On 8/02/2011, at 3:47 AM, Gábor Lehel wrote:
>
> I dunno. As a language extension, would - let's call it BangTypes - be
> backwards-incompatible in any way?
Let's look at an intermediate step first, BangFunctions.
What this does is to say that there are two versions of "->":
t1 -> t2
What we have right now, which might be evaluated or not.
!t1 -> t2
The function wants its argument evaluated. Suppose f is a
value of this type. Then a use of f is rather like a
use of (\x -> x `seq` f x) and we can already write that.
Now if you write
f :: !t1 -> t2
f p1 = e1
...
f pn = en
you're making the function strict whether it would have been strict
or lazy. But again, with BangPatterns we can already do that:
f !(p1) = e1
...
f !(pn) = en
The advantage of BangPatterns is that they can be precisely and
selectively located.
The advantages of BangFunctions include
- the forced strictness is part of the function's (published)
*interface*, not its implementation
- the question of what happens if some patterns for an argument
are banged and some are not does not arise (I *think* that this
can avoid some mistakes)
- it's compatible with BangTypes but simpler.
So in some sense there is (now) nothing new here *except* putting
the information where people can easily see it.
BangTypes could be rather more complicated. Clean 2 offers
lazy, head strict spine lazy, head lazy spine strict, head and spine
strict, head unboxed spine lazy, head unboxed spine strict
for lists, which are all different types; it also offers
strictness-polymorphic lists. I never actually made use of this
because having so many kinds of list made my head spin.
Roughly speading, Clean 1 had BangFunctions, Clean 2 BangTypes.
One of the things that makes me wary of BangPatterns is that it seems
as though it's headed in a BangTypes kind of direction.
Oh, by the way, I got the Clean syntax wrong.
![a] means "lazy list evaluated to WHNF";
[a!] means "value is spine strict".
_______________________________________________
Haskell-Cafe mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/haskell-cafe
