XSLT: Match last text() descendant in an element

Timothy Sullivan

I am working on an XSLT (1.0) template that needs to match the last text node in an element that has a particular attribute and inject . The hard part is I need to know the last descendant, and not necessarily the last child. The schema is not consistent. Some text Some more text Some other text. SELECT ME The closest I have gotten is the following, but it selects the last text() on every element, not the last element with a text node (e3 in the example above). *[ancestor-or-self::*[@condition='optional']]/text()[last()]

Timothy Sullivan

This is what I came up with, it seems to be working.

matthew.sorenson

Are you getting caught by order of precedence? The predicate operator has a higher precedence than / and // operators. Edit 1: Unfortunately, it looks like XSL 1.0 patterns can't handle adding the parentheses to avoid the precedence issue. I tried moving the conditions into the predicate and had some success. *[ancestor-or-self::*[@condition='optional'] and text() and not(descendant::*) and not(following::*//text())] See: [
http://www.xsltcake.com/slices/vGAxJN/3][1] Edit 2: Your new input raises some interesting questions. I have a new solution that works on both of these inputs, but fails in other scenarios. - [
http://www.xsltcake.com/slices/lFUpC7][2] - [
http://www.xsltcake.com/slices/lFUpC7/2][3] Fails (I'm unsure on how to get the order of operations correct on a `[last()]` predicate to grab the last node: - [
http://www.xsltcake.com/slices/lFUpC7/3][4] (this raises a question if you would want to ignore certain text markup elements?) - [
http://www.xsltcake.com/slices/lFUpC7/4][5] It is funny how easy this seems to be with `select`: `(//text())[last()]`. On that note, what are you trying to due with this last text node? Maybe an alternative can be found. [1]:
http://www.xsltcake.com/slices/vGAxJN/3 [2]:
http://www.xsltcake.com/slices/lFUpC7 [3]:
http://www.xsltcake.com/slices/lFUpC7/2 [4]:
http://www.xsltcake.com/slices/lFUpC7/3 [5]:
http://www.xsltcake.com/slices/lFUpC7/4

Timothy Sullivan

Note this is XSL 1.0 (Added to question). I am getting an exception when trying to parse that line due to parenthesis.

matthew.sorenson

I edited my answer with the constraints of XSL 1.0 and that this needs to be a match pattern, not any valid XPath.

Timothy Sullivan

Thanks, that works, though I do have an issue now where there is a trailing

that is messing up the match.
http://www.xsltcake.com/slices/vGAxJN/4

matthew.sorenson

I have a new solution that works on both of these inputs, but fails in other scenarios.
http://www.xsltcake.com/slices/lFUpC7
http://www.xsltcake.com/slices/lFUpC7/2 Fails:
http://www.xsltcake.com/slices/lFUpC7/3 (and raises a question if you would want to ignore certain text markup elements?)
http://www.xsltcake.com/slices/lFUpC7/4 I've edited my answer as well with additional comments.

Unknown

Couldn't you use
to clean up the empty text nodes? I'd think that would take care of the
and
elements messing things up.

Unknown

I haven't tried this and it's been a long time since I looked at XSLT1.0 but would it not work if you put `*[ancestor-or-self::*[@condition='optional']]/text()` in a variable and you then selected `$var[last()]`?

Timothy Sullivan

Matt, one thing to note is following doesn't work in all cases because there will be other text outside of the sub-clause.

Timothy Sullivan

Anyone able to unprove this?
http://www.xsltcake.com/slices/vGAxJN/12

Unknown

The following seems to work for me:

Timothy Sullivan

Look at the comment chain in Matthew Sorenson's answer to see why this isn't enough.

matthew.sorenson

Something I didn't consider before is that this may take precedence over currently existing text node templates (or newly added/existing templates may override this template). For example, the special character translator and footnote reference cleaner templates also match against "text()".

Timothy Sullivan

Ultimately the issue is bad content, but they won't have a fix in time. I need to wrap certain text nodes with [] brackets. I can always call the special character translate directly if needed, but for the content the [] are much more important.