Better errors in bibliographies

[dginev]

Fixes #2099 (using the feedback from #2216). This PR:

1. Moves ltx:ERROR from Meta.class to Misc.class in the schema, as errors are typically visible in the final formats, which fits better with the purpose of Misc.
2. Allows bibitem and bibblock to auto-open, as suggested by Bruce.
   • I tried to double-check for side-effects, and it looks like as long as ltx:biblist is never allowed to auto-open, this change will have limited impact, as bibitem is only allowed in a biblist, and bibblock is only allowed in its respective item.

This approach has the nice side-effect that it will group together any trailing content from an unknown macro as in \myitem{Some content} will now produce a self-contained:

<bibitem>
  <bibblock><ERROR class="undefined">\myitem</ERROR>Some content
  </bibblock>
</bibitem>

Lastly, I think it is worthwhile to quickly open the diff for LaTeXML.model here and check the implied cases where ERROR (dis)appeared, as the change from Meta to Misc led to some interesting differences. As just one example - ltx:chapter used to allow error, but now doesn't, so it auto-opens an ltx:para to host the error instead (for latex such as \chapter{Test}\badmacro).

Here is this PRs visual HTML behavior on the test from #2216 :

Aside: I also tried investigating the Document::makeError method that Bruce mentioned in our recent meeting, but it looks like the TAG_PROPERTIES are consulted quite early - already in computeIndirectModel, which gets cached after it is computed once. That seems a little tricky to interfere with and make exceptions for ltx:ERROR? Could be possible of course, but this solution did the trick as well.

[brucemiller]

This looks promising. Your case for misc vs meta is fairly convincing (though Error is still kinda meta too). Now that that the schema is easier to read, might be a further improvement available? I worry a bit about autoOpen for misbehaved arXiv papers, but perhaps they'll sort themselves out no worse than before.

[dginev]

might be a further improvement available?

Do you mean as part of this PR, or as a follow-up improvement? The autoOpen approach is consistent with the other list environments insofar as ltx:item and ltx:para also auto-open for this kind of error in {itemize} and {enumerate}.

Aside: actually ltx:para isn't declared as autoOpen, but Document.pm auto-opens it as a special case. I just realized that while reviewing code for this comment.

But I got curious and tried one broken test with a bibliography followed by a paragraph/section/chapter. The PR approach leads to some good recovery where it won't slurp in a following paragraph as part of the bibliography, but on the downside it auto-opens empty bibitems for each empty line until the auto-close takes place:

\documentclass{article}
\begin{document}
\begin{thebibliography}{3}
  \xyz

\paragraph Trailing
\end{document}

  <bibliography xml:id="bib">
    <title>References</title>
    <biblist>
      <bibitem>
        <bibblock><ERROR class="undefined">\xyz</ERROR></bibblock>
      </bibitem>
      <bibitem xml:id="bib.bib1">
        <tags>
          <tag>[1]</tag>
          <tag role="refnum">1</tag>
        </tags>
        <bibblock/>
      </bibitem>
      <bibitem xml:id="bib.bib2">
        <tags>
          <tag>[2]</tag>
          <tag role="refnum">2</tag>
        </tags>
        <bibblock/>
      </bibitem>
    </biblist>
  </bibliography>
  <paragraph inlist="toc" xml:id="S0.SS0.SSS0.Px1">
    <title>T</title>
    <para xml:id="S0.SS0.SSS0.Px1.p1">
      <p>railing</p>
    </para>
  </paragraph>

[brucemiller]

All in all, this seems pretty reasonable and doesn't cause problems that I can see in normal cases. We'll have to watch for anything pathological in arXiv bibliographies, tho it may turn out to be the best solution overall.