\\[] fails with negative height lengths

[DonJBridges]

If \\ is given a negative length, for example like in

\begin{align}
x + y = z \\[-1 pt]
1 + 2 = 3
\end{align}

MathJax gives no output. However, negative vertical spacings are extremely useful for fine-tuning the spacing between interwoven formulas, e.g.

\begin{align}
x_{x_{x_{x_x}}} + y &= z \\ %Insert [-6 pt] space here!
2 &= 1 + 1 + 3^{3^{3^3}}
\end{align}

This might also be related to this issue.

[dpvc]

It is not the issue you point to, but it is something that should be fixed. I will have to make some change to the HTML-CSS and SVG output processors to accommodate that, and there are some issues to deal with in terms of the underlying MathML and how to represent it, but it can probably be handled.

In the meantime, you could use

\begin{align}
\smash{x_{x_{x_{x_x}}}} + y &= z \\
2 &= 1 + 1 + \smash{3^{3^{3^3}}}
\end{align}

to get the spacing you are looking for.

[fred-wang]

A simple != test would be to compare the rendering of [-1 pt] and [invalid].
However, it is probably best to write a LaTeXToMathML test when we find how to implement this in MathML.

[dpvc]

I have made changes to allow negative dimensions in \\[...], but they are clipped to 0 since negative dimensions on the rowspacing are not supported by MathML. This is just to prevent the error message from the TeX parser for now. I will need to come up with something better in the future, so am marking this Address later for now.

The issue236 branch of my fork of MathJax includes the new code.

@fred-wang, this also includes code to trim the number of decimals used in columnspacing specifications set up by the AMSmath extension. That may affect some tests if they have the complete decimals that were being produced in the past.

[fred-wang]

OK, I'll just write a test to verify that negative dimensions are clipped to 0 + update the tests with columnspacing.

[fred-wang]

Finally, I kept the idea of a != reftest against [invalid], since we haven't implemented the correct MathML yet:

LaTeXToMathML/issue236.html

The tests that needed to be updated are:

LaTeXToMathML/basic-operators/basic-operators-2.html
LaTeXToMathML/environments/align-1a.html
LaTeXToMathML/environments/align-1b.html
LaTeXToMathML/environments/aligned-1.html
LaTeXToMathML/environments/aligned-2.html
LaTeXToMathML/environments/split-1.html
LaTeXToMathML/matrices/matrix-2.html
LaTeXToMathML/matrices/smallmatrix-1.html
LaTeXToMathML/mhchem/environments-1.html
Parsing/issue230.html

It seems that rowspacing has changed in LaTeXToMathML/matrices/matrix-2.html too. However, the former value contained a unitless "0.4" which I guess was incorrect.

[fred-wang]

Verified in Opera, Chrome and Firefox.

=> In Testsuite, Ready for release

[dpvc]

Thanks for checking which tests needed changing. I had forgotten about the rowspacing fix that added the missing units. Good catch!

[fred-wang]

Finally, I kept the idea of a != reftest against [invalid], since we haven't implemented the correct MathML yet:

I updated the test to compare against 0. We will need to update that later.

[pkra]

Should we create a label "spec bug" (like browser bug)?

[dpvc]

Closed during review of old issues.