Slashdot Mirror
W3C Recommends XSL
An Anonymous Coward writes: "The W3C upgraded
XSL 1.0
to the status of a recomendation today, as they reported in
a press release." From that release: "XSLT 1.0 makes it possible to significantly change the original structure of an XML document (automatic generation of tables of contents, cross-references, indexes, etc.), while XSL 1.0 makes complex document formatting possible through the use of formatting objects and properties."
I won't use any shit that I can't parse with fscanf.
XSL has two parts, XSL:FO and XSLT. XSL:FO is an XML format for the printed page. XSLT is a language for transforming one XML format into another.
XSLT is an incredibly useful language to learn. Imagine being able to take a Docbook file and spit out XHTML or XSL:FO > PDF, or... yes, even plain text.
For an idea of how easy it is I wrote a Docbook to HTML converter for 100 docbook tags in four hours, and I hadn't touched XSLT before.
XSLT is based around rewriting a data tree. You can step around the tree like a file system. In a hardcoded way like //html/body/h1, or relatively like ../body.
The language has loops and built in functions for analysing the tree. For example, how many times does a paragraph (p) occur in this HTML document when it's parent is a table cell (td)? count(td/p)
The downside is the bloated syntax. You know those horror stories about an XML programming language that looked like <xml:if(blah= blah)> do this </xml:if> ? XSL is that. But if you can get past that you'll find one of the most wonderful things for mangling XML into whatever format you want.
Applying this technolgy to page generation and tree transformation is not fantastic. This is what you get when you assemble the same gang of professional committee sitters that tanked SGML, and give them a blank slate.
Want to transform documents and render content? Try DOM and CSS. These are much simpler, more flexible, better understood technologies. CSS is actually supported by your browser, and DOM is actually supportedf by real programming languages.
After having used XSLT about a dozen times, I gave up. I had DSSSL expreience before, so I thought that it would be similar and that I could transfer some of my experience over. I felt like I was more productive in yacc or Antlr for quick projects and for anything that started to become more complex, I felt that the XSLT transformation definition complexity grew polynomially. For very large projects (more than a hundred million element, blame someone else for choosing XML for this, not me) I couldn't find a transformer that worked, even though I made sure that the DTD was regular (not context free).
I thought that the original goal was to make a _simple_ declarative language that handled 80% of the transformations easily and left the other 20% to something else (like using a real language). Even the simplest tasks require too much code in my opinion. My first XSLT project, a learning project, was to write the game of life, as it is with every new tool. I have two versions, the shortest being 150 lines that required the field to be an ugly composition of <o/> and <X/> elements.
There is a very high syntax to semantics ratio. Similar operations require different syntax, such as inserting an element can be done literally, but inserting an attribute requires special instructions (a minimum of 2 elements). There is no continuity in the different ways to reference a variable binding. You use templates to generate the structure of the XML but cannot generate text structure in a similar way, neither content data or attribute value data. Blah blah blah... I could go on. It seems like they ended up with a complex framework without strong expressive power. Why couldn't they just do DSSSL with the XML syntax if they wanted it?
Besides, for me it is difficult to look at with XSLT definitions/commands, obviously in XML, sprinkled around output literal XML elements. But I haven't trained my XSLT syntax eye too extensively. It is probably the same way people feel about looking at my DSSSL code.
p.s., I think they lie when they say it is side effect free and completely declarative.
-j
It took me a whole weekend to write Conway's game of life for my first learning project. My first real application was a simple xml logical to HTML converter. It took about 12 hours. The real catch was when I needed to ouput to be dependant on two different far away element bodies.
XSLT is not based on rewriting the XML tree, but the correct way to view the process is on creating an output tree from an input tree. Rewriting would imply that elements that were not specifically analyzed (how ever you define it) would still be in the output tree. Also, you cannot reference the structure of the newly constructed output tree, only the input tree.
Using XML was supposed to free you of the burden of doing much of the parsing, i.e., have a consistent format for the data. However in XSLT you still have a very large parsing responsibility. They choose XML to represent the top-level transformation structures, but almost every element and attribute that has content has its own mini-language format. XML is too syntax heavy for a pure XML-only syntax, probably.
-j