Sandbox run 01.2024, fixes for Fatal cases (10)

[dginev]

This PR lifts a first small selection of the unfortunate Fatal regressions in the arXiv sandboxes with the latest master:

1. Add a guard for \xyoption to avoid loading the same option twice (as inspired by the original sty, which also does so)

2. allow more recursion in deeply nested argument reads, even if a true infinite loop lurks behind that - the final Fatal printed gets a more informative trace in the genuine loop cases.

3. add a guard in revertScript to only call ->defined_as on Token objects (and not Tokens).

4. recognize and guard the advanced use of \multirow where the height param can be a negative or fractional number. Not implemented yet, but won't Fatal anymore.

5. Fix a surprisingly common regression with xparse and lipsum conflicting (really my hacky expl3 "pool" conflicting with the expl3.sty binding). I believe this was the trickiest issue to detect here. And yes, 1% of sandbox articles included lipsum in a conflicting setting.

   • The minimal example is simply:

   \documentclass{article}
   \usepackage{xparse}
   \usepackage{lipsum}
   \begin{document}
   \end{document}

   I think eventually the expl3 "pool" should disappear entirely, but I will defer that until we start loading latex.ltx raw. I think that would be a better home for that removal.

6. A connected change outside of this PR is updating the biblatex.sty experimental binding to catch up to the new internals of url and hyperref - by simply not using them and loading hyperref to get a working \href and \url.

7. guard literal keyword regexes in listings.sty with quotemeta()

8. guard failed parses from realizeXMNode coming back with a non-node object

9. ensure import.sty paths work with ZIP inputs, by making requested paths absolute w.r.t the "SOURCEDIRECTORY" parameter (source is unpacked to a temp dir for ZIP).

10. Recognize math mode in ulem.sty macros.

This concludes round 1 of my debugging of Fatals. I will rerun the sandboxes with the updated commits and circle back with a comment.

[dginev]

The PR's impact on our arXiv sandboxes is now documented here:
https://gist.github.com/dginev/abb72c989f4474a3a33300ed23ac6c86

[brucemiller]

Interesting that the leading comment ends with "Worry about list of lists". That ->revert would yield a list of both Token and Tokens almost seems like the real bug. Either way, your patch doesn't protect against leading Tokens. It's also unclear that simply skipping over Tokens is the right thing; what if there are unbalanced or leading or trailing braces in there?

I'd think a simpler, though slightly inefficient, approach would be

my @tokens = Tokens($script->revert)->unlist;

[dginev]

Excellent suggestion - and indeed more correct.

Done: Tested with two articles, seems to work well.

[brucemiller]

Looks good! Thanks!!