Thanks for your comments.

We would need a test here, and I think a What's new note for the new math normal option would be good as well.

Will do. I did not add tests yet, because a lot of tests where failing with the text-overaul branch at the time.

Can you link to where you obtained the font files? They do not match what I have in my TexLive installation, but there may be some version mismatch.

I got them from https://ctan.org/tex-archive/fonts/cm/ps-type1/bakoma.

@QuLogic it looks like the tests are still failing with the text-overhaul branch here. Do you know a workaround? It is a bit difficult to pinpoint the tests that I need to update, when everything fails.

It's a bit complicated; you would have to fetch from the text-overhaul-figures branch, then test against those, and commit only the new ones, if any. Of course, if you're only adding a new test, then that is simpler as you can just add it here directly. However, at the moment, that branch is a bit out of sync with text-overhaul as I'm preparing to merge #31107.

If you do update the test images yourself, please keep it as a separate commit so that it can be stripped before merging.

Thanks a lot.
I do expect some existing tests to fail, because the behaviour of \mathit has changed (and even uses an entirely different font for computer modern). Those baseline images would have to be updated in any case. I'll look into that, once you get your branch in sync again. In addition, I would add a small test case for \mathnormal.

The branches should be in sync again.

Taking a quick look at the results, I see in mathfont_cm_10, the mathbf does now look bolder, and in mathfont_cm_20, the mathtt is now fixed-width. Additionally, items like \mathbb are now upright by default; I would have to check if that matches LaTeX though.

There does seem to be something wrong with \mathit though, as in mathfont_cm_{15,16,17} it appears that DejaVu Sans is being used instead of an italic Computer Modern.

Taking a quick look at the results, I see in mathfont_cm_10, the mathbf does now look bolder, and in mathfont_cm_20, the mathtt is now fixed-width. Additionally, items like \mathbb are now upright by default; I would have to check if that matches LaTeX though.

There does seem to be something wrong with \mathit though, as in mathfont_cm_{15,16,17} it appears that DejaVu Sans is being used instead of an italic Computer Modern.

Thanks for testing. Previously, the BakomaFonts class would force all digits (no matter what font was selected) to rm (cmr10) via the latex_to_bakoma lookup table. This would only happen for cm and was not consistent with the other font classes and with LaTeX (https://www.overleaf.com/learn/latex/Mathematical_fonts#Other_mathematical_fonts), although I am not entirely sure if this replicates some ancient version of LaTeX.

mathfont_cm_{15,16,17} looks correct on my end. It looks like cmti to me, as it should.

That said, some other tests still fail here, like mathtext_cm_00. The letters are displayed upright instead of italic in mathnormal mode. There seems to be a logic bug in the other classes. I'll do some debugging.

There does seem to be something wrong with \mathit though, as in mathfont_cm_{15,16,17} it appears that DejaVu Sans is being used instead of an italic Computer Modern.

This is likely due to an outdated local font cache, which does not include cmti10. rm ~/.cache/matplotlib/fontlist* prior to running the tests fixed this for me.

I fixed all of the regressions that I could find. There are still a large number of the computer modern tests failing, which I cannot quite put my finger on.

There does seem to be something wrong with \mathit though, as in mathfont_cm_{15,16,17} it appears that DejaVu Sans is being used instead of an italic Computer Modern.

This is likely due to an outdated local font cache, which does not include cmti10. rm ~/.cache/matplotlib/fontlist* prior to running the tests fixed this for me.

Ah right, I forgot it doesn't update for the internal fonts. You can bump the __version__ in FontManager to force an update.

Some of the remaining test failures with Computer Modern seem to come from metrics being measured on cmti instead of cmmi. One example is _make_space, which measures the width of the 'm' glyph from the 'it' family (which has changed). There are more subtle places, though.

The tests succeed now against the baseline images from QuLogic's branch, except for those few cases that are expected to fail due to the changed behaviour.

The previous failures were caused by the slanted attribute not being set correctly for fontname = normal font in the BakomaFonts class, causing the kerning do be slightly off.

I have updated the announcement to be more vocal about the changed behaviour of mathtext.default: it.
To my understanding, all the review comments are addressed. I would appreciate another review.

Some tests would be nice (you can maybe use check_figures_equal to compare a mathnormal string and the corresponding mixture of mathrm & mathit, not so much to save a few baseline images but more importantly because that would actually test the semantics of mathnormal).

Some tests would be nice (you can maybe use check_figures_equal to compare a mathnormal string and the corresponding mixture of mathrm & mathit, not so much to save a few baseline images but more importantly because that would actually test the semantics of mathnormal).

Both 'normal' (via default math) and 'it' are already covered by the existing tests. A test for explicit \mathnormal is still missing. I'll try to add a test the way you propose.

Note that we cannot compare mathnormal with mathit even for letters, because they can be different fonts/alphabets (for example in the case of computer modern).

I have just rebased the branch against text-overhaul and resolved the merge conflicts due to #31110 being merged. Is there anything else that needs doing?

You can bump the __version__ in FontManager to force an update.

You need to do this when adding new internal fonts.

After clearing the cache, I see 14 test changes. For Computer Modern, we get mathit working as well as mathbf, mathtt and mathsf. But I see that mathbf+mathsf (test 57) doesn't get bold; is there a way to fix that one? For DejaVu Sans/Serif and STIX, these are tests 15&18 where mathit is now really italic. However, it seems that STIX Sans doesn't also get fixed?

Thanks for your feedback!

You can bump the __version__ in FontManager to force an update.

You need to do this when adding new internal fonts.

Done.

After clearing the cache, I see 14 test changes. For Computer Modern, we get mathit working as well as mathbf, mathtt and mathsf. But I see that mathbf+mathsf (test 57) doesn't get bold; is there a way to fix that one? For DejaVu Sans/Serif and STIX, these are tests 15&18 where mathit is now really italic. However, it seems that STIX Sans doesn't also get fixed?

I expect these 14.
Test 57: It now correctly renders sans-serif digits. Previously, numbers were hardcoded to it everywhere. Combining two environments was not implemented for cm (see below).

stixsans: Unfortunately, sans-serif italic digits are not defined as unicode symbols (see https://en.wikipedia.org/wiki/Mathematical_Alphanumeric_Symbols#Digits) nor in Stix 1. So we have to fall back to sans-serif upright here.

A general remark: Combining two environments such as mathbf + mathsf is unique to the hard-coded Stix implementation and is realized via font_class. Neither classic LaTeX nor lualatex (with or without unicode-math) support this -- they simply use the innermost environment. The modern way in lualatex with unicode-math would be a \mathsfbf environment, but that is something for a future PR.