Re: [xslt] performance: xsl:attribute vs. textual attribute

[Kasimier Buchcik <kbuchcik 4commerce de>]

Hi Daniel,

Daniel Veillard wrote:
> On Wed, Dec 22, 2004 at 02:38:17PM +0100, Kasimier Buchcik wrote:
>
> > Hi,
> >
> > Performance of "xsl:attribute" versus textual attributes.
> >
> > This mail is related to the optimization request in mail [1].
>
>
>   That is obsolete I think. AVT are now precompiled as much
> as possible. It might be something else.
>
>
> > Did someone already look into it? Is this doable in Libxml2? I did a
> > little performance testing and the "xsl:attribute" version seems to
> > quite slower.
>
>
>   yes, that should be fixed already, [1] is obsolete.
> [...]
>
>
> > Regarding the mail-thread [2], it seems that XSLT authors are not always
> > aware of performance differences here.
> > As a workaround we are currently changing every stylesheet by
> > substituting attributes for xsl:attribute before transforming.
>
>
>   I bet saxon does a far greater job at trying to optimize intermediate
> structure, libxml2 relies far more on the original tree I think.

Ahh, so that's the difference. In the case of attributes: would it break
something if libxslt would not rely on the original tree? I.e. since
not even the position of attributes is significant. Could the same
mechanism as for "xsl:attribute" be used implicitely by libxslt? Or does
this make no sense at all, due to the internals of libxslt?

Greetings (I still hope to be back to the schemata soon),

Kasimier