> the example given compiles

Surprising! I see your point about the types working out, but this also requires the additional assumption that the module type defined by ast.mli will be ascribed specifically to a module named Ast. I suppose this is consistent with how ocaml associates file names with modules so it works out.


On Tue, Dec 15, 2009 at 6:28 PM, David Allsopp <dra-news@metastack.com> wrote:
Ashish Agarwal wrote:
> If you only have a file ast.mli, you should not be able to write Ast.Sig
because you do not have a module named Ast.

This isn't true - the include statement works at a type system level
(because you're dealing with a signature) and therefore only a .cmi file is
required. It does not generate any work for the linker so a module called
Ast is not actually needed when linking.

> Please double check your example. It cannot be working as you describe.

I'm afraid you should have checked - the example given compiles (try ocamlc
-o foo ast.mli toto.ml with suitable substitutions for the "..."s)

Romain Bardou wrote:

> I have a file ast.mli. It has no .ml implementation as it contains only
> type definitions.

This is fine

> I have a file toto.ml, which contains simply:
>
> module A = Ast

But as this involves a linker instruction you need an actual module - hence
the error you're seeing. I guess you could argue that the module statement
could check to see if Ast only contains type definitions and relax the need
for an actual module but in the general case Ast could contain values and so
you need a module to link against.

> I found a workaround, which is to change ast.mli to put all type
> definitions in a signature, such as:

This workaround works because you take the whole thing back to the type
system so module implementations are not required by the linker.


David