Re: [xsl] Legibility, repetition, nesting

["Imsieke, Gerrit, le-tex gerrit.imsieke@xxxxxxxxx" <xsl-list-service@xxxxxxxxxxxxxxxxxxxxxx>]

Dear David,

It's difficult to propose a solution without seeing the current code, 
but I suggest that if you can't avoid moving the nested for-eaches into 
the type-specific templates (that are invoked by next-match), you can 
encapsulate these for-eaches in a named template that you call from each 
of the specific templates. The named template will accept a function 
item as a parameter. This function item does what the specific template 
body previously did. Higher-order functions work with recent Saxon HE 
versions. This will only introduce a rather trivial redundancy: The same 
named template invocation at several places, each with a distinct 
parameter, but no replication of any complex logic.
Maybe there's a more straightforward way to pass your function items to 
the process logic, without using next-match. But being a huge proponent 
of using next-match myself [1], I would only eliminate it if it were 
really redundant, conceptually.

Gerrit

[1] https://twitter.com/gimsieke/status/581751858638749696


On 26.07.2020 22:31, David Birnbaum djbpitt@xxxxxxxxx wrote:
> Dear xsl-list,
>
> I am grateful to all of those who responded to my inquiry, below.
>
> Packaging the variables inside a map, as several readers suggested, so 
> that I have to pass only one thing when I hand off responsibility from 
> the general template (does the common stuff for all items) to the 
> type-specific ones, reduces the repetition substantially, and calling a 
> function to wrap up the variables, instead of doing it directly inside 
> the template that does the calling, as readers also suggested, improves 
> the legibility. So far, so good.
>
> I thought I had settledB on using <xsl:next-match> to transfer control to 
> a type-specific template for the type-specific processing as needed, but 
> I ran into a problem because I was invoking <xsl:next-match> inside an 
> <xsl:-for-each> (actually, two of them), and "using <xsl:for-each> makes 
> the current template rule null" (Kay, p. 400). I hadnbt known about this 
> limitation before, and if I have understood correctly, it seems to mean 
> that I cannot use <xsl:next-match> inside <xsl:for-each>. I could move 
> the entire for-each processing inside the next-match template, but then 
> I'm repeating the same for-each code and common processing inside all of 
> the next-match templates, and I'm trying to avoid repetition. Using 
> template matching, rather than calling functions, is convenient because 
> I have values for the type-specific templates like <xsl:template 
> match="item[paradigm='1*a%1% B' 9']">, that is, values that cannot be 
> used directly as XML names. Those values come from the original data, 
> and although I could build a mapping table or function to translate them 
> into XML names, it would be more legible, more self-documenting (to a 
> developer who knows the data, that is, me), more convenient, and less 
> error-prone to use them directly, as is possible withB @match values like 
> the one illustrated above.
>
> So: Is there a way I am overlooking for working around the inability to 
> use <xsl:next-match> inside <xsl:for-each>, one that would let me, 
> inside <xsl:for-each> use the original, data-driven <paradigm> values to 
> identify where the type-specific processing should happen? Ideally the 
> method would work with Saxon HE and without extension functions.
>
> Best,
>
> David
>
> On Mon, Jul 20, 2020 at 1:03 PM David Birnbaum djbpitt@xxxxxxxxx 
> <mailto:djbpitt@xxxxxxxxx> <xsl-list-service@xxxxxxxxxxxxxxxxxxxxxx 
> <mailto:xsl-list-service@xxxxxxxxxxxxxxxxxxxxxx>> wrote:
>
>     Dear XSL-list,
>
>     I write here for advice about Best Practice. The question is at the
>     end; because I am constitutionally unable to be concise, the long
>     narrative before it is context.
>
>     I'm developing an XSLT stylesheet that processes 134 different types
>     of items (same generic identifier, distinguished by anB attribute
>     value). For each item, regardless of type, I create several
>     variables, do an <xsl:copy>, and inside that first create some other
>     content that is common to all items, regardless of type, and then
>     use <xsl:choose> to handle the types differently, according to their
>     idiosyncrasies. I began by implementing this as a single template
>     with a long <xsl:choose> deeply nested inside it, which has the
>     advantage of avoiding unnecessary repetition, since the shared
>     operations are outside the <xsl:choose>. It works, but perfectionism
>     is a terrible curse ...
>
>     Perhaps I'm being arbitrarily fastidious, but the <xsl:choose>
>     inside the deep nesting feels awkward; I wind up with one template
>     that runs to more than a thousand lines, where the <xsl:choose> is
>     seven levels deep. This made me wonder whether off-loading the
>     type-specific tasks to separate templates or functions, which could
>     be called from the appropriate place, would keep the main template
>     down to a more manageable size. Specifically, I'd like to be able to
>     put the code blocks currently inside of the <xsl:when> statements
>     somewhere other than deep inside a single main template.
>
>     One implementation of this approach that works, but comes with its
>     own issues, is using an <xsl:next-match> with
>     auxiliaryB lower-priority templates that match item[@type eq 'x'].
>     This lets me break out the type-specific code into separate
>     templates. The reason this is not wholly satisfactory is that I have
>     to pass all of the variables into these separate templates as
>     parameters, so I wind up repeating the same <xsl:param> statements
>     inside each of the secondary templates. That much repetition feels
>     suboptimal.
>
>     The only approach that occurs to me that might simultaneously
>     eliminate repetition and avoid putting all of the processing inside
>     a single thousand-line template, most of which is the <xsl:choose>
>     with all of the type-specific handling inside <xsl:when> children,
>     is to put the type-specific processing into separate files, with an
>     <xsl:when> root, and then <xsl:include> them inside the
>     <xsl:choose>. One downside seems to be that they will not be valid
>     XSLT (they wonbt have the necessary wrapper boilerplate and the
>     variables they use wonbt be defined inside them), which I think I
>     could overcome by using an <oXygen/> "master document", which would
>     cause them to be validated in context. That isn't ideal, since it's
>     tied to a specific development environment, but since that happens
>     to be my usual development environment, the objection is
>     philosophical (= can be ignored in the interest of Getting The Job
>     Done), rather than practical.
>
>     So: Is there a Best Practice approach to breaking out the
>     type-specific treatment of the different types of items that avoids
>     both 1) unnecessary repetition and 2) embedding a single
>     thousand-line <xsl:choose>, which contains all of the type-specific
>     operations, seven levels deep inside a template?
>
>     Thanks,
>
>     David