w32 + haskell

gistfile1.hs

import Data.Bits.Extras
import Data.Hashable

{-
:info w32
w32 :: Integral a => a -> GHC.Word.Word32
  	-- Defined in ‘Data.Bits.Extras’
-}

filter (/= 0) (map (w32 . hash) ([1.0 .. 10000.0]::[Double]))

filter (== 0) (map (w32 . hash) ([1 .. 10000]::[Int]))

this is relevant because of how hashes are computed for use in hyperloglog

https://github.com/ekmett/hyperloglog/blob/master/src/Data/HyperLogLog/Type.hs#L137

I get why this is happening, not sure who to complain to.

To hash a Double (64-bit float), hashable casts it to a Word64 (64-bit unsigned int) with the same bits. That's very fast (roughly free) but you get hashes with the same bit layout as IEEE doubles:

ghci> import Data.Hashable
ghci> hash (8 :: Double) == 0x4020000000000000
True

That's relevant because w32 converts Int to Word32 by throwing out the most significant 32 bits. So

w32 (hash d) == 0

for a lot of "simple" Double values.