literal functions [was: Re: dash.el]

[Joost Kremers]

On Wed, May 13 2020, Richard Stallman wrote:
>   > In Clojure, a function literal is written with `#(...)`, and 
>   > it 
>   > supports `%n` for arguments, where n=1,2,3... (Actually, I 
>   > don't 
>   > know if Clojure supports `%10` and up or if it just goes up 
>   > to 
>   > `%9`, but I guess that doesn't matter much). A single `%` is 
>   > synonymous with `%1`, which is used when there's only one 
>   > argument. Also supported is `%&`, which is similar to `&rest 
>   > <var>` in Elisp, being bound to a list containing the 
>   > remaining 
>   > arguments.
>
> This is rather shell-like.  Can we find a nicer one?

Not sure what a nicer syntax would look like. If nicer means more 
verbose,  function literals might become pointless, because their 
compactness is IMHO the main reason for having them.

Using Unicode would be nice:

   (seq-reduce λ(* 2 (+ α β)) (number-sequence 1 10) 0)

But that runs into obvious problems (font support, input method). 

Personally, I wouldn't be adverse to adopting the Clojure syntax 
for function literals wholesale, except that `#(...)` already has 
a meaning for the reader. Perhaps `#f(...)` can be used instead, 
since it follows the existing pattern (`#` plus a single 
character).

> It doesn't specify how many args the literal function should 
> accept.
> Maybe there is no need for that, but if we want to do that, how 
> could we?

Something like

   #f2(* 2 (+ %1 %2))

with the 2 after #f indicating the number of arguments? Doable, I 
guess. But I'm not sure it gets you anything, unless you also find 
a way to indicate the use of an `&rest` argument. If you don't, 
you'd need to go through the entire form anyway to search for it, 
and then you might as well collect the other arguments as well.

You could of course decide not to support `&rest` arguments. That 
would limit a function literal's usefulness somewhat, but that 
might be an acceptable compromise.

OTOH you might want to use gensyms to represent the arguments in 
the expanded lambda form. In that case you need to go through the 
entire form as well, so no need to indicate the number of 
arguments.

A related question is what to do about function literals such as 
`#f(* 2 (+ %1 %3))`. If the reader collects all %n arguments and 
puts them in the argument list in the right order, this doesn't 
have to be an error. OTOH if you use the #f2 syntax and let the 
reader create an argument list without checking, a form such as , 
`#f2(* 2 (+ %1 %3))` would yield a void variable error upon 
execution.

My personal preference would be to just use `#f(...)` and not 
indicate the number of arguments.

> Is there ever a need for nested literal functions?  To implement 
> them
> fully, we would want to use different arg names at each level.

Which is probably why Clojure doesn't allow nesting them. Function 
literals are not meant to replace lambdas in every context, they 
are mainly meant as short, one-off function arguments to 
higher-order predicates, so it's an acceptable compromise that 
they're not nestable. 

> Does anyone think this is a bad idea?

I obviously don't. :-)

--
Joost Kremers
Life has its moments