Re: [Caml-list] strange error with packed modules and module type of

[Jacques Garrigue <garrigue@math.nagoya-u.ac.jp>]

On 2012/06/08, at 4:53, Milan Stanojević wrote:

> If I just do
> ocamlopt.opt -for-pack G -c foo.mli foo.ml
> ocamlopt.opt -pack -o g.cmx foo.cmx
> 
> that is, forget all the fancyness with std and specifying interface
> for G, the resulting G module has the type
> sig module Foo : sig type t = G.Foo.t val x : t end end
> 
> So somehow the resulting interface of G is defined in terms of itself?!
> 
> Should I just file this problem on ocaml mantis?

This is the right behavior.
To understand it, you should distinguish between the interface (or abstract signature)
and the concrete type of a module (its strengthened signature).
The above type is not the interface of G (which of course should not refer to
itself by name) but its strengthened signature (its exact type).
Since an abstract signature has a meaning independently of the module you use it
for, it is strengthened so that the abstract components of the interface
refer explicitly to the module.

See this dialogue:

# module M : sig type t val x : int end = struct type t = int let x = 1 end;;
module M : sig type t val x : int end
# module N = M;;
module N : sig type t = M.t val x : int end

> p.s the only way I was able to get interface for G is to make a helper
> program that uses G with a wrong signature and get the compiler to
> give me the type in the error message. Is there a better way?

As shown above, you can define another module as an alias.
It will give you the strengthened signature for your original module.

You can also see its abstract signature, using "module type of".

# module type S = module type of M;;
module type S = sig type t val x : int end

Jacques Garrigue