Conflicting definition in hs-boot file

[Feuerbach]

Foo.hs:

module Foo where

import Bar

data Foo a

Foo.hs-boot:

module Foo where

data Foo a

Bar.hs:

module Bar where
import {-# SOURCE #-} Foo

With GHC 7.8.2 I'm getting the error

Foo.hs-boot:3:1:
    Type constructor ‘Foo’ has conflicting definitions in the module
    and its hs-boot file
    Main module: type role Foo phantom
                 data Foo a
    Boot file:   data Foo a

[Feuerbach]

If I add a role annotation to the hs-boot file, this works.

My guess is that since the hs-boot doesn't have the information about the data constructors, it cannot infer the phantom role of the parameter.

Still, I wish I didn't have to deal with this. But if this isn't possible, at least the error message should suggest adding the role annotation to the hs-boot file.

[goldfire]

Replying to Feuerbach:

If I add a role annotation to the hs-boot file, this works

Yes, that's the solution. .

My guess is that since the hs-boot doesn't have the information about the data constructors, it cannot infer the phantom role of the parameter.

Yes. As documented here, in the last bullet point before 4.7.10, role inference in hs-boot files defaults to representational, the most common case.

Still, I wish I didn't have to deal with this. But if this isn't possible, at least the error message should suggest adding the role annotation to the hs-boot file.

Yes, absolutely. I will fix in due course, and clarify some of this in the manual as well.

[Richard Eisenberg <eir@…>]

In ec8781f063d246a79ce1d4eb207dbee4b6317c94/ghc:

Test #9204 in roles/should_fail/T9204

[Richard Eisenberg <eir@…>]

In ee0f34d53291a7223185f83c644a25b54ea16fab/ghc:

Fix #9204 by outputting extra info on boot file mismatch.

[skip ci] -- testsuite wibbles are in next commit

[Richard Eisenberg <eir@…>]

In 90a2bb6fc66e9341ae466ff3ff6c9da438e159c2/ghc:

Testsuite wibbles due to #9204

[phischu]

In the following example the complete data declaration happens to be present in the hs-boot file:

Foo.hs:

module Foo where

import Bar

data P a = P

Foo.hs-boot:

module Foo where

data P a = P

Bar.hs:

module Bar where
import {-# SOURCE #-} Foo

Could you now do role inference in the hs-boot file as well, so that the example compiles? My use-case is that I generate hs-boot files by copying whole declarations. Thank you.

[goldfire]

Yes, the suggestion in comment:8 is reasonable: Use phantom as the default role in hs-boot files when the definition is supplied. Even today, GHC does do role inference in hs-boot files -- it just uses a different default. But there is no reason to change the default if a definition is supplied.

Thanks for the suggestion.

However, because this is a user-facing change, it's not appropriate (in my opinion) to make the change for 7.10. And, it's just barely possible that this will break code. The following compiles today:

A.hs-boot:

module A where

data P a = P

A.hs:

module A where

type role P representational
data P a = P

Under the proposed change, this example would fail due to the missing role annotation in the hs-boot file. That said, I'm still in favor of the change, just documenting what would break.

[simonpj]

OK let's change this for 7.12, and document the change. Richard, you took ownership, so maybe do this whenever it's convenient to you.

Simon

[Richard Eisenberg <eir@…>]

In 72a927267d9c658a2e5d226a855702d348472516/ghc:

Change default roles in hs-boot files. (#9204)

Test case: roles/should_compile/T9204b

[goldfire]

This is a user-facing change, so wait until 7.12.

[thoughtpolice]

Milestone renamed