Greptile Summary

This PR adds support for \overbracket and \underbracket LaTeX commands by extending the existing horizontal brace infrastructure. The implementation follows the established pattern used for \overbrace and \underbrace:

• Adds the new commands to the horizBrace function definition (reuses all existing rendering logic)
• Defines Unicode code points (U+23B4 and U+23B5) for MathML output
• Provides SVG path geometry for bracket rendering, following the pattern of leftlinesegment/rightlinesegment
• Includes comprehensive test coverage mirroring the existing brace tests
• Updates documentation to reflect the newly supported commands

The implementation is clean, minimal, and leverages existing code effectively. All changes are consistent with the codebase patterns.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The changes follow established patterns exactly, reusing proven infrastructure for horizontal braces. The implementation is minimal (only adds function names, Unicode mappings, and SVG paths), includes comprehensive tests, and updates documentation appropriately. No complex logic changes or potential edge cases introduced.
• No files require special attention

Important Files Changed

_{Last reviewed commit: 0d94372}

Thanks for the feedback, I'll work on the dimensions.

I've addressed review comments:

• Thickness and height are closer to LaTeX now (compared with mathtools)

• Also it seems to me that mathtools puts some kern around brackets regardless of the surroundings, I've tried to replicate this too.
• There are fewer unit tests.

I am commenting just to say that this PR is cool. @grigoriy-reshetniak, you are the first contributor other than me to use SVGs to extend KaTeX. I'm happy that someone has been able to follow and extend the methods that I used.

@edemaine has been maintaining KaTeX by himself for a while now, and I'm glad that you have stepped up to give him some support.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.30%. Comparing base (314f856) to head (7459f41).
⚠️ Report is 107 commits behind head on main.

@@            Coverage Diff             @@
##             main    #4147      +/-   ##
==========================================
+ Coverage   93.15%   93.30%   +0.14%     
==========================================
  Files          91       91              
  Lines        6795     6810      +15     
  Branches     1582     1585       +3     
==========================================
+ Hits         6330     6354      +24     
- Misses        429      454      +25     
+ Partials       36        2      -34     

Can you add the screenshots from the Screenshotter actions' uploaded artifacts?

Unfortunately, I'm not able to find them. I'm able to see artifacts from failed builds, like https://github.com/KaTeX/KaTeX/actions/runs/22549692911, but its not available for successful ones (I believe because no diff is introduced), like https://github.com/KaTeX/KaTeX/actions/runs/22796025639?pr=4147.

This is KaTeX output in dev server for

\text{test}\underline{hello}\text{test}\underbracket{hello}_{world}\text{test}\underbrace{hello}_{world}\text{test}
\\[1em]
\text{test}\overline{hello}\text{test}\overbracket{hello}^{world}\text{test}\overbrace{hello}^{world}\text{test}

This is LaTeX output for the

\begin{align*}
  \text{test}\underline{hello}\text{test}\underbracket{hello}_{world}\text{test}\underbrace{hello}_{world}\text{test} \\[1em]
  \text{test}\overline{hello}\text{test}\overbracket{hello}^{world}\text{test}\overbrace{hello}^{world}\text{test}
\end{align*}

It seems to me that it's pretty close now.

(I believe because no diff is introduced

Actually I think this was #4158! Hopefully it works now.

I ran texcmp (from dockers/texcmp) on the screenshotter example and the two example lines from your comment, and they indeed look good:

If I were to nit pick, I'd say the green (KaTeX) under/overbracket is slightly thicker. Did you get the line thickness measurement from LaTeX somewhere, or did you eyeball it? If it's from LaTeX, I'm happy to keep it as is; maybe just a browser rendering thing.

The horizontal spacing looks perfect! And ignore the sub/superscripts looking so different; that's #3972.

I did it by eye. However, I've checked documentation for mathtools and in 3.3.2 Braces and brackets it says:

The default rule thickness is equal to that of \underbrace (app. 5/18 ex)

KaTeX \underbrace is has thickness of 120 (SVG path), but current implementation of \underbracket has thickness of 100.

Per my understanding of documentation \underbracket should be even thicker (or \underbrace thinner to match). At the same time I made it thinner before and it seems to be too thick anyway.

What do you think?

I think I must have red and green backwards. Indeed, the texcmp README says green is LaTeX. (I was confused because LaTeX is the red channel, but light is additive, so...)

So overbracket is indeed too thin, and I'd up it to 120 like the documentation says.
Indeed, I think the right answer is approximately 120, which according to stretchy.js corresponds to 0.120em. This code:

\documentclass{article}
\begin{document}
$
\newlength\LEN
\setlength\LEN{1em}
\showthe\LEN
\setlength\LEN{0.277777ex}
\showthe\LEN
$
\end{document}

measures 1em at 10.00002pt and 5/18 ex at 1.19595pt, so 5/18 ex = 0.119595 em. Very close to 0.120em.

I agree it looks thick, but compatibility is a clearer target. In the future, we could try to make the thickness configurable — if that's even possible...

Here are the texcmp's with 120:

I think this is a bit better: there's now both red and green of roughly equal amount. So the thickness seems right, just vertical offset is slightly off.

I agree that \underbracket thickness should be 0.12 em, per mathtools docs and KaTeX fontMetrics definition of x-height.

To define \underbrace geometry, I converted a KaTeX font into an SVG font, copied a brace glyph, rotated it 90°, and used that path.

If you really want to make the thickness configurable, there is a go-by in current KaTeX settings. delimiter.js reads minRuleThickness from the KaTeX settings and passes a value to svgGeometry, which modifies the thickness of a sqrt vinculum. It's a lot of trouble for something used very infrequently. I recommend sticking with 0.12 em.

Next thing I need to do -- update docker images for screenshotter. I'm not able to run it on arm architecture so I have to take image from artifacts.

Here are the texcmp's with 120:

I reran texcmp and the offsets seem slightly worse:

The top red text seems slightly higher, and the bottom red text seems slightly lower, than before, but should be the other way around.

Here's fresh diff from texcmp:

🎉 This PR is included in version 0.16.42 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