[ZPT] XSLT version of TAL, pushing caching to the client

[Brad Clements]

About 3 years ago I posted on this list asking if anyone else was 
interested in using TAL/METAL to produce XSLT stylesheets that could 
then be further processed.

The folks over at Bitflux have toyed with this a bit:

http://wiki.bitflux.org/Templates_TAL

I've been playing with their TAL processing stylesheet (which they seem 
to have stopped working on)

http://svn.bitflux.ch/repos/public/popoon/trunk/components/transformers/xsltal/tal2xslt.xsl

And have started adding fill-slot/define-slot support to it.

Also, trying to setup some unit tests.

--

All my Zope websites these days are more data driven then content 
driven. 

Each page "class" is a python script that returns an "XML context blob" 
with a stylesheet PI. That stylesheet then generates the HTML (usually 
on the client browser, but on the server if necessary).

By page "class", I mean that many URI's are handled by a single 
pythonscript via traverse_subpath handling, and each script handles a 
specific class of "page types". 

The xslt stylesheets themselves also have xsl:include statements that 
work, sorta like metal macros. 

With further use of the document() function in the xslt stylesheets, 99% of 
browsers can cache individual page fragments in the client's browser.

This approaches the ultimate goal of the ESI (Edge Side Includes) 
specification.

http://www.esi.org/index.html


Throw into that mix AJAX-style interfaces (sorry for buzzword use), and 
the usefulness of XML-backed Zope sites increases dramatically.

Suddenly my RESTful data sources can be exposed, either by 
compositing into the html page directly during HTML generation via the 
document() function.. Or dynamically via XML-HTTP requests, or both!

Also, where appropriate, my data sources can render their output as XML 
or javascript literal or javascript arrays (though the later is client-side 
javascript specific).  Mozilla's transformix engine isn't very fast, so for 
google-suggest style interfaces sending over javascript arrays is better 
than sending XML.

Though in a pinch, I also have used an XSLT file to convert XML back 
into javascript literal format on the client. This works until E4X is 
implemented in client browsers:

http://www.ecma-international.org/publications/standards/Ecma-357.htm


--

For my clients, using XML/XSLT as the basis for web-site architecture 
has dramatically simplified my site designs, while simultaneously 
increasing the flexability in interface creation through the composition of 
data sources and templates (xslt).


Additionally, offline back-end processing via XSL-FO is now possible with 
hardly any additional effort.

--

But back to ZPT and XSLT.

In ZPT, the TALES context space serves as the root of the data source 
tree that is used to expand TAL elements.

ZPT sheets can be tested in-situ as they are being edited by clicking on 
the test tab. 

For TAL/XSLT, the context space comes from the "XML context blob" 
(with the launch point stylesheet PI).

TAL to XSLT requires a pre-processing step, where the tal document is 
converted to xslt. Perhaps this will be a three-step process if we want to 
make METAL expansion a distinct phase (to ease previewing and editing 
of  the TAL page).


This makes working with TAL/XSLT at least a  three-step process. 

1. Metal expansion

2. TAL to XSLT

3. Final rendering of "XML context blob" to XHTML (or whatever). This 
step usually occurs on the client browser.

Different XML context blobs can use the same XSLT stylesheet. So in 
some respects this is somewhat similar to Zope acquisition, where the 
"data context" can be applied to a "template context" through the 
arrangement of the requesting URL.

--


So, is anyone interested in TAL to XSLT?

I guess I should have blogged about this instead.. Can't remember where 
I left my blog...




-- 
Brad Clements,                bkc at murkworks.com    (315)268-1000
http://www.murkworks.com                          
AOL-IM or SKYPE: BKClements