If you’re going to import Data.Semigroup as a sort of backwards compatibility shim, then I don’t think you should import it qualified. To avoid the ambiguity error, how about “import Data.Monoid hiding ((<>))”? On Mon, Apr 30, 2018 at 05:57 Joey Hess wrote:

I tried to follow the instructions at https://prime.haskell.org/wiki/Libraries/Proposals/SemigroupMonoid and I got a <<loop>> when my code runs with ghc 8.2.2.

My code looked like this:

import Data.Monoid import Prelude

data InfoVal v = NoInfoVal | InfoVal v deriving (Show)

instance Monoid (InfoVal v) where mempty = NoInfoVal mappend _ v@(InfoVal _) = v mappend v NoInfoVal = v

Following the "recommended variant" for compatible code, I changed that too:

{-# LANGUAGE CPP #-}

import Data.Monoid import qualified Data.Semigroup as Sem import Prelude

data InfoVal v = NoInfoVal | InfoVal v deriving (Show)

instance Monoid (InfoVal v) where mempty = NoInfoVal #if !(MIN_VERSION_base(4,11,0)) mappend = (<>) #endif

instance Sem.Semigroup (InfoVal v) where _ <> v@(InfoVal _) = v v <> NoInfoVal = v

This loops because <> comes from Data.Monoid not from Data.Semigroup, so mappend = mappend.

Note that I diverged slightly from the instructions to get to this wrong code; I imported Data.Semigroup qualified. Without the qualification, the above code fails to compile, with Ambiguous occurrence ‘<>’

So, the transition instructions don't produce code that ghc 8.2.2 can build, and when the obvious fix is made to get it to compile, it compiles into a loop, that is *not* a loop when compiled with newer versions of ghc.

-- see shy jo _______________________________________________ Haskell-Cafe mailing list To (un)subscribe, modify options or view archives go to: http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe Only members subscribed via the mailman list are allowed to post.

-- -- Dan Burton