[xslt] RE: [xml] periods legal in an NMTOKEN

["David Cramer" <dcramer broadjump com>]

You aren't kidding when you say verbose. I guess it normally is quite by
contrast :)

Turns out xsltproc is correct--I was feeding it some questionable xml in
the form: attr="foo " (where attr is an NMTOKEN). Not sure if it is
strictly a 'validity error'; nsgmls doesn't complain about it. 

I promise that the message I mentioned isn't coming from an xsl:message
tho. Maybe some of the code is doing more than you think?

Total processing time (including a fo renderer and ghostscript) for one
of our larger docs:
xsltproc: 681 seconds  
Saxon:    580 seconds  

That surprises me--I'd expected xsltproc to be faster. Maybe I can do
some profiling to figure out where the expensive stuff is in any case.
Btw., these are docbook documents, preprocessed extensively, then run
through the docbook xsls (v. 1.50.0 + our customizations).

Other interesting differences:
* Saxon outputs things like — as unicode characters, while
xsltproc leaves them as numeric entity references: e.g. ­.
* Saxon sorts AaBbCc, xsltproc sorts ABC.. abc...

There are other differences--I can tell because of pagination
differences, but I haven't figured out where they are yet. In the fo I
can see that generated id values are different and the order of
attributes is different--other things are harder to make out in the sea
of tags and line break differences. 

Thanks for your help,
David


> -----Original Message-----
> From: Daniel Veillard [mailto:veillard@redhat.com]
> Sent: Wednesday, September 11, 2002 5:25 PM
> To: David Cramer
> Cc: xslt@gnome.org; xml@gnome.org
> Subject: Re: [xml] periods legal in an NMTOKEN
> 
> 
> On Wed, Sep 11, 2002 at 05:19:54PM -0500, David Cramer wrote:
> > Sigh. I prepared a small test case and xsltproc does fine 
> with NMTOKENs.
> > Apparently it's something else--either a bug in our xsl that Saxon
> > misses or a bug in xsltproc that our stuff reveals. I'll 
> try to get a
> > test case that's small enough to be useful. 
> > 
> > Any chance there's a hidden --quiet switch on xsltproc? It 
> processes the
> > document fine, despite its complaints. 
> 
>   Well, the problem is that xsltproc does *not* do validity
> checks like the message you seems to report. It loads the DTD
> in order to expands entities and defaulted attributes but does
> not do checks or report them, so the message seems issued by
> your stylesheets actually. Why, I have no idea ...
>   Run xsltproc with the -v verbose mode and save the stderr
> output on a file, you should be able to spot where the message get
> issued and debug your problem.
> 
> Daniel