Re: Rethinking XSLT 2.0 design

[Norman Walsh <ndw@nwalsh.com>]

Jirka Kosek <jirka@kosek.cz> writes:
>> 1. It's very expensive. The entire document gets processed at least
>> twice. While the idea of simplifying the "downstream" design seemed
>> very attractive, I think the cost is too high.
>
> I think that multiple passes over document are necessary anyway -- I
> think that profiling should be default in XSLT 2.0 stylesheets. Also I'm
> now working on transclusions use-cases document for DocBook TC and in
> DocBook specific transclusion mechanisms is emerging in my head. This
> means another pass over the document.

I agree that it should be possible to do all those things, and that
they should be easy. I'm not sure I agree that they should all always
happen whenever you process a document.

In an, *ahem*, database context, for example, I can imagine wanting
to factor some of those processes in different ways. :-)

> I'm not sure whether "normalization" was good or bad, but I think that
> its processing burden is not so high compared to stylesheets load time.

Yeah. Maybe. In some contexts. :-)

>> 4. I made a DB4 to DB5 conversion phase part of that normalization,
>> and that makes the problems even worse (three phases, an even broader
>> disconnect).
>
> What about not supporting DB4 in XSLT 2.0 stylesheets? It will simplify
> things and users can always run their own DB4->DB5 process before
> stylesheets are applied.

Yes, I think that's probably the right answer. More generally, I think
I want to decompose all the functionally separate phases. We might
want to provide a stylesheet that does all the steps, but I (think I)
also want the ability to apply the different phases at different
times.

                                        Be seeing you,
                                          norm

-- 
Norman Walsh <ndw@nwalsh.com>      | Ignorance is a precious commodity:
http://www.oasis-open.org/docbook/ | once it's gone, you can't get it
Chair, DocBook Technical Committee | back.

PGP signature