Re: [xml] Any outstanding bugs in closing parser contexts or running with different memory handlers?

[Sean Chittenden <sean chittenden org>]

> > ctxtClose is a blatant rip off from someplace inside of the libxml
> > sources:
> >
> > static void ctxtClose(FILE *f) {
> >   if (f != stdin) {
> >     fclose(f);
> >   }
> > }
> >
> >
> > ... so I'm not really sure why I'm getting the following cryptic
> > error:
> >
> > ==19278== Invalid read of size 1
> > ==19278==    at 0x40308758: (within /lib/libc-2.2.5.so)
> > ==19278==    by 0x431B10BC: ctxtClose (ruby_xml_parser.c:17)
>
> > ==19278== Invalid read of size 2
> > ==19278==    at 0x40308789: (within /lib/libc-2.2.5.so)
> > ==19278==    by 0x431B10BC: ctxtClose (ruby_xml_parser.c:17)
> > ==19278==    by 0x432BCDAA: xmlFreeParserInputBuffer (xmlIO.c:1575)
>
> > ==19278== Invalid read of size 4
> > ==19278==    at 0x403087CB: (within /lib/libc-2.2.5.so)
> > ==19278==    by 0x431B10BC: ctxtClose (ruby_xml_parser.c:17)
> > ==19278==    by 0x432BCDAA: xmlFreeParserInputBuffer (xmlIO.c:1575)
>
> > What's interesting is that the read size grows from 1 to 4.  Why I'm
> > not sure, and think that there could be a stray bit being written to
> > in Ruby, but I'm not 100% sure.  Since valgrind only runs on Linux,
> > I'd perfectly happy to chalk this up as a linux-ism, but it crashes on
> > FreeBSD too.  :-/  Anyone have any thoughts?  -sc
>
>   Hum, how are those InputBuffer created ? The fact that the read
> size changes may simply be due to buffer access informations. Note
> that some optimizations from GCC can actually lead to such invalid
> read warnings from valgrind, it can also be processor dependant
> (e.g. Athlon vs. PIII).

xmlCreateMemoryParserCtxt()

I'm going back and recompiling without any -O optimizations to see if
that helps, but I'm seeing this on gcc 2.9.4 and 3.2.

> Make sure you have the latest valgrind 1.0.1, if needed get in touch
> with valgrind author, and get a debug version of the C library to
> see exactly where this occurs *inside*.

1.0.3 work?  I'm going to see if I can't get a hold of a debug versions
of the core libs, but the only linux box I have access to isn't mine.
:)  Thanks for the pointers though.  -sc

-- 
Sean Chittenden