Combinators for n-ary functions (use apply instead of funcall)

[vapniks]

N-ary versions of -on, -not, -orfn, -andfn
New combinators -rotate-args and -reverse-args for rotating/reversing
arglist of n-ary functions.

Altered items:
-on2, -rotate-args, -reverse-args, -not2, -orfn2, -andfn2,

[vapniks]

You might want to replace the original -on, -not, -orfn, -andfn with the new versions

[magnars]

Thanks! I'll ping @Fuco1 for this one.

[Fuco1]

Personally I find &rest and apply a bit meh and would prefer more explicit applicative style, but of course that's not easily possible in elisp.

The main issue here, however, is speed. Funcall is slow as it is but apply is much much slower. I did a quick benchmark calling

(-sort (-on '< 'string-to-int) numbers)
(-sort (-on2 '< 'string-to-int) numbers)

3000 times on a list of 1000 strings (all files byte-compiled) and the results are

-sort + -on  dash-test      1           7.989396416   7.989396416
-sort + -on2 dash-test      1           30.05434008   30.05434008

This is definitely something you could feel in the result product. I suspect the rest would be equally slowed down.

I can't imagine right now a use case for "ternary -on", but I'm sure there is. For the sake of speed, I'd therefore suggest we keep both variants. the -blabla2 versions could be named -blablan maybe? (to show they are n-ary), and it would be nice to add some note to docstring about the possible performance drawback, e.g. "for binary use, consider faster `-bla'".

Also we should add the types to all the functions... but I'm not sure how to express variable arguments. Maybe a.... So -on2 would be `(b... -> c) -> (a -> b) -> a... -> c

[vapniks]

How about adding fast binary versions with funcall since these are most likely to be used (e.g. (-orfn2 'string< 'string=) which is what prompted me to write those functions).
We could name the binary versions -orfn2, -andfn2, etc. and name the n-ary versions -orfnn, -andfnn, etc.

[Fuco1]

What you mean adding? The present functions all use funcall when possible. -on is mostly useful as binary combinator, -orfn and -andfn as unary (all filters take unary predicates)

Edit: Oh, you mean like adding binary, ternary, ... using funcall and then a generic one for n-ary use? Dunno if that wouldn't be too much noise.....

[vapniks]

or how about an extra parameter to indicate the arity? (implemented as macros)