incompatible with record fields that are actually duplicated -- "symbol already defined"

[peterwicksstringfield]

barbies-th works fine with DuplicateRecordFields enabled (thanks to the name unmangling patches); but it can't handle record fields that are actually duplicated. Possibly this is an accepted limitation, but it would be nice to document it I think. The error message is not very friendly:

... Assembler messages: ... Error: symbol `testlibzm0zi1zi0zi0zminplace_Lib_qux_info' is already defined ... testlibzm0zi1zi0zi0zminplace_Lib_qux_info

@Elvecent reported this as well, at the bottom of the thread in #3.

GHC 8.10.7
barbies == 2.0.3.1
barbies-th == 0.1.10

Example:

{-# LANGUAGE TemplateHaskell #-}
{-# LANGUAGE FlexibleInstances #-}
{-# LANGUAGE TypeFamilies #-}
{-# LANGUAGE DuplicateRecordFields #-}

module Lib where

import Barbies.TH (declareBareB)

declareBareB
  [d|
    data Foo = Foo
      { qux :: Integer
      }
    |]

declareBareB
  [d|
    data Bar = Bar
      { qux :: Integer
      }
    |]

cabal-version:   3.4
name:            testlib
version:         0.1.0.0
build-type:      Simple

common warnings
    ghc-options: -Wall

library
    import:           warnings
    exposed-modules:  Lib
    build-depends:
        base ^>=4.14.3.0,
        barbies == 2.0.3.1,
        barbies-th == 0.1.10
    hs-source-dirs:   src
    default-language: Haskell2010

peter@gtower:~/2022october8_reproduce_barbies_error$ cabal build
Build profile: -w ghc-8.10.7 -O1
In order, the following will be built (use -v for more details):
 - testlib-0.1.0.0 (lib) (first run)
Preprocessing library for testlib-0.1.0.0..
Building library for testlib-0.1.0.0..
[1 of 1] Compiling Lib              ( src/Lib.hs, /home/peter/2022october8_reproduce_barbies_error/dist-newstyle/build/x86_64-linux/ghc-8.10.7/testlib-0.1.0.0/build/Lib.o, /home/peter/2022october8_reproduce_barbies_error/dist-newstyle/build/x86_64-linux/ghc-8.10.7/testlib-0.1.0.0/build/Lib.dyn_o )
/tmp/ghc80775_0/ghc_2.s: Assembler messages:

/tmp/ghc80775_0/ghc_2.s:1298:0: error:
     Error: symbol `testlibzm0zi1zi0zi0zminplace_Lib_qux_info' is already defined
     |
1298 | testlibzm0zi1zi0zi0zminplace_Lib_qux_info:
     | ^

/tmp/ghc80775_0/ghc_2.s:1329:0: error:
     Error: symbol `testlibzm0zi1zi0zi0zminplace_Lib_qux_closure' is already defined
     |
1329 | testlibzm0zi1zi0zi0zminplace_Lib_qux_closure:
     | ^
`gcc' failed in phase `Assembler'. (Exit code: 1)

I tried to minimize the testcase further by expanding the splices and minimizing the generated code, but I was unable to achieve further minimization. I expanded one splice using the code-action offered by HLS. I found that the generated code is referencing a symbol called (Unit :: (Type -> Constraint) -> Type -> Constraint). I stubbed out such a type like this:

{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE ConstraintKinds #-}
{-# LANGUAGE UndecidableSuperClasses #-}
import Data.Kind (Constraint, Type)
class c t => Unit (c :: Type -> Constraint) (t :: Type) where
  {- empty body -}

When I added that bit, the generated code started compiling. That is, without the "symbol already defined" error.

Thanks for the library, seems cool.

[fumieval]

Thank you for reporting this, I was able to reproduce this on GHC 9.2.4 too. That's a shame... This looks like a GHC bug because such error messages are most likely not intended to be shown to users.

[fumieval]

There's a GHC ticket: DuplicateRecordFields doesn't work with TH-generated records