`mjx-row > *` CSS style has impact on large documents

[krassowski]

Issue Summary

When very large documents which include mathjax are edited the mjx-row > * contributes to the high cost of style recalculation.

Most of the CSS selectors in MathJax are neatly scoped, use child combinator (>), and avoid using universal selector (*), which altogether makes recent versions of MathJax quite performant when it comes to style calculation. However this one rule stands out:

'mjx-row > *': { display: 'table-cell' }

https://github.com/mathjax/MathJax-src/blob/1a2ef74c0ac0620e7b8de46402c9dce3b95ade52/ts/output/chtml.ts#L150

I see that elsewhere mjx-row > mjx-cell is used. Is the use of the universal selector here intentional, or was it just a placeholder when the code was initially written?

For context, this rule was initially added in mathjax/MathJax-src#50

Steps to Reproduce:

1. Create an HTML document with deep nesting and high number of DOM nodes
2. Load MathJax
3. Enable CSS statistics collection in Chrome/Edge Performance tab
4. Start Chrome profiler
5. Make any modification to the DOM
6. See that a few ms can be spent in "Recalculate Styles" on matching mjx-row > *

Technical details:

• MathJax Version: 3.0+
• Browser: Chrome and Edge 141

Supporting information:

Irrelevant

[krassowski]

Looking at the code, I think the solution to preserve the functionality would be something like:

mjx-row > mjx-cell,
mjx-row > mjx-over, // maybe? not sure
mjx-row > mjx-base,
mjx-row > mjx-under {
  display: table-cell;
}

but I don't know enough about MathJax internals to be confident if this covers all the potential children of mjx-row row.

[dpvc]

I think there is on other situation missing from your list, which is mjx-row > mjx-den (used in fractions). Otherwise, I think you have found the full list. I will do some more testing, but I think that is correct.

There are also some * rules for multi-character stretchy assemblies (like really large parentheses, or really wide overbraces); are those causing problems, too?

[krassowski]

Hmm that's interesting, the other rules with * do not show up in profiling in my environment. I think they are not causing problems as of now, but it would be ideal to have as few rules as possible include *.

[dpvc]

I have made a PR that removes the * CSS rules, as suggested.