Errors when using \\ (new linebreak from v4) in siunitx extension

[limefrogyank]

Issue Summary

I don't know if this is a bug in MathJax or my own extension yet, but it only happens when I use the \\ newline macro. I converted my extension to MathJax v4 and got it to build without errors. I have a demo page that uses many different iterations of the siunitx macros and all of them worked except some examples that used the \displaylines macro. An example would be:

\displaylines{
    \sisetup{fill-angle-degrees}
    \ang{-1;;} \\
    \ang{;-2;} \\
    \ang{;;-3} \\
    \sisetup{fill-angle-degrees = false}
}

Generates:

This example can be seen halfway down the page here: https://limefrogyank.github.io/siunitx-pcc/index.html

It looks like the Mml generator tried to add <mtd> table elements inside the "equation" that needed to be generated instead of just on the outside.

When I tried exploring using the new \\ macro and just removed the \displaylines section, I started getting some really weird results. Sometimes, it works just fine. Sometimes, it would stall loading for maybe 30+ seconds. Sometimes, it would process quickly but give odd results where an mml node seemed to have been inserted out of place. For example, $\ang[fill-angle-seconds]{;-2;} \\ \ang[fill-angle-degrees]{;3;} \\$ generates:

when it should look more like:

-2' 0"
0° 3' 

Previous example here: https://limefrogyank.github.io/siunitx-pcc/test.html

I've inspected the MmlNode object that is created after macro call and it seems to look fine. My best guess is that something inside the \\ macro is rearranging everything.

Steps to Reproduce:

You can use the demo pages I linked.
https://limefrogyank.github.io/siunitx-pcc/test.html
https://limefrogyank.github.io/siunitx-pcc/index.html

They are using code from the v4upgrade branch, not the main branch, in case you feel like inspecting my code. The relevant angle code is here: https://github.com/limefrogyank/siunitx-pcc/blob/v4upgrade/ts/angMethods.ts

Technical details:

• MathJax Version: 4.1.1
• Client OS: Windows
• Browser: Edge 146.0.3856.109

I am using the following MathJax configuration:

 MathJax = {
        loader: {
            load: ['[custom]/siunitx.js', '[tex]/gensymb', "[tex]/html","[tex]/color",], //gensymb is required for \degree and \ohm
            paths: { custom: './browser/' }
        },
        tex: {
            packages: { '[+]': ['siunitx', 'gensymb', 'html', "color"] },  //gensymb is required for \degree and \ohm
            inlineMath: { '[+]': [['$', '$']] }
            /* global options go here, see IOptions in options.ts for properties */
            // Uncomment example below: 
            // perMode: 'fraction'
        }
    };

and loading MathJax via

<script type="text/javascript" id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@4/tex-svg.js">

Lastly, there are other console errors, but I can't pursue them yet.

Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
Tracking Prevention blocked access to storage for <URL>.
tex-svg.js:1 No version information available for component [custom]/siunitx.js
(anonymous) @ tex-svg.js:1
/favicon.ico:1  Failed to load resource: the server responded with a status of 404 ()
tex-svg.js:1 Speech generation error: Cannot read properties of undefined (reading 'getAttribute')
speechError @ tex-svg.js:1

[limefrogyank]

The code infinite loops and bad formatting has been "fixed". It was an issue with 'mrow' vs 'inferredMrow'. However, the linebreaks now don't do anything at all.

https://docs.mathjax.org/en/latest/upgrading/whats-new-4.0/linebreaking.html#line-breaking-macros-in-tex

According to the docs, it only breaks on an operator mo, but will insert one if none exist in that location. I don't see any insertion...

[dpvc]

the linebreaks now don't do anything at all.

They work for me. I see this:

According to the docs, it only breaks on an operator mo, but will insert one if none exist in that location.

That is talking about the \goodbreak, \badbreak, \nobreak, and \break macros (and \\ when it appears outside of a table layout macro or environment). The \displaylines macro creates a table where the \\ ends each line, so the documentation you cite doesn't apply here. This produces a table, and so doesn't need an mo for line breaks.

This is what your first expression above produces:

<math xmlns="http://www.w3.org/1998/Math/MathML">
  <mtable rowspacing=".5em" columnspacing="1em" displaystyle="true">
    <mtr>
      <mtd>
        <mrow fake="argument">
          <mrow>
            <mo rspace="0em" lspace="0em">-</mo>
            <mn>1</mn>
          </mrow>
          <mrow data-mjx-texclass="ORD" class="MathML-Unit">
            <mi mathvariant="normal" class="MathML-Unit">&#xB0;</mi>
          </mrow>
        </mrow>
      </mtd>
    </mtr>
    <mtr>
      <mtd>
        <mrow fake="argument">
          <mrow>
            <mo rspace="0em" lspace="0em">-</mo>
            <mn>0</mn>
          </mrow>
          <mrow data-mjx-texclass="ORD" class="MathML-Unit">
            <mi mathvariant="normal" class="MathML-Unit">&#xB0;</mi>
          </mrow>
          <mrow>
            <mn>2</mn>
          </mrow>
          <mrow data-mjx-texclass="ORD" class="MathML-Unit">
            <msup aria-label="degree-minutes">
              <mi></mi>
              <mi mathvariant="normal" class="MathML-Unit">&#x2032;</mi>
            </msup>
          </mrow>
        </mrow>
      </mtd>
    </mtr>
    <mtr>
      <mtd>
        <mrow fake="argument">
          <mrow>
            <mo rspace="0em" lspace="0em">-</mo>
            <mn>0</mn>
          </mrow>
          <mrow data-mjx-texclass="ORD" class="MathML-Unit">
            <mi mathvariant="normal" class="MathML-Unit">&#xB0;</mi>
          </mrow>
          <mrow>
            <mn>3</mn>
          </mrow>
          <mrow data-mjx-texclass="ORD" class="MathML-Unit">
            <msup aria-label="degree-seconds">
              <mi></mi>
              <mi mathvariant="normal" class="MathML-Unit">&#x2033;</mi>
            </msup>
          </mrow>
        </mrow>
      </mtd>
    </mtr>
  </mtable>
</math>

Note, however, that <mrow fake="argument"> is not valid MathML, as fake is not a valid attribute. You should use data-fake or better data-siunitx-fake so that the MathML will validate, and to help avoid conflicts with other uses of data attributes.

