There is currently a problem with the Previews. The current wrapper only works with the chtml output but not with the SVG output. Here is a working prototype:

....
        startup: {
		// MathJax creates anchor tags from MathML elements with href attributes.
		// But it does not add the title attributes from these elements
		// that we need for the extension Popups
		ready() {
			const { MML } = window.MathJax._.core.MmlTree.MML;
			MML.a = MML.mrow;
			const { ChtmlWrapper } = window.MathJax._.output.chtml.Wrapper;
			ChtmlWrapper.prototype.handleHref = function ( parents ) {
				if ( !this.node.attributes.hasExplicit( 'href' ) ) {
					return parents;
				}
				const attrs = { href: this.node.attributes.get( 'href' ) };
				if ( this.node.attributes.hasExplicit( 'title' ) ) {
					attrs.title = this.node.attributes.get( 'title' );
				}
				return parents.map(
					( parent ) => this.adaptor.append( parent, this.html( 'a', attrs ) )
				);
			};
			const { SvgWrapper } = window.MathJax._.output.svg.Wrapper;
			SvgWrapper.prototype.handleHref = function ( parents ) {
				const href = this.node.attributes.get( 'href' );
				if ( !href ) {
					return parents;
				}

This issue seems to only apply to the SVG rendering mode. MathML and MathJax both display the TeX formula.

In T414289#11531820, @FrederikHennecke1 wrote:

Firefox 140.7.0esr (64-bit) on Debian 13.

Firefox 140.7.0esr (64-bit) on Debian 13.

In T414289#11531134, @Physikerwelt wrote:

In the MathWG we defined that it should be possible to replace all mrow with a and stop outputting mrow at all see
https://codepen.io/physikerwelt/pen/XJKMMyo

<math xmlns="http://www.w3.org/1998/Math/MathML" class="mwe-math-element mwe-math-element-inline">
    <mrow data-mjx-texclass="ORD">
        <mstyle displaystyle="true" scriptlevel="0"><a href="https://en.wikipedia.org/wiki/Energy" title="energy">
                <mi>E</mi>
            </a>
            <mo stretchy="false">=</mo><a href="https://en.wikipedia.org/wiki/Mass" title="mass">
                <mi>m</mi>
            </a>
            <msup><a href="https://en.wikipedia.org/wiki/Speed_of_light" title="speed of light in vacuum">
                    <mi>c</mi>
                </a>
                <mrow data-mjx-texclass="ORD">
                    <mn>2</mn>
                </mrow>
            </msup>
        </mstyle>
    </mrow>
</math>
<br>
<math xmlns="http://www.w3.org/1998/Math/MathML" class="mwe-math-element mwe-math-element-inline">
    <mrow data-mjx-texclass="ORD">
        <mstyle displaystyle="true" scriptlevel="0">
                <mi>E</mi>
            <mo stretchy="false">=</mo>
                <mi>m</mi>
            <msup>
                    <mi>c</mi>
                <mrow data-mjx-texclass="ORD">
                    <mn>2</mn>
                </mrow>
            </msup>
        </mstyle>
    </mrow>
</math>

This is the rendering in Firefox. Chrome and Webkit look normal. But this issue is a bit wonky. I did screenshot it in my default Firefox browser(140.7.0esr ). This issue did not show up when I tried to compare it with BackstopJS so it may just be my local font.

In T414289#11512003, @Physikerwelt wrote:

@FrederikHennecke1 do you think it is a bug in MathJax that the title is not used? If yes, we should file an issue here https://github.com/mathjax/mathjax/issues

I just tried to find all formulas in enwiki with QIDs with a short python script but I could only find 161 unique formulas. Is this correct? If that is the case I think that we don't necessarily have to rewrite the cache to cache these formulas. I also tested the performance of with / without caching and got the following results:

I found one problem with the new font:

I did a visual regression test and I noticed two issues:

The current MathML looks like:

In T375238#11333288, @Physikerwelt wrote:

I looked at \grave{a} and was unable to identify the underlying problem. The generated MathML code is identical to the MathML code generated by mathoid. If I render the resulting MathML code with the interactive Mathjax demo from https://mathjax.github.io/MathJax-demos-web/input/tex2mml.html the result looks fine

It looks like this was fixed in commit id c5188b11172842b1afe3a6dcabb48a05ad04413d in T401623.

