In what order do templates in an XSLT document execute, and do they match on the source XML or the buffered output

xslt

Here is something that has always mystified me about XSLT:

  1. In what order do the templates execute, and
  2. When they execute, do they match on (a) the original source XML, or (b) the current output of the XSLT to that point?

Example:

<person>
  <firstName>Deane</firstName>
  <lastName>Barker</lastName>
</person>

Here is a fragment of XSLT:

<!-- Template #1 -->
<xsl:template match="/">
  <xsl:value-of select="firstName"/> <xsl:value-of select="lastName"/>
</xsl:template>

<!-- Template #2 -->
<xsl:template match="/person/firstName">
  First Name: <xsl:value-of select="firstName"/>
</xsl:template>

Two questions about this:

  1. I am assuming that Template #1 will execute first. I don't know why I assume this — is it just because it appears first in the document?
  2. Will Template #2 execute? It matches a node in the source XML, but by the time the we get to this template (assuming it runs second), the "firstName" node will not be in the output tree.

So, are "later" templates beholden to what has occurred in "earlier" templates, or do they operate on the source document, oblivious to what has been transformed "prior" to them? (All those words are in quotes, because I find it hard to discuss time-based issues when I really have little idea how template order is determined in the first place…)

In the above example, we have a template that matches on the root node ("/") that — when it is done executing — has essentially removed all nodes from the output. This being the case, would this pre-empt all other templates from executing since there is nothing to match on after that first template is complete?

To this point, I've been concerned with later templates not executing because the nodes they have operated on do not appear in the output, but what about the inverse? Can an "earlier" template create a node that a "later" template can do something with?

On the same XML as above, consider this XSL:

<!-- Template #1 -->
<xsl:template match="/">
  <fullName>
    <xsl:value-of select="firstName"/> <xsl:value-of select="lastName"/>
  </fullName>
</xsl:template>

<!-- Template #2 -->
<xsl:template match="//fullName">
  Full Name: <xsl:value-of select="."/>
</xsl:template>

Template #1 creates a new node called "fullName". Template #2 matches on that same node. Will Template #2 execute because the "fullName" node exists in the output by the time we get around to Template #2?

I realize that I'm deeply ignorant about the "zen" of XSLT. To date, my stylesheets have consisted of a template matching the root node, then are completely procedural from there. I'm tired of doing this. I would rather actually understand XSLT correctly, hence my question.

Best Solution

I love your question. You're very articulate about what you do not yet understand. You just need something to tie things together. My recommendation is that you read "How XSLT Works", a chapter I wrote to address exactly the questions you're asking. I'd love to hear if it ties things together for you.

Less formally, I'll take a stab at answering each of your questions.

  1. In what order do the templates execute, and
  2. When they execute, do they match on (a) the original source XML, or (b) the current output of the XSLT to that point?

At any given point in XSLT processing, there are, in a sense, two contexts, which you identify as (a) and (b): where you are in the source tree, and where you are in the result tree. Where you are in the source tree is called the current node. It can change and jump all around the source tree, as you choose arbitrary sets of nodes to process using XPath. However, conceptually, you never "jump around" the result tree in the same way. The XSLT processor constructs it in an orderly fashion; first it creates the root node of the result tree; then it adds children, building the result in document order (depth-first). [Your post motivates me to pick up my software visualization for XSLT experiments again...]

The order of template rules in a stylesheet never matters. You can't tell, just by looking at the stylesheet, in what order the template rules will be instantiated, how many times a rule will be instantiated, or even whether it will be at all. (match="/" is an exception; you can always know that it will get triggered.)

I am assuming that Template #1 will execute first. I don't know why I assume this -- is it just because it appears first in the document?

Nope. It would be called first even if you put it last in the document. Template rule order never matters (except under an error condition when you have more than one template rule with the same priority matching the same node; even then, it's optional for the implementor and you should never rely on such behavior). It gets called first because the first thing that always happens whenever you run an XSLT processor is a virtual call to <xsl:apply-templates select="/"/> . The one virtual call constructs the entire result tree. Nothing happens outside it. You get to customize, or "configure", the behavior of that instruction by defining template rules.

Will Template #2 execute? It matches a node in the source XML, but by the time the we get to this template (assuming it runs second), the "firstName" node will not be in the output tree.

Template #2 (nor any other template rules) will never get triggered unless you have an <xsl:apply-templates/> call somewhere in the match="/" rule. If you don't have any, then no template rules other than match="/" will get triggered. Think of it this way: for a template rule to get triggered, it can't just match a node in the input. It has to match a node that you elect to process (using <xsl:apply-templates/>). Conversely, it will continue to match the node as many times as you choose to process it.

Would [the match="/" template] pre-empt all other templates from executing since there is nothing to match on after that first template is complete?

That rule preempts the rest by nowhere including <xsl:apply-templates/> in it. There are still plenty of nodes that could be processed in the source tree. They're always all there, ripe for the picking; process each one as many times as you want. But the only way to process them using template rules is to call <xsl:apply-templates/>.

To this point, I've been concerned with later templates not executing because the nodes they have operated on do not appear in the output, but what about the inverse? Can an "earlier" template create a node that a "later" template can do something with?

It's not that an "earlier" template creates a new node to be processed; it's that an "earlier" template in turn processes more nodes from the source tree, using that same instruction (<xsl:apply-templates). You can think of it as calling the same "function" recursively, with different parameters each time (the nodes to process as determined by the context and the select attribute).

In the end, what you get is a tree-structured stack of recursive calls to the same "function" (<xsl:apply-templates>). And this tree structure is isomorphic to your actual result. Not everyone realizes this or has thought about it this way; that's because we don't have any effective visualization tools...yet.

Template #1 creates a new node called "fullName". Template #2 matches on that same node. Will Template #2 execute because the "fullName" node exists in the output by the time we get around to Template #2?

Nope. The only way to do a chain of processing is to explicitly set it up that way. Create a variable, e.g., $tempTree, that contains the new <fullName> element and then process it, like this <xsl:apply-templates select="$tempTree">. To do this in XSLT 1.0, you need to wrap the variable reference with an extension function (e.g., exsl:node-set()), but in XSLT 2.0 it will work just as is.

Whether you're processing nodes from the original source tree or in a temporary tree that you construct, either way you need to explicitly say what nodes you want to process.

What we haven't covered is how XSLT gets all its implicit behavior. You must also understand the built-in template rules. I write stylesheets all the time that don't even include an explicit rule for the root node (match="/"). Instead, I rely on the built-in rule for root nodes (apply templates to children), which is the same as the built-in rule for element nodes. Thus I can ignore large parts of the input, let the XSLT processor automatically traverse it, and only when it comes across a node I'm interested in will I do something special. Or I could write a single rule that copies everything recursively (called the identity transform), overriding it only where necessary, to make incremental changes to the input. After you've read "How XSLT Works", your next assignment is to look up the "identity transform".

I realize that I'm deeply ignorant about the "zen" of XSLT. To date, my stylesheets have consisted of a template matching the root node, then are completely procedural from there. I'm tired of doing this. I would rather actually understand XSLT correctly, hence my question.

I applaud you. Now it's time to take the "red pill": read "How XSLT Works"