Re: All teidata.pointer attributes

Thanks so much for this Syd. It is certainly more involved than I first 
thought.

I had been looking at the recursive code that generates the Attributes 
section of an element's description. That wasn't getting me anywhere 
fast. Then I did a quick inventory and realized the great majority of 
pointer attributes are such across all elements. So I just manually 
typed those in, handled the exceptions brute-force, and moved on.

What I have now, then, is fragile. Any schema changes will require 
manually tweaking my XSLT. When I get back to generalizing, I'll come 
back to what you have here—or maybe that wiki article. ;-)

Thanks again!
John


On 2/14/2017 2:58 AM, Syd Bauman wrote:
> John --
>
> Sorry it has taken so long for me to get to posting this. Was off at
> the TEI Council meeting, a one day TEI conference, and a 3-day XML
> conference.
>
> Lou is indeed correct, to do this properly you need to chase class
> references. (Because not only might a pointer attribute be a member
> of a class, that class might be a member of a class.) You can do it
> using any of the following as the base from which you figure out
> which element & attribute pairs are pointers.
>
>   * Your customization ODD -- very hard, as you have to
>     "compile" the ODD by merging it with the P5 ODD first;
>
>   * Your compiled ODD; or
>
>   * Your RELAX NG schema.
>
> No matter which base you choose, with XSLT 1 or XSLT 2 you have to
> use a 2-step process. (I think you can do it in 1 step in XSLT 3, but
> I'm not sure as I've never tried. :-)
>
> Step 1 reads in the compiled ODD or the RELAX NG schema and writes
> out an XSLT stylesheet. Step 2 runs that XSLT stylesheet using your
> document instance as input. The output is (hopefully) exactly what
> you want: a copy of that document instance decorated with attributes
> that say "this element has a pointer attribute".
>
> I have written sample implementations of step 1 for both of those
> inputs. You can find them at
>    http://paramedic.wwp.neu.edu/~syd/temp/TEI/JPM_generate_ptr_attr_flagger_from_ODD.xslt
> and
>    http://paramedic.wwp.neu.edu/~syd/temp/TEI/JPM_generate_ptr_attr_flagger_from_RNG.xslt
>
> IIRC Lou has already explained how to get your hands on the RNG;[1]
> and generating the compiled ODD can be a bit tricky, so I suspect
> you're going to prefer the latter.[2] But Peter insisted I write the
> former, too, as it is somewhat more appropriate and cleaner. (The RNG
> version, e.g., relies on the fact that TEI attribute class names
> start with "att.". So it's somewhat more fragile. On the other hand,
> I don't think anyone wants to change that naming convention. :-)
>
> Caveats:
>   * I have not tested these carefully
>   * These are not necessarily the "right" or "best" way to do this,
>     nor particularly good examples of XSLT.[3]
>
> If someone has a better way to do this, or can suggest improvements
> to my XSLT, I'm all ears. If there is community interest I could
> spiff these up a bit and put them up on the TEI wiki.
>
> Anyway, there are a variety of ways of combining those two steps into
> one, the most obvious of which is to use XProc. (But one could also
> use a shell script, ant, or make.)
>
> Notes
> -----
>   [1] Given that you use oXygen, run the "TEI ODD to RelaxNG XML"
>       transformation scenario using your customization ODD as the input
>       document. (If you don't see the various ODD transformations when
>       you click "Configure Transformation Scenario(s)" (cmd-shift-C or
>       ctl-shift-C), remember to choose "Show all scenarios" from the
>       little blue gear in the upper right corner of the dialog box.) At
>       the moment the output is put into a directory called out/ that is
>       in the same directory as the input.
>       
>   [2] Furthermore, I've just discovered a minor problem with use of the
>       compiled ODD, which some may consider a bug. The declarations for
>       at least some elements that are never used are still in the ODD.
>       (E.g., the specification of the <typeNote> element is still in
>       the compiled "tei_drama.odd.odd" file, even though there is no
>       <typeDesc> element, so <typeNote> can never occur.) They have
>       been removed by the time you get to the RELAX NG. Worth noting
>       that these won't change your output -- you just end up trying to
>       match an element/attribute combination (e.g.,
>       typeNote/@scribeRef) that can never be there. Oh well, there's
>       another 18 microseconds of your life you will never get back. :-)
>
>   [3] E.g., I suspect that in the RNG version, the template that
>       matches "data" in "class" mode and the template named
>       "chase-attr-class" could be combined into a single more readable
>       template. And in that "data" mode "class" template it might be
>       clearer to test for something to return by checking count($all)
>       rather than the length of the joined string.