piece1.tex does not work at all in latexml

[kohlhase]

I am trying to get symdefs for slides/missing/piece1.tex to work, but I cannot get them through latexml.

[deyan]

It converts for me, it just chokes on the "&" alignment characters. I don't think it is legal to use them in symdefs like this, did you mean to escape them, or?

[deyan]

As we discussed, we probably need to rework the piecewise treatments of LaTeX's array environment, so that we can properly write down the semantics.

• \arrayline - marks up a line of the array, equates to a table row on the XML side.
• \arraycell - marks up a cell of the array, equates to a table cell on the XML side.

This approach allows \symdef{} definitions to become more reusable, as they can be plugged easily into arbitrary arrays. It also allows for defining array skeletons that one can instantiate with different cells. On the TeX side, we probably want to use \futurelet to enable parsing a sequence of arbitrarily many \arraycell commands (or, instead, look for exactly two in the {array} environment, or as many as have been defined in a matrix or table case).

We should probably extend this markup to tables and matrices in the future.

[kohlhase]

I have added the LaTeX side of this to the presentation package. Actually this was very simple. But the documentation took quite some space (see section 2.7 in trunk/sty/presentation/presentation.pdf. Deyan, please re-read this and correct any problems. I will now change missing/en/piece1.tex to the new primitives, so you have a test case. The ball is in your field now.

[kohlhase]

we should probably think about other layout primitives as well. Maybe we can fix the symdefs with tikzinside this way as well.

[deyan]

In order to make the renderings work, we need to allow <XMCell> and <XMRow> to appear in <XMath>, whenever there is a <rendering ancestor>.

Once that is done I can test the conversion and see if LaTeXML behaves properly.

[kohlhase]

what is the content model of XMCell? I guess that of XMRow is XMCell+

[deyan]

The content mdoel of XMCell is XMath.class.

 http://dlmf.nist.gov/LaTeXML/manual/schema/schema.module.LaTeXML-math.xhtml#schema.element.XMCell

[kohlhase]

Could you post a couple of XML trees that need to be valid? Currently I do not really see what is needed in the schema.

[deyan]

• The rendering for:

   \symdef{piece}[2]{\arrayline{\arraycell{#1}}{\text{if}\;#2}}
  

  is:

        <rendering>
          <Math xmlns="http://dlmf.nist.gov/LaTeXML" xml:id="piece1.m1" text="[arg:1ifarg:2]">
            <XMath>
              <XMRow>
                <XMCell>
                  <XMTok role="ID">arg:1</XMTok>
                </XMCell>
                <XMCell>
                  <XMText>
                    <text>if</text>
                  </XMText>
                  <XMHint name="thickspace"/>
                  <XMTok role="ID">arg:2</XMTok>
                </XMCell>
              </XMRow>
            </XMath>
          </Math>
        </rendering>
  

---

• And the rendering for:

   \symdef{otherwise}[1]{\arrayline{\arraycell{#1}}{\text{else}}}
  

  is:

        <rendering>
          <Math xmlns="http://dlmf.nist.gov/LaTeXML" xml:id="piece1.m2" text="[arg:1else]">
            <XMath>
              <XMRow>
                <XMCell>
                  <XMTok role="ID">arg:1</XMTok>
                </XMCell>
                <XMCell>
                  <XMText>
                    <text>else</text>
                  </XMText>
                </XMCell>
              </XMRow>
            </XMath>
          </Math>
        </rendering>
  

[deyan]

The schema breaks when you ask for an <XMRow> in <XMath>. Same happens if you ask for <XMCell> in <XMath>. If those two are allowed, one can use \arrayline and \arraycell throughout \symdef commands without any problems.

[kohlhase]

I think we fixed this, @Deyan, please test

[deyan]

The XMath side of things is now working smoothly.

However, be reminded that we have no translations for <XMArray>, <XMRow> and <XMCell> in the OpenMath? post processor of LaTeXML. We need to think about how to represent arrays in OpenMath? (or you need to tell me, if this is known).

[kohlhase]

I do not think that there is any way to represent arrays in OpenMath?, and there should not be one as they are purely presentational devices. This should just return an error (in the XML tradition with severe error handling).

[kohlhase]

Note for this that we only use these XMRow and XMCell in the renderings and thus for the pmml part of latexmlpost. See also  https://trac.mathweb.org/LaTeXML/ticket/1514 for a discussion.

[deyan]

I think I have handled all problems that were popping up here. You can update&reinstall LaTeXML's branch and take a look.

Please close if things look ok, let me know otherwise.

[kohlhase]

note quite: In missing/en/piece1.tex I have

\begin{omtext}[title=Testing piecewise]
  $|x|=\piecewise{\piece{x}{x>0}\piece{-x}{x<0}\otherwise{0}}$
\end{omtext}

and that gives me

         <om:OMOBJ>
            <om:OMA>
              <om:OMS cd="relation1" name="eq"/>
              <om:OMA>
                <om:OMS cd="latexml" name="absolute-value"/>
                <om:OMV name="x"/>
              </om:OMA>
              <om:OMA>
                <om:OMS cd="piece1" name="piecewise"/>
		<om:OMA>
		  <om:OMS cd="arith1" name="times"/>
		  <om:OMA>
                    <om:OMS cd="piece1" name="piece"/>
                    <om:OMV name="x"/>
                    <om:OMA>
                      <om:OMS cd="relation1" name="gt"/>
                      <om:OMV name="x"/>
                      <om:OMI>0</om:OMI>
                    </om:OMA>
                  </om:OMA>
                  <om:OMA>
                    <om:OMS cd="piece1" name="piece"/>
                    <om:OMA>
                      <om:OMS cd="arith1" name="minus"/>
                      <om:OMV name="x"/>
                    </om:OMA>
                    <om:OMA>
                      <om:OMS cd="relation1" name="lt"/>
                      <om:OMV name="x"/>
                      <om:OMI>0</om:OMI>
                    </om:OMA>
                  </om:OMA>
                  <om:OMA>
                    <om:OMS cd="piece1" name="otherwise"/>
                    <om:OMI>0</om:OMI>
                  </om:OMA>
                </om:OMA>
              </om:OMA>
            </om:OMA>
          </om:OMOBJ>

Note that there is a spurious application of times. I looked and found that also in situations, where we have purely sTeX markup, so it must be an artefact of piecewise.

[deyan]

We need to make the schema more flexible so that it tolerates our hackery (<render> elements get passed in the form of <XMTok> elements during the .tex.xml XMath stage), namely:

• Allow <XMTok> in <XMRow> and <XMArray>
• Allow <render> in <m:tr> and <m:table>

This will be enough to get the renderings looking as expected, but we should be careful, as we might break the auto-open and auto-close functionality of arrays... Could the schema be restricted to allow the two above, only when having a <rendering> ancestor? Then we would be safe.

If not, this might force us to reconsider using <XMTok> as a <render> carrier, since <XMArray> and <XMRow> are extremely strict about their content ("proper" tables only allowed).

It seems a bit tricky. Also, I just committed the \arrayline macro in presentation.dtx, sorry about that.

[kohlhase]

we should take this up again after the rewrite, or is this fixed already?

[kohlhase]

there is another instance of a problem with \arrayline in cfl/en/BNF.tex, we should look at this as well.

[deyan]

We looked at the example and we reconfirmed we need to allow <render> in <m:tr> and <m:table>, and keep it allowed in <m:td>.

Respectively, allow <XMTok> in <XMRow> and <XMArray>

[kohlhase]

I have updated the schema, and it now allows all of these, but it still generates spurious XMRow and XMCell in the notation declaration for piecewise. We should look at this together.