can't add mode="replace" to attribute class

[emylonas]

Trying to make changes for TEIC/TEI#1867. This works for the one element, <constraintSpec> which has@mode="replace" and '@mode="change" seems to work for several other elements. but compile fails when new class is added and attribute definition is given `mode="replace".

[sydb]

That is, specifying mode="replace" on an <attDef ident="type"> inside a <classSpec> when the class itself is also a member of att.typed (and thus already had @type) causes a build failure:

/var/tei/P5/p5odds-examples.rng:28131:27: error: duplicate attribute "type"

It shouldn’t.

[martindholmes]

@sydb and @martindholmes have narrowed down the problem to what's happening in the odds/classatts.xsl file, where some clever processing takes place to handle cases where attDefs in elementSpecs override those inherited from classSpecs of which the elementSpec is a member. This process is not done for classSpecs (yet). Simply editing the match attribute to cause classSpecs to be processed as well as elementSpecs fails because it generates another kind of error, whereby nested attribute classes end up duplicated on elements, so that simple fix doesn't work. However we think that it is possible to fix this by passing classSpecs through the same process, with some modification to the process.

[martindholmes]

@sydb and @martindholmes went back through the commits and tested the build, and determined that the problem was introduced by the merge at #343; that particular merge contains lots of individual commits, but only one file seems to have been changed:

https://github.com/TEIC/Stylesheets/pull/343/files

[sydb]

Stylesheets group comes up w/ the following replacement for a template in odds/odd2odd.xsl:

  <xsl:template match="tei:elementSpec[@mode eq 'change']
                      |tei:classSpec[@mode eq 'change']
                      |tei:macroSpec[@mode eq 'change']
                      |tei:dataSpec[@mode eq 'change']"
                mode="pass0">
    <xsl:variable name="myGI" select="local-name(.)"/>
    <xsl:variable name="myIDENT" select="@ident"/>
    <xsl:variable name="odd2odd-CHANGE-key-replacement"
      select="ancestor::tei:schemaSpec//
      (tei:elementSpec|tei:classSpec|tei:macroSpec|tei:dataSpec)
      [ @mode eq 'change']
      [ $myGI eq local-name(.) ]
      [ @ident eq current()/@ident ]"/>
    <xsl:choose>
      <xsl:when test="count( $odd2odd-CHANGE-key-replacement ) gt 1">
        <xsl:if test=". is $odd2odd-CHANGE-key-replacement[@ident eq $myIDENT][1]">
          <xsl:copy>
            <xsl:copy-of select="@*"/>
            <xsl:for-each select="$odd2odd-CHANGE-key-replacement[ @ident eq $myIDENT ]">
              <xsl:apply-templates select="node()" mode="pass0"/>
            </xsl:for-each>
          </xsl:copy>
        </xsl:if>
      </xsl:when>
      <xsl:otherwise>
        <xsl:copy>
          <xsl:copy-of select="@*"/>
          <xsl:apply-templates select="*|text()|comment()|processing-instruction()" mode="pass0"/>
        </xsl:copy>
      </xsl:otherwise>
    </xsl:choose>
  </xsl:template>

[martindholmes]

The above function didn't actually work, but subsequent tests show that the problem with simplePrint is that it consists initially only of specGrpRefs, and those refs are not expanded early enough in the process. So we need a pass before pass0 which will expand all specGrpRefs.

[ebeshero]

Quick summary of where we stopped the meeting: We were constructing a variable as a document-node() with guidance from the Kay book, to store the expanded specGrpRefs and pass them to the key() definition as its third argument, which defines its context and must be a document-node. This will, we hope, give us a means to do a pass before pass0!

[sydb]

Do tell, @ebeshero, y’all are talking about using the XSLT 2.0 <xsl:document> instruction, yes? That’s where my head is going right now. I played with it — dividing pass0 into two passes — for perhaps half an hour this morning. But I am no longer convinced this is a stylesheets bug!
I think it may be a tei_simplePrint problem. Namely, that it (P5/Exemplars/tei_simplePrint.odd) has two <classSpec type="atts" ident="att.global.rendition" mode="change"> elements. Is that supposed to be allowed?
My instinct is “yes, someday we would love it if multiple mode=change declarations worked, but it doesn’t now and hasn’t for the last decade, so use only one.” Well, at least that’s what I vaguely recall saying in workshops. But it was working just a few months ago, eh? Or was that second <classSpec> added? @martindholmes do you remember? If not, I will try to look that up again later today.
Anyway, if validated against P5/Exemplars/tei_customization.isosch, this duplicate <classSpec> is not caught, but 4 duplicate <elementSpec>s are. (As well as 4 duplicate @xml:id values, and a bunch of stuff that should not really be a problem.) The error messages are “Current ODD processors will not correctly handle more than one ＜elementSpec＞ with the same @Ident” on lines 3430, 4646, 4982, and 5113.

[sydb]

Oh, but that’s silly, isn’t it — this ticket is about mode="replace" in attribute classes other than simplePrint. We were just using simplePrint as a test case yesterday. (Possibly a red-herring test case, though.)

[martindholmes]

We should definitely fix simplePrint, and add some Schematron to prevent this duplication. It makes no sense in a single ODD file.

[martindholmes]

@sydb and @martindholmes have added fixes (TEIC/TEI/#6589fca) (along with @lb42 ) to simplePrint to avoid cases where there are multiple *Spec elements with the same @ident. This should fix the build process for the moment.

[lb42]

This change does cause the make to fail a little further on: something is apparently going wrong in the generation of the -examples.rng file, or the nvdl file which calls it. I can't reproduce the problem using the current "bleeding edge" framework in oXygen, which obstinately goes on doing the right thing, so I am still puzzled.

BUILD SUCCESSFUL
Total time: 10 seconds
../run-onvdl tei_simplePrint.nvdl tei_simplePrint.odd
/usr/bin/onvdl
/home/lou/Public/TEI/P5/Exemplars/tei_simplePrint.odd:606:79: error: attribute "rhyme" not allowed at this point; ignored
/home/lou/Public/TEI/P5/Exemplars/tei_simplePrint.odd:700:48: error: attribute "part" not allowed at this point; ignored