Frank Mittelbach wrote:

> I agree with you that arg spec is getting (and got) in the wrong direction,
> too many letters too much encoded in convention. This needs one more serious
> round of discussions and thought. However, beside "perfect" conventions one
> also always has to think about the look and feel of the resulting code, e.g.,
> even if it is more logical perhaps to use "t" and "f" for true and false
> branches "T" and "F" work better because they stand out and that is what you
> need for these arguments.

From writing the few things I've tried in expl3, I'd say the the
arg-specs I've use most (by some way) are N, T, F, n and c, plus p in
\def:Npn (although nowhere else).  Probably o and x pop up in similar
numbers, with everything else very rarely: I've used f a couple of
times, d perhaps once or twice and w for things like \tlp_map_break:w
(where I'm not sure it helps).  (I've also been forced to use a couple
of :D primitives, but I'm sure that this will be sorted.)

So one could possibly rationalise to:

N, T, F, n, c (all unchanged), x (perhaps also covering current X), w
(covering current w and p) and e (or s = special) (other expansion, read
docs, to cover d, f, current E, ...), plus of course D.

That is nine letters, with e and w covering a multitude of things, i.e.
these are both "special" and the user needs to read the rules for using
these functions.  I'd say that ten or fewer letters would be a worthy aim.
-- 
Joseph Wright