\newtheorem does not use trivlist settings

[mbertucci47]

I know there is some discrepancy expected in output between the tagging code \newtheorem and the usual one. However, I think with minimal effort they could look much closer than they currently do. Right now \@begintheorem calls \UseInstance{block}{theorem}{} but this could be changed to \UseInstance{blockenv}{theorem}{par-skip=\parskip,parindent=\parindent}. This makes the heading indented but adding \noindent before \mode_leave_vertical: seems to fix this without affecting tagging (I don't claim this is the correct fix).

\documentclass{article}
\usepackage{kantlipsum}

\newtheorem{theorem}{Theorem}

\begin{document}

\begin{theorem}
\kant[1][1-2]

\kant[2][1-2]
\begin{enumerate}
\item abc
\item \kant[3][1-2]

\kant[4][1-2]
\end{enumerate}
\end{theorem}

\end{document}

Here's the output without tagging.

Now adding

\DocumentMetadata
  {
    lang=en-US,
    pdfversion=2.0,
    pdfstandard=ua-2,
    testphase={phase-III,math,title,table,firstaid}
  }

to the example above gives

With the proposed fix:

% !TeX TXS-program:compile = txs:///lualatex
\DocumentMetadata
  {
    lang=en-US,
    pdfversion=2.0,
    pdfstandard=ua-2,
    testphase={phase-III,math,title,table,firstaid}
  }
\documentclass{article}

\usepackage{kantlipsum}

\makeatletter
\ExplSyntaxOn
\def\@begintheorem#1#2{
  \UseInstance{blockenv}{theorem}{par-skip=\parskip,parindent=\parindent} % Changed
  \tagpdfparaOff
  \noindent % Added
  \mode_leave_vertical:
  \MakeLinkTarget{\l__block_thm_current_counter_tl}
  \tag_struct_begin:n{tag=Caption}
   \group_begin:
   \bfseries
   \tag_mc_begin:n {}
     #1\
   \tag_mc_end:
    \tag_struct_begin:n{tag=Lbl}
      \tag_mc_begin:n {}
          #2
      \tag_mc_end:
    \tag_struct_end:
    \group_end:
  \tag_struct_end:
  \tagpdfparaOn
  \__block_start_para_structure_unconditionally:n { \PARALABEL }
  \itshape
  \hskip\labelsep
  \ignorespaces
}
\ExplSyntaxOff
\makeatother

\newtheorem{theorem}{Theorem}

\begin{document}

\begin{theorem}
\kant[1][1-2]

\kant[2][1-2]
\begin{enumerate}
\item abc
\item \kant[3][1-2]

\kant[4][1-2]
\end{enumerate}
\end{theorem}

\end{document}

Similar changes would need to be made for the amsthm firstaid.

Edit

Hmm I see the proposed fix also affects the indent in enumerate. Not sure about that at the moment.

[FrankMittelbach]

Your examples, seem to be a white lie @mbertucci47 :-). You say Now adding ... which fooled me at first. In reality, I think your first display was typeset with pdftex and the second with tagging was typeset with luatex. It is a bit surprising that we see so much change in the paragraphs, but I guess that is CM to LM.

Anyhow, your proposed solution goes in the wrong direction in my opinion. You do not want to hardwire such settings somewhere in the middle of the code (this is worse than allowing \@listi to mess things up, I would say). What you want to do instead is to make use of the template/instance machinery.

So a better solution, in my opinion, might be this:

\DeclareInstance{blockenv}{theorem}{display}
{
  env-name       = theorem-like,
  tag-name       = theorem-like,
  tag-class      = ,
  tagging-recipe = standalone,
  inner-level-counter  = ,
  level-increase = false,
  setup-code     = ,
  block-instance = theoremblock ,   % <--- do not reuse the displayblock, use your own
}

\DeclareInstance{block}{theoremblock-0}{display}
  {
    parindent      = \parindent ,
    par-skip        = \parskip ,
  } 

That doesn't solve this, unfortunately:

Hmm I see the proposed fix also affects the indent in enumerate. Not sure about that at the moment.

That is a bug in the code; I guess I confused myself trying to support the various legacy setting possibilities. For the moment that could be (partly) repaired with

\DeclareInstance{block}{list-1}{display}{  parindent  = 0pt } 

but this way you can't alter the \listparindent globally any longer. The correct fix has to happen in the code.

[mbertucci47]

In reality, I think your first display was typeset with pdftex and the second with tagging was typeset with luatex.

@FrankMittelbach Wow I did not even notice that, yes you are right. Sorry about that!

What you want to do instead is to make use of the template/instance machinery.

Yes that's definitely better. When I naively add your proposed code to my example it seems to indent the entire block, but I suspect some other changes being made in latex3/latex2e#1561 might fix this.

