This otoh: http://benchmarksgame.alioth.debian.org/u64q/program.php?tes...
>Foreign.Ptr (and companions) are hardly ever used in normal Haskell code unless you're optimizing for that last bit of performance<
Do you think "optimizing for that last bit of performance" doesn't happen in the wild?
>IMO they should at least use idiomatic code<
Provide some!
http://benchmarksgame.alioth.debian.org/play.php#contribute
GHC 7.8.2 installed, cabal 1.2 stuff available as-needed.
Of course it does, but the cost increases non-linearly and you have to find the right trade-off for the situation at hand. Do you want maintainability or absolute performance?
> Provide some!
Of course one could, but that would that accomplish? Everyone who thinks these benchmarks have value just looks at the fastest-performing version of $YOUR_LANG_BENCHMARK. Hint: Mostly it won't be idiomatic regardless of the value of $YOUR_LANG.
Do you think the source code for those optimized for the last bit of performance programs tells us something about that?
>Of course one could, but what would that accomplish?<
Could one? Perhaps there'd be complaints that your "idiomatic programs" were just naïve.
A passive website like the benchmarks game can never be enough, by itself, to educate and inform all (or even most) readers -- reading selectively is far too commonplace and errare humanum est.
The benchmarks game is a resource. There are enough examples that when someone draws a misinformed conclusion, we can usually show them a counter-example and ask them to think.
But only if someone contributes those (whatever they consider idiomatic to be) idiomatic programs as counter-examples.
FWIW, I don't Don Stewart[0] would consider that idiomatic either. He just knows a bit more than the rest of us about how to squeeze more performance out of GHC-compiled code and was willing to use non-idiomatic code to do so.
Also FWIW, personally I don't think language shootouts are very informative most of the time, but... IMO they should at least use idiomatic code -- otherwise they tell you even less than they already do about what real-life performance is like.
[0] Just an example, I know he didn't contribute the particular benchmark you linked to. He did contribute a few of the other programs AFAIK. I think the following sentences also apply to (essentially) all the Haskell shootout contributors.