[Haskell-cafe] Re: A tale of three shootout entries

[Simon Marlow]

Simon Peyton-Jones wrote:
> | There may well have been changes to the strictness analyser that make
> | some of the bangs (or most) unnecessary now. Also, its very likely
> | I didn't check all combinations of strict and lazy arguments for the
> | optimal evaluation strategy :)
> |
> | If it seems to be running consitently faster (and producing better Core
> | code), by all means submit. I don't think this is a ghc bug or anything
> | like that though: just overuse of bangs, leading to unnecessary work.
> 
> You might think that unnecessary bangs shouldn't lead to unnecessary work -- if GHC knows it's strict *and* you bang the argument, it should still only be evaluated once. But it can happen.  Consider
> 
>         f !xs = length xs
> 
> Even though 'length' will evaluate its argument, f nevertheless evaluates it too.  Bangs say "evaluate it now", like seq, because we may be trying to control space usage.  In this particular case it's silly, because the *first* thing length does is evaluate its argument, but that's not true of every strict function.
> 
> That's why I say it'd be good to have well-characterised examples.  It *may* be something like what I describe. Or it may be a silly omission somewhere.

A little addition to what Simon mentioned above: while it is definitely 
true that adding unnecessary bangs can cause a slowdown, the slowdown 
should be much less with 6.8.1 because in the common case each evaluation 
will be an inline test rather than an out-of-line indirect jump and return.

So, with 6.8.x, you should feel more free to sprinkle those bangs...

Cheers,
	Simon