Heiko Oberdiek a écrit :
>> (This uses \@nil.) Putting the second split into a macro to test it against 
>> \@empty is safe, but one might dislike it as "slow".
> 
> I prefer "safe".
> 
I agree.

> An expandable test could be used, e.g.:
>   \ifx\\##2\\% or something else as \\

Is it "allowed" to use e-TeX commands inside the kernel? If so,

\expandafter\ifx\expandafter\\\detokenize{##2}\\% or something else as \\

is the safest test, as I'm sure you know.

Anyway, depending on the intended use of \in@, certain resctrictions (such as
"no unbalanced \if" or "no # token" or "no \@nil token" are probably acceptable,
as long as they are properly documented.

Manuel.