Daniel Fagerstrom wrote:<snip/>
You're right no need to optimize until we know if it is needed. AFAICS the only reasonable way of accessing the XML input is things like XPath that in general need access to the complete XML-tree. And in that case we can reuse XSLT processor technology if needed. Going for something like STX-style access would be far to complecated for the intended audience.
Why is STX considered more complex than XSLT?
I mean for writing a simple to use template language that works like XSLT: XML->XML driven by a stylesheet. If the template writer want to render data in another order than that of the input document, the streaming, non-buffering nature of STX will bite them. They need to write rules that are trigered by the correct input events and buffer the data. This requires some thinking even for experienced programmers.
In XSLT you have random access to the whole tree and don't need to care about temporal order of things. Much easier.
Sure it doesn't have arbitrary source tree access, but the syntax is very similar to XSLT by design. Instead of <xsl:template>, you have <stx:template>.
I have written a number of STX style sheets for a number of different application, where avoiding buffering was important for performance. I would say that syntax is its only similarity with XSLT. Or do you have different experience.
I still think that STX would act as an effective macro expansion mechanism instead of having macros in Java/SAX.
As said in another letter: macro expansion is written in the template language it self.
/Daniel
