Hcfyk

I'm trying to write myself a package, similar to the exercise package, so I can prepare problem sets for my students. I wish to delay the output of problems and/or solutions until a suitable place in the document has been reached. I have something that sort of works but it's kind of clunky. For example, this works pretty well:

usepackage{environ}

NewEnviron{testb}{globalexpandafterletcsname barendcsnameBODY}

I use csname because the names are dynamically generated. This doesn't work so well if the command is anything other than BODY. For example, if I do letbar{BODY}, latex has an aneurysm (it binds bar to just {).

There is an alternate way of doing this, which is:

NewEnviron{testc}{globalexpandafteredefcsname fooendcsname{BODY}}

This sort of works. The following thing works:
begin{testc}hiend{testc}, but this gives LaTeX an aneurysm: begin{testc}bf hiend{testc}. (The error message is incomplete). I tried to debug it myself but I get hopelessly tangled up in packages that I don't understand. If you do it with tiny instead of bf you get the wildly different LaTeX error message: ! TeX capacity exceeded, sorry [input stack size=5000].

How can I squirrel away for later, not just BODY, but some sophisticated combination of BODY and other stuff, while having it robust to what's between the begin and end?

Edit: ok multiple solutions below

With unexpanded you can avoid worrying about protected@xdef.

documentclass{article}

usepackage{environ}



NewEnviron{exercise}{%

  xdefsavedexercises{%

    unexpandedexpandafter{savedexercises}%

    noexpandbegin{printedexercise}%

    unexpandedexpandafter{BODY}%

    noexpandend{printedexercise}%

  }%

}

newcommand{printexercises}{%

  savedexercises

  gdefsavedexercises{}%

}

newcommand{savedexercises}{}



newtheorem{printedexercise}{Exercise}



begin{document}



Here we talk about addition and show that $1+1=2$.



begin{exercise}

Compute $1+2$

end{exercise}



Here we talk about integrals.



begin{exercise}

Compute the following integrals:

begin{itemize}

item $displaystyleint_0^x e^{-t^2},dt$



item $displaystyleint_1^x frac{e^t}{t},dt$, for $t>0$.

end{itemize}

end{exercise}



Now we can print the exercises.



printexercises



end{document}

I used newtheorem just for the example.

With xparse released 2019-03-05 or later:

documentclass{article}

usepackage{xparse}



ExplSyntaxOn



NewDocumentEnvironment{exercise}{+b}

 {

  tl_gput_right:Nn g_loisel_exercises_tl

   {

    begin{printedexercise}

    #1

    end{printedexercise}

   }

 }{}



NewDocumentCommand{printexercises}{}

 {

  tl_use:N g_loisel_exercises_tl

  tl_gclear:N g_loisel_exercises_tl

 }



tl_new:N g_loisel_exercises_tl



ExplSyntaxOff



newtheorem{printedexercise}{Exercise}



begin{document}



Here we talk about addition and show that $1+1=2$.



begin{exercise}

Compute $1+2$

end{exercise}



Here we talk about integrals.



begin{exercise}

Compute the following integrals:

begin{itemize}

item $displaystyleint_0^x e^{-t^2},dt$



item $displaystyleint_1^x frac{e^t}{t},dt$, for $t>0$.

end{itemize}

end{exercise}



Now we can print the exercises.



printexercises



end{document}

If you want BODY to be saved then add stuff dynamically, you are likely best using two macros:

documentclass{article}

usepackage{environ}

NewEnviron{testb}{%

  globalexpandafterletcsname barendcsnameBODY

  expandafterxdefcsname barplusendcsname{%

    expandafternoexpandcsname barendcsname

    noexpandbf Hi

  }%

}

begin{document}



begin{testb}

  bfseries

  Hi

end{testb}

showbarplus

end{document}

If you want to avoid using BODY you could use xparse

documentclass{article}

usepackage{xparse}

NewDocumentEnvironment{testb}{+b}{expandaftergdefcsname barendcsname{#1}}{}

begin{document}



begin{testb}

  bfseries

  Hi

end{testb}

showbar

end{document}

A trick for triggering expansion is using romannumeral:

When due to romannumeral (La)TeX does gather together a sequence of digits trailed by a space as the number which it has to convert, expandable tokens get expanded.

