:
> > On Mon, Feb 20, 2006 at 01:13:57PM -0500, Christopher R. Palmer wrote:
> > > Digging into this a little more, it looks like xsltInit is called every 
> > > time that a new stylesheet is parsed.  xsltInit uses 
> > > xsltRegisterAllExtras which always updates all of the hash entries and 
> > > the hash function is (not surprisingly) not thread safe for updates.
> > >
> > > CVS tells me that xsltInit was changed to fix bug #165201 (calling 
> > > xsltInit multiple times) where previously it had a local static 
> > > initialized flag.
> > >
> > > The best fix I can suggest is to export xsltInit and have a static 
> > > initialized flag.  To not break the fix of bug #165201, add an internal 
> > > function to signal that the library is no longer initialized.
> > >
> > > I've attached such a patch for possible inclusion in the code base.
> >
> >  Okay, this seems to make sense, and exporting the init function might
> > be the only way to garantee the absence of race in multithreaded uses.
> > I guess you tested it, I have no good way to test multithreaded uses,
>
> I tested it as well as you can test a race condition...  Before the change, 
> my multi-threaded service was crashing at least once a day and it hasn't 
> crashed in several days now.

  Good, good, as I said it looks logical to me :-)

Daniel

-- 
Daniel Veillard      | Red Hat http://redhat.com/
veillard redhat com  | libxml GNOME XML XSLT toolkit  http://xmlsoft.org/
http://veillard.com/ | Rpmfind RPM search engine http://rpmfind.net/