handle empty case in Tokens::stripBraces

[dginev]

Fixes #2452

The classic loops in Tokens::stripBraces were missing an edge case where the braces held nothing - and an empty Tokens would be the appropriate response.

Interestingly, the bug reported in #2452 was absent from the Rust port, as I had used Rust's Iterator abstraction (with peekable), which safely avoids the need of index arithmetic.

[dginev]

I also added the report from the issue as a new test here, to spot any regressions (e.g. during porting)

[xworld21]

I suggest adding {{}} (and maybe {{{}}} too), to be sure that the correct number of braces is stripped out. When I reported the issue, I fixed it locally by stripping all braces, but that was wrong too.

[dginev]

Great point, the check only comes back empty for {} and {{}} in keyval.

The stripBraces fix is correct for the intended semantics of "stripBraces", but it looks like keyval should only strip them twice at most. So it should be called twice? ...

[dginev]

Alright, a little scary but I updated the test to request exactly 2 braced strips for this case, extending stripBraces to allow for a number of layers.

[dginev]

Btw, there is a semi-related curiosity that I recently learned thanks to Norbert Preining, about there being a maximum number of strips in usepackage braces. If you look at the following example:

\documentclass{article}
\usepackage{{{{{{graphicx}}}}}} % works
\usepackage{{{{{{{graphicx}}}}}}} % fails

\begin{document}
oops.
\end{document}

pdflatex will successfully unwrap 6 braces, but will keep the seventh and try to load {graphicx}.sty in the second case. I consider it a feature that latexml does not emulate that accurately, and would succeed with both 6 and 7 braces. I also hope this is rarely used by real authors...