When in the end a number is gathered together which is not positive, as the result of the conversion (La)TeX will not deliver any token at all.

Thus one can nicely (ab?)use romannumeral for triggering a lot of expansion-work and flipping-arguments-around-work as long as one ensures that in the end romannumeral will not find a positive number.

Here is a variant of egreg's answer which does with romannumeral and exchange instead of xdef and unexpanded.

documentclass{article}

usepackage{environ}



newcommandexchange[2]{#2#1}



NewEnviron{exercise}{%

  expandaftergdefexpandaftersavedexercisesexpandafter{%

    romannumeral0expandafterexchangeexpandafter{BODY}{%

      exchange{ }{expandafter}savedexercises

      begin{printedexercise}%

    }%

    end{printedexercise}%

  }%

}

newcommand{printexercises}{%

  savedexercises

  gdefsavedexercises{}%

}

newcommand{savedexercises}{}



newtheorem{printedexercise}{Exercise}



begin{document}



Here we talk about addition and show that $1+1=2$.



begin{exercise}

Compute $1+2$

end{exercise}



Here we talk about integrals.



begin{exercise}

Compute the following integrals:

begin{itemize}

item $displaystyleint_0^x e^{-t^2},dt$



item $displaystyleint_1^x frac{e^t}{t},dt$, for $t>0$.

end{itemize}

end{exercise}



Now we can print the exercises.



printexercises



end{document}

If you wish to wrap the name of the macro that is to be defined into csname..endcsname, i.e., if you wish to use csname savedexercisesendcsname rather than savedexercises, you can take advantage of the fact that (La)TeX does expand expandable tokens while due to csname gathering the name of a control sequence token and hereby searching for the matching endcsname:

documentclass{article}

usepackage{environ}



newcommandexchange[2]{#2#1}



NewEnviron{exercise}{%

  expandaftergdefcsname savedexercisesexpandafterendcsnameexpandafter{%

    romannumeral0expandafterexchangeexpandafter{BODY}{%

      exchange{ }{expandafterexpandafterexpandafter}csname savedexercisesendcsname

      begin{printedexercise}%

    }%

    end{printedexercise}%

  }%

}

newcommand{printexercises}{%

  csname savedexercisesendcsname

  expandaftergdefcsname savedexercisesendcsname{}%

}

expandafternewcommandexpandafter{csname savedexercisesendcsname}{}



newtheorem{printedexercise}{Exercise}



begin{document}



Here we talk about addition and show that $1+1=2$.



begin{exercise}

Compute $1+2$

end{exercise}



Here we talk about integrals.



begin{exercise}

Compute the following integrals:

begin{itemize}

item $displaystyleint_0^x e^{-t^2},dt$



item $displaystyleint_1^x frac{e^t}{t},dt$, for $t>0$.

end{itemize}

end{exercise}



Now we can print the exercises.



printexercises



end{document}

Be aware that with the approaches presented by now you cannot use printexercises for having exercises occur in arbitrary places. You can have exercises occur only in places of the document which in the source correspond to places behind the exercise-environments.

Perhaps an environment which does read its content under verbatim-catcode-régime for unexpanded-writing it to .aux-file in a way where from the .aux-file it gets read back under verbatim-catcode-régime also for defining a macro where scantokens will be applied to, and thus some sort of re-implementation of the label-ref-mechanism or of the tableofcontents-mechanism for verbatimized stuff might make it possible to make exercises printable throughout the entire document.

Implementing such a mechanism might be a nice challenge. But before taking that into consideration at all, exact information is needed on the intended usage and the desired "user-interface", i.e., what additional things you wish to be able to specify in which ways, etc, ...