anonymous function + lable bug?

anonymous function + lable bug?

[Chris Hecker]

Is this a bug?

# let f = fun ~x -> x;;
val f : x:'a -> 'a = <fun>

# let g = function ~x -> x;;
Characters 17-18:
Syntax error

I thought these two were equivalent?

Chris

Re: anonymous function + lable bug?

[Jacques Garrigue]

From: Chris Hecker <checker@d6.com>

> Is this a bug?
> 
> # let f = fun ~x -> x;;
> val f : x:'a -> 'a = <fun>
> 
> # let g = function ~x -> x;;
> Characters 17-18:
> Syntax error
> 
> I thought these two were equivalent?

They are not:

Similar to the above example,
# let f = fun x y -> x;;
val f : 'a -> 'b -> 'a = <fun>
# let f = function x y -> x;;
Syntax error       ^

and the other way round,
# let f = fun Some x -> x;;
              ^^^^
The constructor Some expects 1 argument(s),
but is here applied to 0 argument(s)
# let f = function Some x -> x;;
val f : 'a option -> 'a = <fun>


The equivalence (in the no-label case) is between
        function p1 -> e1 | ... | pn -> en
and
        fun x -> match x with p1 -> e1 | ... | pn -> en

fun and function only become equivalent when the matching is trivial,
and there is no need to use an explicit match ... with
This is a true equivalence: the compiled code is identical, so you can
really think of function as a shorthand for the explicit form with
match, when you take only a single non-labelled argument.

It would be possible to add special rules to allow
        function ~l:p1 -> e1 | ... | ~l:pn -> en
This was actually the case in olabl, but it was very rarely used, and
complicates the parser and type checker.

Jacques
---------------------------------------------------------------------------
Jacques Garrigue      Kyoto University     garrigue at kurims.kyoto-u.ac.jp
		<A HREF=http://wwwfun.kurims.kyoto-u.ac.jp/~garrigue/>JG</A>

Re: anonymous function + lable bug?

[Sven LUTHER]

On Sun, Nov 05, 2000 at 05:11:11AM -0800, Chris Hecker wrote:
> 
> Is this a bug?
> 
> # let f = fun ~x -> x;;
> val f : x:'a -> 'a = <fun>
> 
> # let g = function ~x -> x;;
> Characters 17-18:
> Syntax error
> 
> I thought these two were equivalent?

No, function has only one argument, and can do more kind of polymorphism than 
fun which can take more than one argument as in :

let f = fun x y z -> x + y + z
f 5 6 7 evaluates to 18.

I think the use of function is preferable over fun, which is there for
compatibility reasons.

not sure what the ~ does, i think it is one of the label stuffs.

Try :

# let f ~x = x ;;
val f : x:'a -> 'a = <fun>

Don't know why there is a problem with the above though.

Friendly,

Sven Luther

Re: anonymous function + lable bug?

[Chris Hecker]

>> I thought these two were equivalent?
>They are not:

Right, I didn't mean to say they're completely equivalent (I've read the FAQ), just that I expected equivalence in that case.

>It would be possible to add special rules to allow
>        function ~l:p1 -> e1 | ... | ~l:pn -> en

Ah, now I see, it's because the parameter in function is actually the pattern in an implicit match?

Is there ever a reason to use function (assuming you don't mind parenthesizing constructors and whatnot in fun arguments)?  Does the implied match allow some simplification or optimization or something?  Why do they both exist?

As I said, I've read the FAQ, but I can't see why you'd ever use function when you've got fun...

Thanks,
Chris

Re: anonymous function + lable bug?

[Jacques Garrigue]

From: Chris Hecker <checker@d6.com>

> >> I thought these two were equivalent?
> >They are not:
> 
> Right, I didn't mean to say they're completely equivalent (I've read
> the FAQ), just that I expected equivalence in that case.
> 
> >It would be possible to add special rules to allow
> >        function ~l:p1 -> e1 | ... | ~l:pn -> en
> 
> Ah, now I see, it's because the parameter in function is actually
> the pattern in an implicit match?

Exactly.

> Is there ever a reason to use function (assuming you don't mind
> parenthesizing constructors and whatnot in fun arguments)?  Does the
> implied match allow some simplification or optimization or
> something?  Why do they both exist?

function is not necessary: you can replace it everywhere with fun and
match, and the code is exactly optimized the same way. Actually some
pass of the compiler expands function in fun+match.

However, in many cases function is a nice and intuitive short-hand for
directly matching on an argument. The reason for it to be here seems
mostly historical: one can define let in terms of fun (+ special
typing), and match in terms of function (no special typing
required). As I remember it, SML (/NJ?) was actually seeing case
(=match) as syntactic sugar for fn(=function).

Jacques Garrigue

Re: anonymous function + lable bug?

[Chris Hecker]

>No, function has only one argument, and can do more kind of polymorphism than 
>fun which can take more than one argument as in :

Is this actually true that function can be more polymorphic?  What's an example?

Chris