Joseph Wright wrote:
> Okay, I'd suggest all in \toks_ (remembering that we'll need to pick up
> the primitives not the LaTeX2e versions for l3in2e). I guess we will
> need to do the same as LaTeX2e for the initex version of expl3, and
> remove direct access to the primitives in user space?

Thinking again, I'm not sure quite what is best.  Possibilities include
having a dedicated "every" name-space:

\everyeof	\l_every_eof_toks
\everypar	\l_every_par_toks
\everymath	\l_every_math_toks

Splitting into different modules:

\everyeof	\l_ior_eof_toks
\everypar	\l_<some-module>_par_toks
\everymath	\l_<some-module>_math_toks

(good modules for these are not obvious)

Putting everything in the toks module:

\everyeof	\l_toks_eof_toks
\everypar	\l_toks_par_toks
\everymath	\l_toks_math_toks

Over all, I think I prefer the first option, as these do all go
together. (If we go for _every_, do we have a very short module
"l3every" for this? If not, where do these things go?)

There is also the question of whether to provide a wrapper for
assignment to these, something like:

\cs_new:Npn { \ior_eof_set:n } #1 {
  \toks_set:Nn \l_ior_eof_toks {#1}
}

or

\cs_new:Npn { \every_eof_set:n } #1 {
  \toks_set:Nn \l_every_eof_toks {#1}
}

I'd probably argue that this is unnecessary.  The \every... toks are all
quite low-level, and I think that it's perfectly logical if they are
given _toks names that they are then treated like any other toks.
-- 
Joseph Wright