Javier, > The "expandability" of \add@accent has not changed. It is still > expandable, if its definition is: > > \def\add@accent#1#2{{% > \setbox\@tempboxa\hbox{#2% > \global\mathchardef\accent@spacefactor\spacefactor}% > \accent#1 #2}\spacefactor\accent@spacefactor} > > The problem is that \accent@spacefactor is not yet defined in \edef's > prior to an actual use of any accent. That is a LaTeX bug which can be > fixed by letting \accent@spacefactor to \relax or by defining > it with \mathchardef (in both cases it expands into itself). this is an interesting point (what you are saying is some assignments are harmless as they do not fail horribly even though they do not expand really and all we have to do in that particular case is making sure that we keep the status of that macro expansion this way). agreed, but as you will be aware this wasn't really the point that i was trying to make (and it doesn't make my point invalid either). the point was that there are macros containing dangerous assignments (mostly internal \def's or \let's) which will die horribly if put into an \edef and my claim is that without the LaTeX protection mechanism (which essentially prohibits the use of \edef everywhere where the content of \edef is not fully controlled by the macro writer) there is no cure for this problem: a) you need macros containing dangerous assignments (containing \let say) b) current coding practice in TeX does need the functionality of \edef now the LaTeX solution right now is to say: don't use edef use \protected@edef and friends and all is well. my question was, are there better approaches? (the plain TeX approach by the way is: know your macros and don't use them in a way that you end up expanding such a dangerous beast inside an \edef :-) coming back to your example: > Example: > > \documentclass{minimal} > > \edef\agrave{\`a} % Doesn't work > \setbox0\hbox{\`a} % We use an accent > \edef\agrave{\`a} % Now it does work (but \agrave looks horrible! :-) ) run like that it unfortunately doesn't show your point but rather shows mine as you will get: [ some tracing omitted ] \?\` ->\UseTextAccent {OT1}\` \UseTextAccent #1#2#3->\let \@curr@enc \cf@encoding \@use@text@encoding {#1}#2{ \@use@text@encoding \@curr@enc #3}\@use@text@encoding \@curr@enc #1<-OT1 #2<-\` #3<-\fi {undefined} ! Undefined control sequence. \UseTextAccent #1#2#3->\let \@curr@enc \cf@encoding \@use@text@encoding {#1}... in other words we never reach the case of the \accent@spacefactor but die in \UseTextAccent which already contains a \let and does expand in this way if used in \edef in contrast to \protected@edef (where it would behave correctly) so i think my questions are still valid and i wonder if i get some more feedback best frank ps which does not mean that we shouldn't fix the kernel to have a definition for \accent@spacefactor