9
votes

In Haskell, is there ever a situation where for a data type

{-# LANGUAGE BangPatterns #-}
import Control.DeepSeq

data D = D Int

the instance

instance NFData D where
  rnf (D !_) = ()

can have a different effect than the instance with another outer !:

instance NFData D where
  rnf !(D !_) = ()

My research:

So I think no, these two are always equivalent, but I'm asking anyway to have one super clear answer to refer people to.

1

1 Answers

6
votes

Indeed this is correct. We can see what is evaluated using :sprint in GHCi, which shows us what thunks have been evaluated.

With no bang patterns:

λ data D = D Int
λ d1 = D 1
λ :sprint d1
d1 = _
λ f1 (D _) = 0
λ f1 d1
0
λ :sprint d1
d1 = <D> _ -- Only D evaluated

With an inner bang pattern:

λ d2 = D 2
λ :sprint d2
d2 = _
λ f2 (D !_) = 0
λ f2 d2
0
λ :sprint d2
d2 = <D> 2 -- Everything evaluated

With an outer bang pattern:

λ d3 = D 3
λ :sprint d3
d3 = _
λ f3 !(D _) = 0
λ f3 d3
0
λ :sprint d3
d3 = <D> _ -- Only D evaluated

With an inner and outer bang patterns:

λ d4 = D 4
λ :sprint d4
d4 = _
λ f4 !(D !_) = 0
λ f4 d4
0
λ :sprint d4
d4 = <D> 4 -- Everything evaluated

From this we can easily see that the patterns !(D !_) and (D !_) are equivalent, and moreover that patterns of the form !(D ...) are redundant.