Underfull \hbox with amsmath

[ethey]

When using a split environment inside an align one, I always get an Underfull \hbox warning if Babel is loaded.

Here is a minimal working example:

\documentclass{article}

\usepackage{babel}
\usepackage{amsmath}

\begin{document}

\begin{align}
  \begin{split}
    a &= a \\
      &\quad + 0
  \end{split} \\
  \begin{split}
    a &= a \\
      &\quad + 0
  \end{split} \\
\end{align}

\end{document}

If I remove Babel, then I do not have any warning. Visually I do not observe any difference in the PDF that is produced, and this only seems to happen with LuaLaTeX, and not pdfLaTeX nor XeLaTeX. So I am not sure if that should be reported here.

I have tried with an up to date version of TeXLive 2021.

[FrankMittelbach]

I don't get any underfull box with babel from your example, but I get a third equation number because of the \\ after the last split. When I add \listfilesto the preamble it reports

 *File List*
 article.cls    2020/04/10 v1.4m Standard LaTeX document class
  size10.clo    2020/04/10 v1.4m Standard LaTeX file (size option)
   babel.sty    2021/05/16 3.59 The Babel package
     nil.ldf    2021/05/16 3.59 Nil language
 amsmath.sty    2020/09/23 v2.17i AMS math features
 amstext.sty    2000/06/29 v2.01 AMS text
  amsgen.sty    1999/11/30 v2.0 generic functions
  amsbsy.sty    1999/11/29 v1.2d Bold Symbols
  amsopn.sty    2016/03/08 v2.02 operator names
l3backend-pdftex.def    2021-05-07 L3 backend support: PDF output (pdfTeX)
 ***********

as version numbers.

[u-fischer]

@FrankMittelbach you used pdftex, but the report says the problem is with lualatex (and I get underfull hboxes there).

[FrankMittelbach]

@u-fischer ok one should read to the end, sorry. So yes, it is there in LuaTeX and to report it report here seems correct, as it comes from

\bbl@mathboxdir ->\ifcase \bbl@thetextdir \relax \everyhbox {\textdir TLT\relax
 }\else \everyhbox {\textdir TRT\relax }\fi 

Thus \begindir TLT ends up in each and every hbox and that is basically deadly ... in particular here

\split@ ->\hbox to\wd \thr@@ {}&\vcenter {\vbox {\moveleft \wd \thr@@ \boxz@ }}

Tex doesn't report underful boxes when they are empty (so the above was fine but with the TLT node added it isn't any longer empty, thus ... :-)

At least that is how I read it.

[FrankMittelbach]

Well, one could argue that such nodes just setting direction should not change the state of the box being empty or not, but I'm not sure that is going to be a sell for the LaTeX developers --- but maybe it is.

[blefloch]

Alternatively, to stay with changing stuff that's under our control, change \bbl@mathboxdir to

\def\bbl@mathboxdir{\ifcase \bbl@thetextdir \relax
  \everyhbox {\bbl@mathboxdir@ L}\else\everyhbox{\bbl@mathboxdir@ R}\fi}
\def\bbl@mathboxdir@#1{\@ifnextchar\egroup{}{\textdir T#1T\relax}}

(Not tested but the idea is to suppress the \textdir TLT when the box is empty.)

[FrankMittelbach]

cool approach @blefloch , can't think of a reason why that shouldn't work.

[ethey]

I have tested on my document, and none of the spurious bad boxes are reported anymore.

Also, I almost did not want to report this because I could not figure out where this came from. Basically what you did @FrankMittelbach was to use \showoutput and noticed the \begindir TLT?

[FrankMittelbach]

Yes (after reading your report when Ulrike pointed out my initial mistake). From there it was a little guesswork, but likely that @jbezos introduced it so I'll cut down your example further and then did a \traceon on it searching for suspicious usages of TLT :-) the rest was easy.

[ethey]

Thanks!

[jbezos]

You have sent 8 replies before I even read it! 🙂 Seems easy to fix (actually, there is very likely another issue, because this function should not be enabled by default, only with bidi =).

[jbezos]

@blefloch Nice. It fails if there is some macro (eg, \@empty), but it's a very good solution.