GitLab

Consider three + one modules:

{-# LANGUAGE TypeFamilies #-}
module DF where

data family DF a

{-# LANGUAGE TypeFamilies #-}
module DI_A where
import DF

data instance DF [a] = DF_A a

{-# LANGUAGE TypeFamilies #-}
module DI_B where
import DF

data instance DF [a] = DF_B a

module Main where

import DF
import DI_A
import DI_B

main :: IO ()
main = print 'x'

When compiling or interpreting this, I get:

% ghci-9.2.0.20210422 -Wall Main.hs
GHCi, version 9.2.0.20210422: https://www.haskell.org/ghc/  :? for help
[1 of 4] Compiling DF               ( DF.hs, interpreted )
[2 of 4] Compiling DI_A             ( DI_A.hs, interpreted )
[3 of 4] Compiling DI_B             ( DI_B.hs, interpreted )
[4 of 4] Compiling Main             ( Main.hs, interpreted )

DI_A.hs:5:15: error:
    Conflicting family instance declarations:
      DF [a] -- Defined at DI_A.hs:5:15
      DF [a] -- Defined at DI_B.hs:5:15
  |
5 | data instance DF [a] = DF_A a
  |               ^^
Failed, three modules loaded.

Note: DI_A was checked without any warnings.

Then (potentially a lot later) the conflict is found and reported.

The explanation is probably because type familyies use the same check, and there defining instances with the same RHS in the different modules is fine, as there is no conflict (thought in that case warning would still make sense!). However with data familyies that will never be the case, the conflict is guaranteed.

I expect some orphan warning already for the new datatype defined.

We get orphan warnings for the instances if we do data instance DF [a] = DF_A a deriving Show, as neither Show nor DF is defined in that module, but we can argue that DF [a] = DF_A is defined in this module, but the data-instance itself is orphan!