> I have discovered a bug in the handling of hi files in the
> presence of
> -fglasgow-exts. Below is a module, MyModule.hs, which uses
> Word internally
> but exports a function of type Integer->Integer. There is
> also a Main.hs
> module which uses this function, but of course does not
> import Word since it
> doesn't need to.
>
> Under ghc-3.02 I get the expected behaviour: I give
> -fglasgow-exts when
> compiling MyModule.hs, but not when compiling Main.hs. I do,
> however, need to
> give -fglasgow-exts when linking.
>
> Under ghc-4.01 (well, the head of the CVS tree as of
> yesterday), however, I
> need to give -fglasgow-exts when compiling Main.hs, too.
> This is surely wrong!
Thanks for the report, but I think GHC is behaving as expected here.
Compiling with -fglasgow-exts gives you access to the extended libraries,
and if you compile one module with -fglasgow-exts you can reasonably expect
to have to compile any modules that refer to it with -fglasgow-exts too.
This probably needs documenting somewhere.
The low-level reason in this case was that the interface file recorded that
Word contained some instances, and effectively re-exported them from
MyModule.hs.
I'm not sure why it went through with 3.02, that needs more investigation.
However, 3.03 behaves the same as 4.01.
Cheers,
Simon
--
Simon Marlow
Microsoft Research Ltd., Cambridge, U.K.
