[Haskell-cafe] To [] Or Not To []

Johannes Waldmann johannes.waldmann at htwk-leipzig.de
Mon Mar 13 10:43:02 UTC 2017

Hi Olaf -

> I'd be interested whether there is a way to check 
> which of my lists in the source code the compiler managed to "deforest" away. 
> Which intermediate files should I look at? What are the tools to inspect?

Good question, and I don't know an easy answer.

The general advice is running ghc with "-ddump-simpl"
but I find it quite challenging to scan the output.

Here is a simple case where it works:

$ cat Fuse.hs

main = print $ sum $ map (^2) [1 .. 1000 :: Int]

$ ghc -fforce-recomp -ddump-simpl -O0 Fuse.hs

no optimisation - shows that "main" calls "map" etc.

$ ghc -fforce-recomp -ddump-simpl -O2 Fuse.hs

fusion works - nice non-allocating inner loop
(lists gone, and Int replaced by Int#)

Rec {
-- RHS size: {terms: 18, types: 3, coercions: 0}
Main.$wgo [InlPrag=[0], Occ=LoopBreaker]
  :: GHC.Prim.Int# -> GHC.Prim.Int# -> GHC.Prim.Int#
[GblId, Arity=2, Caf=NoCafRefs, Str=DmdType <S,1*U><S,U>]
Main.$wgo =
  \ (w_s5kV :: GHC.Prim.Int#) (ww_s5kZ :: GHC.Prim.Int#) ->
    case w_s5kV of wild_Xn {
      __DEFAULT ->
          (GHC.Prim.+# wild_Xn 1#)
          (GHC.Prim.+# ww_s5kZ (GHC.Prim.*# wild_Xn wild_Xn));
      1000# -> GHC.Prim.+# ww_s5kZ 1000000#
end Rec }

but the challenge is to find the path from "main" to that,
wading through several other functions that may or may not be related.

I can imagine a source annotation like "in the code compiled
from this function  f,  that constructor  C  should never be called"
but this is certainly not easy. Do we really mean "never", or do we mean
"only a bounded number of times" (that is, not in the inner loop).
Perhaps there is no code for  f  itself, because it gets inlined.

But yes, *some* automated analysis (and human-readable print-out)
of the code after simplification would be nice.

This could be done as a compiler plug-in?

- J.

More information about the Haskell-Cafe mailing list