Separate compilation fails with existential records

[guest]

Consider M.hs:

> {-# OPTIONS_GHC -fglasgow-exts #-}
> module M where data D = forall a. C { l :: a }

Then consider Main.hs:

> import M
> main = print () where x = C { l = () }

A normal compilation works:

    $ ghc --make Main.hs
    [1 of 2] Compiling M                ( M.hs, M.o )
    [2 of 2] Compiling Main             ( Main.hs, Main.o )
    Linking Main ...

However, separate compilation does not:

    $ touch Main.hs
    $ ghc --make Main.hs
    [2 of 2] Compiling Main             ( Main.hs, Main.o )

    Main.hs:2:26:
        Constructor `C' does not have field `l'
        In the expression: C {l = ()}
        In the definition of `x': x = C {l = ()}
        In the definition of `main':
            main = print ()
                 where
                     x = C {l = ()}

Where should I start looking to fix this? :-)

Thanks, Audrey

[simonpj]

Audrey, when you added the "record fields in existentials" feature, you didn't arrange that the info is recorded in interface files.

I discovered and fixed this when cleaning up IfaceSyn? to add FC to GHC; but the fix is only in the HEAD. The patch is not easy to incorporate in the branch, because the FC patch was very big.

Fixing the 6.6 branch is possible, but would mean a change in interface file format, which we strongly discourage between patch-level releases.

So I'm inclined to leave it alone in the branch, and encourage you to try the HEAD.

[guest]

Understood. May I commit a doc change to explain that it's a known limitation, not expected to be lifted in the 6.6.x series, then?

Thanks, Audrey

[igloo]

Documenting this in the 6.6 branch, if not already done, sounds like a good idea.

[igloo]

I've documented this in the 6.6 branch users guide.