Joseph Wright a écrit :
[log in to unmask]" type="cite"> 
[log in to unmask] wrote:
  
\begingroup
\catcode`\:=13
\catcode`\.=13\relax
\gdef\chemin#1{%
\begingroup
\catcode`\:=13
\catcode`\.=13\relax
\let.\cheminsommet
\let:\cheminface
\endlinechar=-1
\everyeof{}%
\scantokens{#1}%
\endgroup
}
\endgroup
    

A couple of things here. First, active characters tend to be bad news. In my own siunitx package, I've moved from using an approach like yours in v1 to trying a different system in v2, based on \tl_replace_all_in:Nnn:

  \cs_set:Nn \my_int_function:n #1 {
    \tl_set:Nn \l_my_tmp_tl {#1}
    \tl_replace_all_in:Nnn \l_my_tmp_tl { . } { \cheminsommet }
    \tl_replace_all_in:Nnn \l_my_tmp_tl { : } { \cheminface   }
   % Do more stuff with the input
  }

The reason for this is two-fold. First, active characters can cause a lot of problems (see what happens when two packages try to do something like this).
Note that those characters would be set active only in the call of the function; outside of it the catcode is preserved (or so I think). In fact, in my first version of the \chemin macro, I did some kind of replace, but without expl3 that meant an ugly thing with a lot of ad-hoc looping code. Won't search/replace be slower than setting . and : to macros ? Anyway, the question here is more to play with expl3 than anything else, getting rid of chemin as it is now isn't a priority at all... :)
[log in to unmask]" type="cite"> 
Second, the \tl_replace_all_in:Nnn approach means that there is only one level of replacement, which is often what people expect if they protect characters. I think the general feeling is that the team are aiming to avoid active characters as far as possible.
Fair enough.
[log in to unmask]" type="cite"> 
If you do want to use \tl_rescan:Nnx, what is wrong with

\group_begin:
  \char_make_active:N \:
  \char_make_active:N \.
  \cs_gset:Npn \my_int_function:n #1 {
    \tl_set_rescan:Nnx \l_my_tmp_tl
      {
        \char_make_active:N \:
        \char_make_active:N \.
        \cs_set_eq:NN . \cheminsommet
        \cs_set_eq:NN : \cheminface 
      } 
      {#1}
    % Stuff with \l_my_tmp_tl
  }
\group_end:

or

\group_begin:
  \char_make_active:N \:
  \char_make_active:N \.
  \cs_gset_eq:NN . \cheminsommet
  \cs_gset_eq:NN : \cheminface 
  \cs_gset:Npn \my_int_function:n #1 {
    \tl_set_rescan:Nnn \l_my_tmp_tl
      {
        \char_make_active:N \:
        \char_make_active:N \.
      } 
      {#1}
    % Stuff with \l_my_tmp_tl
  }
\group_end:

(The later makes active . and : globally available, so works without x-type expansion of the tl. I'd personally go for the first of these two solutions as it again is independent of what others do.)
I think I have tried such solutions, wouldn't TeX complain about nonexistent \char_make_active because a colon cannot anymore be in the name of a macro ?

Julien