Re: [xsl] Matching string values across element boundaries

[Michael Sokolov <msokolov@xxxxxxxxxxxxxxxxxxxxx>]

David - I think the answer whether there is any improvement to be made 
to your system will depend in detail on just how the matching algorithm 
works.  Clearly if it expects a string, you have to give it one, and you 
are left with something like your approach.  If you're willing to 
revisit the matching algorithm (I expect you don't want to - it sounds 
hairy), you could probably also change the markup generation.  One idea 
that springs to mind is the highlighters available in search platforms: 
these typically operate on text only, remembering the position of every 
word, and allow you to mark them with tags in a highlighting pass, which 
you can later coalesce using XSLT or some other markup-aware process.  
If you can cast the matching problem as a search problem, you could 
leverage MarkLogic, or Lucene or something like that.  Maybe that would 
be better than what you have, I don't know.

-Mike

On 4/8/2013 2:15 PM, David Sewell wrote:
> I expect this has been discussed here before, but I can't locate any relevant
> discussion, so here goes.
>
> We have input data with many unmarked short-title citations that look like this:
>
>     Sprague, <hi rend="italic">Braintree Families</hi>
>
> We want to wrap them inside another element, in our case a <ref> to the
> bibliographic expansion. We have a venerable chain of XSLT 2.0 transforms that
> does this, and pretty well, by preprocessing the data to convert all those <hi>
> tags into a pair of unique ASCII characters, so that we can do string-matching
> operations within a single text node that now includes something like
>
>     Sprague, "Braintree Families%
>
> which is easy to handle with xsl:analyze-string. then once we've wrapped all the
> strings we need to, we post-process with xsl:analyze-string to put the <hi>
> elements back in.
>
> In practice, given the proper regexes, this works quite well and provides the
> desired output, but I always feel a bit guilty about the hackishness of the
> approach. Given that the citations are quite variable in structure (usually but
> not always containing <hi> elements, with various combinations of text nodes at
> start and end), I've never come up with a good general-purpose way to operate
> purely on elements and text nodes without the convert-tags-to-characters step.
> Is there one (or more)?
>
> David S.