As for the mrow versus inferredMrow your original was valid MathML. The problem was occurring due to semantic enrichment that is part of the speech generation (this is on by default in v4, but was an opt-in in v3). The extra mrow does seem to void the problem, however.

[dpvc]

@zorkow, the following table causes a problem with semantic enrichment:

<math xmlns="http://www.w3.org/1998/Math/MathML">
  <mtable rowspacing=".5em" columnspacing="1em" displaystyle="true">
    <mtr>
      <mtd>
          <mrow>
            <mn>0</mn>
          </mrow>
          <mrow data-mjx-texclass="ORD" class="MathML-Unit">
            <mi mathvariant="normal" class="MathML-Unit">&#xB0;</mi>
          </mrow>
          <mrow>
            <mn>2</mn>
          </mrow>
          <mrow data-mjx-texclass="ORD" class="MathML-Unit">
            <msup aria-label="degree-minutes">
              <mi></mi>
              <mi class="MathML-Unit">&#x2032;</mi>
            </msup>
          </mrow>
      </mtd>
    </mtr>
  </mtable>
</math>

It seems to have to do with the empty base for the prime symbol, as well as being tied to the fact that there are two pieces: 0 degrees and the 2 minutes.

[dpvc]

@limefrogyank, as a side note, the <mi class="MathML-Unit">&#x2032;</mi> will render better as an mo rather than an mi, and you won't need the mathvariant.

[limefrogyank]

Thanks for the clues for looking in the right place. First, I apologize for altering the test.html page after posting the issue. It wasn't showing the error after all. The test page was supposed to show the newline macro without the displaylines macro.

In any case, I believe I tracked it down to the cause. It turns out I was creating a TeX parser mapping for \\degreeminute and \\degreesecond. For reasons I don't remember, I was generating MathML that put the unicode prime and doubleprime inside an msup element after wrapping it with the mi. Honestly, I don't know why I did that. Perhaps one of the degree symbols looked like that at one point.

I ended up removing the msup and it looks really good now. The ClearSpeak sounds good as well. I don't see those errors in the console anymore. The new MathML generated for this TeX $\ang[fill-angle-degrees]{;-2;} \\ \ang[fill-angle-seconds]{;3;} \\$ looks like below.

<math xmlns="http://www.w3.org/1998/Math/MathML">
  <mrow>
    <mrow>
      <mo rspace="0em" lspace="0em">-</mo>
      <mn>0</mn>
    </mrow>
    <mrow data-mjx-texclass="ORD" class="MathML-Unit">
      <mi mathvariant="normal" class="MathML-Unit">&#xB0;</mi>
    </mrow>
    <mrow>
      <mrow>
        <mrow>
          <mn>2</mn>
        </mrow>
        <mo>&#x2062;</mo>
        <mrow data-mjx-texclass="ORD" class="MathML-Unit">
          <mi mathvariant="normal" aria-label="degree-minutes" class="MathML-Unit">&#x2032;</mi>
        </mrow>
      </mrow>
      <mspace linebreak="newline"></mspace>
      <mo>&#x2062;</mo>
      <mrow>
        <mrow>
          <mn>3</mn>
        </mrow>
        <mo>&#x2062;</mo>
        <mrow data-mjx-texclass="ORD" class="MathML-Unit">
          <mi mathvariant="normal" aria-label="degree-minutes" class="MathML-Unit">&#x2032;</mi>
        </mrow>
      </mrow>
      <mo>&#x2062;</mo>
      <mrow>
        <mrow>
          <mn>0</mn>
        </mrow>
        <mo>&#x2062;</mo>
        <mrow data-mjx-texclass="ORD" class="MathML-Unit">
          <mi mathvariant="normal" aria-label="degree-seconds" class="MathML-Unit">&#x2033;</mi>
        </mrow>
      </mrow>
    </mrow>
  </mrow>
  <mspace linebreak="newline"></mspace>
</math>

I might end up reverting those preferredMrow to mrow changes I made earlier, too.

BTW, how might one add versioning to the extension directly to get rid of this console warning? It's not part of the package.json...

[limefrogyank]

function angleChars(parser: TexParser, mchar: string | TexToken | StackItem) {
    if (mchar && mchar instanceof TexToken) {
        const def = mchar.attributes || {};
        def.mathvariant = TexConstant.Variant.NORMAL;
        def.class = "MathML-Unit";
        //const emptyToken = parser.create("token", "mi");
        const symbolToken = parser.create("token", "mi", def, mchar.char);
        //const msupNode = parser.create("node", "msup", [emptyToken, symbolToken]);
        parser.Push(symbolToken);
    }
}

https://github.com/limefrogyank/siunitx-pcc/blob/b8fb97a930367456da7d9b8d8eb69169531f2112/ts/siunitx.ts#L98-L108

Commented out lines are what were removed.