\unvbox does not unbox the box

[xworld21]

Couldn't let it sit until after the bank holidays for obvious reasons.

The implementation of the \un*box commands is wrong at

LaTeXML/lib/LaTeXML/Engine/TeX_Box.pool.ltxml

Lines 586 to 591 in 1301850

	DefPrimitive('\unvbox Number', sub {
	my $box = 'box' . $_[1]->valueOf;
	my $stuff = LookupValue($box);
	adjustBoxColor($stuff);
	AssignValue($box, undef);
	(defined $stuff ? $stuff->unlist : List()); });

They should replace the boxes with their contexts, but $stuff->unlist is typically a one element list consisting of the <ltx:block> wrapper. The difference is visible here:

\documentclass{article}
\newbox\mybox
\begin{document}
  Vertical box below.

  \setbox\mybox=\vbox spread 3cm{This box is 3cm high}\box\mybox

  Another vertical box below, or is it?

  \setbox\mybox=\vbox spread 3cm{This box should be 3cm high.}\unvbox\mybox

  Was it 3cm?
\end{document}

The XML output may look the same (LaTeXML won't save the height), but the @vattach attribute is the visible symptom that the box has not been unwrapped.

A hasty workaround is to call $whatsit->getArg(2) instead of ->unlist, and to stick some kind of newline (\nointerlineskip?). A proper solution should at least involve some error checking (TeX throws an error when the box types are mismatched, e.g. \mybox=\hbox{}\unvbox\mybox) and some thinking about the type of newline required. I haven't checked the TeX book for the details.

(Raw beamer uses a lot of \unvbox and it creates nested blocks like there's no tomorrow. ->getArg(2) flattens the output nicely which suggests I am generally right.)

[dginev]

This may have to do with latexml's still evolving model of how a \vbox is represented at the digestion stage. Currently the contents are stored inside a Whatsit property, called content_box. I recall Bruce and I discussed that at times earlier in 2024, but can't recall too many of the details at present...

But a thought that comes to mind is that if the box contents are in that property, we may want to upgrade the Whatsit::unlist method to recurse into those. As in:

# in lib/LaTeXML/Core/Whatsit.pm
sub unlist {
  my ($self) = @_;
  if (my $box = $self->getProperty('content_box')) {
    return $box->unlist; }
  else {
    return ($self); } }

I am not quite confident that this is correct, so adding it as a comment for now.

Edit: If this approach is deemed reasonable, I made PR #2459 for an easy resolution.

---

Aside 1: latexml also stores the spread of a \vbox in the whatsit height attribute, but that doesn't appear to be added to a ltx:para. In my mind that also ties into other discussions about absolute widths and heights reaching the HTML, and thinking of ways to preserve that information while allowing responsive CSS styling.

Aside 2: Interpreting beamer raw sounds a little bit too low-level @xworld21 . Are you starting a new beamer binding from scratch, or are you thinking of upgrading the beamer.cls.ltxml that was contributed in #1652 ?

[xworld21]

Aside 2: Interpreting beamer raw sounds a little bit too low-level @xworld21 . Are you starting a new beamer binding from scratch, or are you thinking of upgrading the beamer.cls.ltxml that was contributed in #1652 ?

I am not sure where I am going with this, yet. I thought I would read raw the TODO items of the existing binding, but I discovered that raw beamer is feasible (except it takes 30 seconds to load beamer! :( ). Also, I am strongly interested in visual accuracy over e.g. responsiveness, at least in the very short term.

[dginev]

Studying the markup problem, I found an example:

\documentclass{article}
\newbox\boxA\newbox\boxB\newbox\boxC\newbox\boxD\newbox\boxE
\begin{document}
  
\setbox\boxA=\vbox spread 1cm{1cm Box}
\setbox\boxB=\vbox spread 2cm{2cm Box}
\setbox\boxC=\vbox{\unvbox\boxA\unvbox\boxB}
\setbox\boxD=\copy\boxC
\setbox\boxE=\vbox{\unvbox\boxC\unvbox\boxD}

combined vbox: \fbox{\box\boxE}
\end{document}

Which in the PDF simply has 4 lines (two pairs of two) on top of each other.
In the XML, today's latexml creates:

<inline-block framed="rectangle" vattach="bottom">
  <block vattach="bottom">
    <p vattach="bottom">1cm Box</p>
    <p vattach="bottom">2cm Box</p>
  </block>
  <block vattach="bottom">
    <p vattach="bottom">1cm Box</p>
    <p vattach="bottom">2cm Box</p>
  </block>
</inline-block>

Is the main request of this issue to reduce that to:

<inline-block framed="rectangle" vattach="bottom">
  <p vattach="bottom">1cm Box</p>
  <p vattach="bottom">2cm Box</p>
  <p vattach="bottom">1cm Box</p>
  <p vattach="bottom">2cm Box</p>
</inline-block>

If so, I can see how this could also be done as a DefRewrite ...