"Extra \else" error when combining breqn, thm-restate, and cleveref

Question

I have a document that uses breqn, thm-restate (thmtools), and cleveref. I can't get it to compile.

Here is a minimal example:

\documentclass{article}

\usepackage{thm-restate}
\usepackage{breqn}
\usepackage{cleveref}

\begin{document}

\begin{restatable}{theorem}{foo}
  Test
\end{restatable}

\end{document}

The output of pdflatex contains many errors like this:

! Extra \else.
<argument> \else 

l.11 \end{restatable}

! Extra \else.
\cref@ifstreq ... \@firstoftwo \else \let \@tempc 
                                                  \@secondoftwo \fi \expanda...
l.11 \end{restatable}

! Extra \else.
\cref@ifstreq ...i \expandafter \endgroup \@tempc 
                                                  {#3}{#4}

I have tried multiple (I think all) orders to load the packages.

My feeling is that this can be fixed by changing the definition of \cref@ifstreq but this is beyond my current LaTeX skills.

Is there a workaround? Should I report a bug? (In which of the packages?)

I use TeX Live 2019, which has breqn 0.98g, thm-restate/thmtools 67, and cleveref 0.21.4. These are all the latest packages from CTAN except for thmtools, which I tried to update to 68 but that did not help either.

Answer 1

Wow, that's a nice finding (if you can call it that). . .

The order doesn't matter, but all three packages play a role in this (though cleveref is innocent).

First, breqn redefines LaTeX's \@ifstar to use \@temp(a|b|c) instead of \reserved@(a|b|c) (see here about these), so when you define a macro \def\test{\@ifstar{A}{B}}, then use \test\else, \@tempc ends up becoming the same as \else (which is dangerous).

thmtools's code does so that \begin{restatable}{theorem}{foo} eventually becomes \expandafter\foo\ifprint\else*\fi, then the \expandafter expands the \ifprint (the name is actually \ifthmt@thisistheone) which, in your case, is true, because you used \begin{restatable}, not \begin{restatable*}. After the \expandafter finishes you are left with \foo\else*\fi (yes, this is bad :-).

Now guess what is the first thing \foo does?
YES! \@ifstar, and now \@tempc is \else.

cleveref is just unfortunate enough to use \@tempc in \cref@ifstreq here:

    \ifx\@tempa\@tempb%
      \let\@tempc\@firstoftwo%
    \else%
      \let\@tempc\@secondoftwo%
    \fi%

which is fine, until \@tempc is the same as \else. The \ifx test is false, so you have:

    \iffalse
    \@tempc % \else
      \@firstoftwo%
    \else%
      \let\@tempc\@secondoftwo%
    \fi%

and you can see where this is going. . .

Here's a patch that fixes thm-restate. The patch changes the \expandafter\foo\ifprint\else*\fi thing to make sure that the entire \if...\fi block is completely expanded before \foo (thus \@ifstar) has the chance to look at it.

\usepackage{xpatch} % Needs to be loaded before breqn
\makeatletter % The patch needs to appear after thm-restate is loaded
\xpatchcmd\thmt@restatable % fixing thm-restatable
  {\csname #3\@xa\endcsname\ifthmt@thisistheone\else*\fi}
  {\begingroup\edef\x{\endgroup
     \expandafter\noexpand\csname #3\endcsname
       \ifthmt@thisistheone\else*\fi}\x
  }{}{\FailedToPatch}
\makeatother

Full document:

\documentclass{article}
\usepackage{xpatch} % Needs to be loaded before breqn

\usepackage{thm-restate}
\usepackage{breqn}
\usepackage{cleveref}

\makeatletter % The patch needs to appear after thm-restate is loaded
\xpatchcmd\thmt@restatable % fixing thm-restatable
  {\csname #3\@xa\endcsname\ifthmt@thisistheone\else*\fi}
  {\begingroup\edef\x{\endgroup
     \expandafter\noexpand\csname #3\endcsname
       \ifthmt@thisistheone\else*\fi}\x
  }{}{\FailedToPatch}
\makeatother

\begin{document}
\begin{restatable}{theorem}{foo}
  Test
\end{restatable}
\end{document}