Re: XML::LibXML-1.56

Hi Joachim, All,

can you please try to reproduce the problem with latest CVS version of
LibXML and LibXSLT?

cvs -z3 -d:pserver:anonymous@xxxxxxxxx:/home/cvs co XML-LibXML

PmmREFCNT_dec is the most usual place of crashes because most freeing
happens there. It doesn't imply the problem is in PmmREFCNT_dec.

Maybe I didn't follow this thread carefully enough, but I think the
subject isn't correct here. If the sigsegv happens within LibXSLT's
self tests on functions, it is most probably problem related to
LibXSLT not LibXML.

-- Petr

Joachim Bauernberger <jbauernberger@xxxxxxxxxxxxxxxx> writes:

> Hi,
>
> sill me!!!
>
> <duh>I got the subject of this thread wrong....</duh>
>
> What I'm are debugging here is XML::LibXML-1.56 + XML::LibXSLT-1.53
> (XML::LibXML-1.53 is probably fine I dunno ...)
>
> Regards,
> ~/joachim
>
> On Wednesday 10 December 2003 12:24, Joachim Bauernberger wrote:
>> Hi,
>>
>> i just had some spare time and was able to confine the problem further to
>> the PmmREFCNT_dec() function in perl-libxml-mm.c of XML-LibXML-1.56:
>>
>> I have added some debug printf statements to demonstrate where it crashes
>>
>> /* decrements the proxy counter. if the counter becomes zero or less,
>>    this method will free the proxy node. If the node is part of a
>>    subtree, PmmREFCNT_def will fix the reference counts and delete
>>    the subtree if it is not required any more.
>>  */
>> int
>> PmmREFCNT_dec( ProxyNodePtr node )
>> {
>>     xmlNodePtr libnode = NULL;
>>     ProxyNodePtr owner = NULL;
>>     int retval = 0;
>>
>>     if ( node != NULL && node) {
>>         printf("## Before PmmREFCNT(node)\n");
>>         retval = PmmREFCNT(node)--;
>>         printf("## node->count=%d\n",node->count);
>>         printf("## PmmREFCNT(node)=%d\n",PmmREFCNT(node));
>>         printf("## retval=%d\n",retval);
>>         if ( PmmREFCNT(node) <= 0 ) {
>>             xs_warn( "NODE DELETATION\n" );
>>             libnode = PmmNODE( node );
>>             if ( libnode != NULL ) {
>>                 if ( libnode->_private != node ) {
>>                     xs_warn( "lost node\n" );
>>                     libnode = NULL;
>>                 }
>>                 else {
>>                     libnode->_private = NULL;
>>                 }
>>             }
>>
>>             PmmNODE( node ) = NULL;
>>             if ( PmmOWNER(node) && PmmOWNERPO(node) ) {
>>                 xs_warn( "DOC NODE!\n" );
>>                 owner = PmmOWNERPO(node);
>>                 PmmOWNER( node ) = NULL;
>>                 if( libnode != NULL && libnode->parent == NULL ) {
>>                     /* this is required if the node does not directly
>>                      * belong to the document tree
>>                      */
>>                     xs_warn( "REAL DELETE" );
>>                     PmmFreeNode( libnode );
>>                 }
>>                 xs_warn( "decrease owner" );
>>
>>              /*
>>
>> At this point owner is a dangling pointer which will crash when
>> dereferencing it in the next call to PmmREFCNT_dec( ) -> remember this func
>> is recursive ...
>>              */
>>
>>                 printf("## Before 1 PmmREFCNT_dec(owner)\n");
>>                 PmmREFCNT_dec( owner );
>>
>>              /* we won't get here */
>>                 printf("## After 1 PmmREFCNT_dec()\n");
>>             }
>>             else if ( libnode != NULL ) {
>>                 xs_warn( "STANDALONE REAL DELETE" );
>>                 PmmFreeNode( libnode );
>>             }
>>             Safefree( node );
>>             /* free( node ); */
>>         }
>>     }
>>     else {
>>         xs_warn("lost node" );
>>     }
>>     printf ("Returning retval=%d\n",retval);
>>     return retval;
>> }
>>
>>
>> ~~~~~~~~~~~~~~~~~~~~~
>> ~~~~~~~~~~~~~~~~~~~~~
>>
>> Having a quick look at the state of "owner" shows that it is probably not
>> fit for being passwd around:
>>
>> (gdb) p *node->owner
>> $10 = {_private = 0x30312f74, type = 1668183398, name = 0x6e6f6974 <Address
>> 0x6e6f6974 out of bounds>, children = 0x742e73, last = 0x4015a6e8, parent =
>> 0x41, next = 0x0,
>>   prev = 0x0, doc = 0x0, ns = 0x0, content = 0x0, properties = 0x0, nsDef =
>> 0x82eb5e8}
>>
>> But where does owner get srewed up in the first place? I doubt that it's
>> anywhere in perl-libxml-mm.c !!
>>
>> I will try to look deeper into this when I have more time at hand, -> also
>> I am not familiar with the code so the person who wrote this might be able
>> to spot the problem faster than me ...
>>
>> Note that you have to add the -g switch to CFLAGS in the XML-LibXML
>> Makefile to get any meaningful debugging with gdb.
>>
>> Cheers,
>> ~/joachim
>>
>> On Tuesday 09 December 2003 08:55, Joachim Bauernberger wrote:
>> > Hi Grant,
>> >
>> > On Monday 08 December 2003 18:49, Grant McLean wrote:
>> > <snip/>
>> >
>> > > > [t][jmbdev]# perl ./10functions.t
>> > > > 1..14
>> > > > ok 1
>> > > > ok 2
>> > > > ok 3
>> > > > ok 4
>> > > > ok 5
>> > > > ok 6
>> > > > Segmentation fault
>> > >
>> > > One obvious difference between running 'make test' and running the test
>> > > directly is that make test uses the recently built version of LibXML
>> > > sitting in the blib directory whereas your simple perl command to run
>> > > the test script uses the version of LibXML installed on your system.
>> >
>> > That version of LibXML was already installed from before, so it should
>> > not make a difference (I also double checked by using the -Include
>> > switches as you did below but it's all the same...)
>> >
>> > cheers,
>> > ~/joachim
>> >
>> > > You could try editing the command line generated by 'make test':
>> > >
>> > > PERL_DL_NONLAZY=1 /usr/bin/perl -Iblib/arch -Iblib/lib \
>> > > -I/usr/lib/perl/5.6.1  -I/usr/share/perl/5.6.1 t/10functions.t
>> > >
>> > > or you could use the 'prove' script that comes with recent versions of
>> > > Test::Harness to achieve a similar thing.
>> > >
>> > > Of course if you've already run 'make install' then the files from blib
>> > > should have been copied to /usr/local/lib already.
>> > >
>> > > Cheers
>> > > Grant
>

-- 
_______________________________________________
Perl-XML mailing list
Perl-XML@xxxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe: http://listserv.ActiveState.com/mailman/mysubs