Intent of MathML-Core B.3 is unclear

[davidcarlisle]

The mathml-core Pull Request #302 makes some minor fixes to the tables in appendix B.3 but the initial paragraph of the appendix does not make the intent of the tables clear.

https://www.w3.org/TR/mathml-core/#combining-character-equivalences

For example grave (all accent mappings are similar):

The first table ("combining") maps the spacing grave U+0060 to the combining grave U+0300

The second table ("non combining") maps U+0300 to U+0060

It isn't clear why there are two tables, if the intended effect is to specify that either U+0300 or U+0060 may be used with <mover> to construct a grave accent.

As recent tests show, for example at

https://mathfonts.github.io/math-accents

currently which characters work as accents, especially as stretchy accents, is highly font and viewer dependent. It would be good to point to this appendix B.3 to give guidance on which characters are intended to work, but the text here is very unclear.

@fred-wang as I commented in the PR I couldn't find the script that made these tables (although clearly they relate to <combref and <noncombref entries in unicode.xml so we must have had some xslt or python once?)

This corresponds to issue w3c/mathml#247 on the full spec.

[davidcarlisle]

There are only three characters where the mappings between combining and non-combing are not simply inverses, possibly these are related to the incorrect IDs fixed in the PR

2190 maps to 20D6 in the first table, with no reverse map in the second.

0312 maps to 00B8 and 0338 maps to 002F in the second table with no reverse map in the first.

If those entries were added the two tables would be the same apart from the order of the columns.

[davidcarlisle]

I wrote

but the initial paragraph of the appendix does not make the intent of the tables clear.

some searching the git history shows that this whole appendix was a data dump (by me) for discussion back in 2019 at

9219999

but it hasn't changed since then.

I propose to extend the existing PR #302 to delete the second table (which would have the same data as the first) and add some words that say that the tables show combining and non combining characters representing accent constructs, and to give a non normative recommendation (or normative?) recommendation that a MathML processor treats the characters as equivalent input, and for wide accents uses whichever character has a suitable construct in the current font.