Native function link name should be an attribute

[brson]

Instead of

    fn close(fd: int) -> int = "_close";

we should write

    #[link_name = "_close"]
    fn close(fd: int) -> int;

[lht]

Should native mod be moved to attribute also to keep it consistent with native fn?

So

native "cdecl" mod ssl = "crypto" {...}

becomes

#[link_name = "crypto"]
native "cdecl" mod ssl {...}

[lht]

Okay, I guess the answer is yes. Issue #547 (Replace some native module syntax with attributes).

[brson]

Also native mods that are crate directives.

[lht]

Now I'm about to start hacking crate directive. The plan is:

1. change
   mod os_fs = "posix_fs.rs";
   to
   #[path = "posix_fs.rs"] mod os_fs;
path is chosen over filename, considering it might indicate a directory.
2. Remove option::t from ast::cdir_src_mod and ast::cdir_dir_mod.
3. Add native support, so we can parse

#[link_name="real_foo"]
native mod foo;```

I'm not totally sure if moving filename/path to attribute is the plan, please confirm?

[brson]

Yes. What you propose sounds good.

[lht]

Are native item declarations in .rc files allowed? As hinted by the few native rc test cases.

native.rc:
native "cdecl" mod rustrt {
  fn str_buf(str s) -> int;
}

native mod libc = target_libc {
  fn puts(int s) -> ();
}

but the reference states a crate is a sequence of directives.

[brson]

Seems to be a feature that was never implemented, and I don't think there's any need.

[lht]

I guess this can be closed? Or should we remove the tests:

./run-pass/native-mod.rc
./run-pass/native.rc

[brson]

Let's leave the tests as a reminder that somebody wanted that feature at some point.