Re: [xml] endianness problem in regression suite

[Nicolai Langfeldt <janl linpro no>]

Daniel Veillard wrote:
> On Thu, Nov 20, 2003 at 10:17:19AM +0100, Nicolai Langfeldt wrote:
>
> > Building libxml2 on Solaris/Sparc.  In the tests I see a problem in the 
> > XML regression tests.  From Makefile.am:
>
> [...]
>
> > Testing utf16bom.xml
> > Binary files ./result/utf16bom.xml and result.utf16bom.xml differ
> >
> > I've run xmllint by hand on result/utf16bom.xml.  In emacs hexl-mode I 
> > see this in the original file
> >
> > 00000000: fffe 3c00 3f00 7800 6d00 6c00 2000 7600  ..<.?.x.m.l. .v.
> > 00000010: 6500 7200 7300 6900 6f00 6e00 3d00 2200  e.r.s.i.o.n.=.".
> >
> > And this is the xmllint output:
> >
> > 00000000: feff 003c 003f 0078 006d 006c 0020 0076  ...<.?.x.m.l. .v
> > 00000010: 0065 0072 0073 0069 006f 006e 003d 0022  .e.r.s.i.o.n.=."
> >
> > I'm told that feff == byte-order mark and that fffe == undefined 
> > character.
>
>
>   Well the file is flagged as UTF-16, which unfortunately has 2 variant
> onle little-endian, and the other big-endian. Seems the libxml2 UTF-16
> serialization code uses the platform native endianness instead of always
> using little-endian. The files are still well formed, but maybe this need
> to be fixed, xmlInitCharEncodingHandlers() already tests the endianness
> of the architecture and it's used for the UTF-8 to UTF-16 input conversion.
> This could also be the behaviour of your iconv() library taking over 
> libxml2 default UTF-16 routines, that would need to be checked under a 
> debugger.

config.log shows that configure finds both iconv and zlib.  This is the 
solaris 8 bundled iconv (and zlib).

I don't know much about neither xml or unicode.  I'm quite happy if the 
output file is well- formed.  But perhaps the compare operation needs to 
be more complicated to take this into account.

But I'm sure I can run this under a debugger if you can hint about what 
to look for.

Nicolai