One other (that I think I would remove raher than deprecate is

• namespaced atributes for undefined behaviour.

Not specifying these will have no effect on anything except validators, and validators can choose to use mathml4-legacy schema that makes them valid (currently they are valid in the draft full schema, but I think we could change that). Currently (inherited from core) mathml-presentation allows data-* attributes I think we should allow that in Content as well then drop other and namespaced attributes.

This comment confirms the keywords mentioned in David's comment from issue #294 are indeed not used by mainstream conversion tools in early 2022.

The attributes themselves are quite rare, and what little use I spotted was with numeric values (largely in em or px units, if anyone is curious).

From the full conversion engines I surveyed: latexml, tex4ht, tralics, pandoc, plastex. They do not seem to emit the attribute-keyword pairs in question.

From the formula conversion tools, TeXzilla and katex have no reliance on the keywords.

From what I can see, the MathJax-src repository has most thorough coverage on the subject, but the code is already defensive - e.g. in "MathML 2 compatibility" input guards, or in test files. I don't see reliance on the keywords for present-day output.

I used local grep and github's code search functionality to conduct this survey - hopefully no major oversights.

At the June 19 meeting we agreed to:

1. Deprecate linebreak on mspace
2. Leave maction in but remove the text for the samples since they are implementation specific
3. Deprecate longdiv and radical from menclose (they are only suggested, so not really deprecated, just delisted)

@davidcarlisle: you wrote

One other (that I think I would remove raher than deprecate is

• namespaced atributes for undefined behaviour.

What attrs are you referring to? other was removed long ago but you list it further down in your comment.

@NSoiffer mathml2 and 3 suggested attributes in (any) namespace should be allowed (and the schema does allow them) for local use for any purpose not specified in the spec. so

<math xmlns:my="http://example.com/whatever">
    <mi my:thing="foo">x</mi>
...

This is very natural in an XML namespace world, but it really reacts badly with the html parser. xmlns:my and my:thing end up as attributes in no-namespace with local names xmlns:my and my:thing and so a DOM that can not be serialised as XML (as the web platform xml serialiser enforces namespace well formed xml)

we added data-thing="foo" to match HTML which uses data-* for the same purpose as the namespaced attributes were used. It does not sit so well with RelaxNG Schema which can not specify that but custom validators such as validator.nu can remove data-* before validating.

So, the suggestion was to deprecate using namespaced attributes in favo(u)r of data-* attributes

Specifically

https://w3c.github.io/mathml/#fund_globatt

Also allowed are attributes from the xml namespace, such as xml:lang, and attributes from namespaces other than MathML, which are ignored by default.