[FrankMittelbach]

but I suspect some other changes being made in latex3/latex2e#1561 might fix this.

It needed more than my handwaving reply (due to one or two bugs discovered this way. I think it is now correct and will show up in the next dev release in a couple of days.

[FrankMittelbach]

@mbertucci47 Its now in the dev release an that should be available by now, so please give it a try and see if my additions and changes have improved the situation

[mbertucci47]

@FrankMittelbach Looks good to me. The only difference I see in my example between tagging and no tagging is with tagging, the distance between the theorem name and number is slightly less than without tagging. With pdflatex (so computer modern, not latin modern), it's enough to change line breaks.

[mbertucci47]

To clarify, I think that issue was there before the most recent change.

Here are screenshots showing the difference. Both were compiled with pdflatex-dev.

Without tagging

With tagging

The Theorem 1 with tagging looks a little cramped to me. The issue is still there with lualatex-dev but due to font metrics it's not enough to change the line break.

[FrankMittelbach]

Here are screenshots showing the difference. Both were compiled with pdflatex-dev.

it is not the \labelsep; it is the fact that pdflatex by default still uses OT1 fonts, but tagging always T1 fonts and they are different. If you run both with \usepackage[T1]{fontenc} the linebreaking changes not so much. But there is another difference in that the text body is somewhat offset on the page. Nothing to do with the list stuff, that looks like something entirely different (and should probably get some test file and an issue for itself.

[mbertucci47]

Good point, but I don't know what you mean by "the linebreaking changes not so much". If I load \usepackage[T1]{fontenc} explicitly I get identical-looking output to the images above (still cramped with tagging, not cramped without).

But there is another difference in that the text body is somewhat offset on the page.

I'm not able to replicate this. Adding \usepackage{showframe} they both look aligned to the left and right.

To avoid confusion, here is the test file I'm using.

\DocumentMetadata
  {
    lang=en-US,
    pdfversion=2.0,
    pdfstandard=ua-2,
    testphase={phase-III,math,title,table,firstaid}
  }
\documentclass{article}
\usepackage{showframe}
\usepackage[T1]{fontenc}
\usepackage{kantlipsum}

\newtheorem{theorem}{Theorem}

\begin{document}

\begin{theorem}
\kant[1][1-2]

\kant[2][1-2]
\begin{enumerate}
\item abc
\item \kant[3][1-2]

\kant[4][1-2]
\end{enumerate}
\end{theorem}

\end{document}

and commenting/uncommenting the \DocumentMetadata declaration.

[FrankMittelbach]

Good point, but I don't know what you mean by "the linebreaking changes not so much". If I load \usepackage[T1]{fontenc} explicitly I get identical-looking output to the images above (still cramped with tagging, not cramped without).

hmm, probably means I'm not able to compare the right files :-) yes, confirmed there is this difference and it looks like an incompatibility: we don't box the theorem heading so the space inside contributes to the line strech and that seems to be enough to alter the line breaking a lot (actually for the better in my opinion in that case, but that isn't the point here)

But there is another difference in that the text body is somewhat offset on the page.

I'm not able to replicate this. Adding \usepackage{showframe} they both look aligned to the left and right.

That looks like a typical Heisenberg: showframe makes the problem go away for me. But the cause is that your document is letterpaper (implicitly) but by distribution default on a4paper pdf over hre and if tagging is loaded it becomes real letterpaper on letterpaper PDF. Since your system default is letterpaper output both are identical at your end.
And if showframe is loaded that loads graphics which also fixes everything to letterpaper at my end.

So all in all what remains is put an \mbox around the heading of the theorem.

[mbertucci47]

Ah apologies for the confusion caused by my silly American defaults :)

I can replicate the shifting issue by setting a4paper. It seems like it's not the tagging code but the backend code because if I leave \DocumentMetadata but comment out testphase=... then the shift is still there. And like you said, loading showframe (rather graphics) makes the problem go away.

[mbertucci47]

With my letter paper default and lualatex-dev, toggling the commented lines in the example that follows gives the following behavior.

Bad output

• No \DocumentMetadata{}, no \usepackage{graphics}. This prints a4-sized text on letter-sized paper.

Good output

• No \DocumentMetadata{}, yes \usepackage{graphics}. Correctly prints a4-sized text on a4-sized paper.
• Yes \DocumentMetadata{}, no \usepackage{graphics}. Same as above.
• Yes \DocumentMetadata{}, yes \usepackage{graphics}. Same as above.

With my letter-default, the shifted text only occurs with the first, bad output. I don't know if this is expected behavior so I will let someone else open an issue if needed.

%\DocumentMetadata{} % toggle
\documentclass[a4paper]{article}
%\usepackage{graphics} % toggle
\usepackage{kantlipsum}

\begin{document}

\kant[1-4]

\end{document}