In T388985#11227827, @Physikerwelt wrote:

Thank you. That seems to better match my expectations. I'll try to run your github on my machine to see if I can match those values. In https://ceur-ws.org/Vol-1785/W48.pdf we show the rendering time over the input length (Fig. 3). I find that a bit more intuitive (but more complicated) than the median.

I thought I did but I was mistaken. Seems like I forgot to set one specific cache value and the caching did not work. I fixed the benchmark and now the results are completely different:

benchAllMissCore........................I49 - Mo195.291ms (±2.60%)
benchAllHitCore.........................I49 - Mo18.568ms (±4.47%)

I created a small benchmarking script for LocalChecker: https://github.com/FrederikHennecke/mediawiki-extensions-Math-testing/tree/master/benchmarks
I used memcached for caching. The results show that the difference between a hit and a miss in the cache is relatively small. As for the input data, I used the reference.json file from the tests. Btw, this is my first time working with the cache, so take these results with a grain of salt.

benchAllMissCore........................I49 - Mo152.840ms (±1.30%)
benchAllHitCore.........................I49 - Mo148.083ms (±2.42%)

In T388985#11178481, @Physikerwelt wrote:

Thank you that’s a great plan. However I think it’s most effective if we discuss the caching and rendering question a bit more in detail before proceeding with the implementation.

1. do we have current measurements of the performance and the impact of caching?

2. can we estimate how the proposed changes would affect the performance? (Now, we have almost no qid / vs future with x % qid)

3. can we generate a code pen, with a few complex and simple examples that demonstrates how the rendering will look like when implementation is done. That would help us to test the impact on even individual browsers.

I propose the following changes for this task:

Looks like the patch has two types of issues: first, the errors from T402362, T402434, and T402660, which should all be fixed.
And the other issues are from commit d562047, where the default tex checker was changed from restbase to local. The local mhchem parser does seem to be missing some features.

In T389712#11018564, @Physikerwelt wrote:

I just updated the progress marks...
@FrederikHennecke1 is 5d complete?

I propose the following steps for Step 5:

I just noticed that one step is missing: we have to replace MMLbase::getEmpty. This is only called for MMLmspace or MMLmrow and returns the empty Element <mspace\>. With our current implementation, we get the full tag just without an inner text: <mspace><\mspace>. Either we change the tests, or we change MMLDomVisitor::getHTML from $this->dom->saveHTML( $this->dom->documentElement ) to $this->dom->saveXML( $this->dom->documentElement, LIBXML_NOEMPTYTAG ). That should automatically close all empty elements while keeping everything else the same.

For Task 4.c, I have the following points:

1. Keep

if ( $node instanceof MMLleaf ) {
	$textNode = $this->dom->createTextNode( $node->getText() );
	$element->appendChild( $textNode );
	return;
}

instead of

Ok, then we can keep supporting the DOMVisitor. We can still change the visitor later on if needed.

I noticed some performance issues with the new MMLDomVisitor implementation.
I merely tested the old .encapsulate function against the new toString implemenation. I also wrote several other classes to test different XML/HTML implemntations.
Here are the times for 10,000 repetitions:

I created a patch, where we automatically take the symbols and the urls from Wikibase property "in defining formula" and add them as anchor tags to the corresponding MathML Element. Also, for MathJax, the links are set via href in the MathML elements, because MathJax does currently not support anchor tags (but will soon in ver 4).

In T381046#10393040, @Physikerwelt wrote:

@FrederikHennecke1 Can you document as a step-by-step guide what you did to get the popups working locally?

Small note: The interaction with Popup is currently broken (see T381310). So an older version of Extension:Popup should be used to test.

In T381046#10364536, @Physikerwelt wrote:

1. Did you analyze why the popup is not showing?

Because the popup module is currently looking for image tags inside .mwe-math-element elements with data-qid attributes, and the MathML mode is not using any images or the data-qid attribute.

1. Is there any output in the browser's console?

No.

1. How is the situation with other browsers?

I just also tested in MS Edge 131.0.2903.70 and Safari 14.1. where it's also broken.