cgrep 8.0.0/8.1.0 doesn't build

[felixonmars]

[19 of 29] Compiling CGrep.Output     ( src/CGrep/Output.hs, dist/build/cgrep/cgrep-tmp/CGrep/Output.dyn_o )

src/CGrep/Output.hs:173:12: error:
    • Illegal equational constraint GHC.Exts.Item a ~ B.Builder
      (Use GADTs or TypeFamilies to permit this)
    • When checking the inferred type
        mkMatch :: forall {a}.
                   (GHC.Exts.Item a ~ B.Builder, Semigroup a, GHC.Exts.IsList a) =>
                   a -> Output -> a
      In an equation for ‘jsonOutput’:
          jsonOutput outs
            = pure
                $ mconcat . intersperse (B.char8 '\n')
                    $ [B.byteString "{ \"file\":\""
                         <> B.byteString fname <> B.byteString "\", \"matches\":["]
                        <>
                          [mconcat $ intersperse (B.char8 ',') (foldl mkMatch ... outs)]
                            <> [B.byteString "]}"]
            where
                fname | (Output f _ _ _) <- head outs = f
                mkJToken chunk
                  = B.byteString "{ \"col\":"
                      <>
                        B.int64Dec (cOffset chunk)
                          <>
                            B.byteString ", \"token\":\""
                              <> B.byteString (cToken chunk) <> B.byteString "\" }"
                mkMatch xs (Output _ n _ ts) = xs <> [...]
    |
173 |            mkMatch xs (Output _ n _ ts) =
    |            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^...

I'm building with GHC 9.2.8 on Arch Linux x86_64.

[loqs]

Triggered by 3a6a67c8ccf99472cf85f06970c861130a4aaa34 cross checked by reverting that commit from v8.0.0 and v8.1.0 and both build. Edit: Work around add GADTs or TypeFamilies to Default-Extensions:

--- a/cgrep.cabal
+++ b/cgrep.cabal
@@ -38,6 +38,7 @@ Executable cgrep
   Hs-Source-Dirs:      src
   Default-Extensions: FlexibleContexts
                       FlexibleInstances
+                      GADTs
                       GeneralisedNewtypeDeriving
                       DerivingStrategies
                       MultiWayIf

[leedagee]

It's the OverloadedLists feature caused this. By enabling it, the type became ambiguous for the compiler. Possible solutions may include:

• Turning off OverloadedLists
• Turning on GADTs or TypeFamilies
• Annotating the type to be a simple [Builder] instead of a IsList type constraint

The last one may be the most straightforward one which causes minimal side-effects.

diff --git a/src/CGrep/Output.hs b/src/CGrep/Output.hs
index 7fed0da..9fbbf77 100644
--- a/src/CGrep/Output.hs
+++ b/src/CGrep/Output.hs
@@ -170,6 +170,7 @@ jsonOutput outs = pure $ mconcat . intersperse (B.char8 '\n') $
         [ mconcat $ intersperse (B.char8 ',') (foldl mkMatch [] outs) ] <> [B.byteString "]}"]
      where fname | (Output f _ _ _) <- head outs = f
            mkJToken chunk = B.byteString "{ \"col\":" <> B.int64Dec (cOffset chunk) <> B.byteString ", \"token\":\"" <> B.byteString (cToken chunk) <> B.byteString "\" }"
+           mkMatch :: [B.Builder] -> Output -> [B.Builder]
            mkMatch xs (Output _ n _ ts) =
                xs <> [B.byteString "{ \"row\": " <> B.int64Dec n <> B.byteString ", \"tokens\":[" <>
                         mconcat (intersperse (B.byteString ",") (map mkJToken ts)) <> B.byteString "] }" ]

[loqs]

Could the reason this has not been detected / produced before be because GADTs is enabled by default in GHC2024?

[loqs]

@awgn do have you any preference on which solution to use to resolve the issue?

[loqs]

@felixonmars with your packaging of ghc 9.4.8 which has GADTs enabled by default the issue is no longer reproducible for me. Can you still reproduce it?

[felixonmars]

Indeed I can no longer reproduce it. Thanks for noticing!