Add support for DllExport on Windows

[kvanbere]

It currently does not seem to be possible to export a function from a DLL on Windows. It's not done as one would expect when building with --lib.

[kvanbere]

It seems that rustc-hash.dll indeed contains exports on Windows, and therefore it must be possible to somehow build such that code is exported. The following method does not produce exports;

Crate:

// Crate linkage metadata
#[link(name = "thing", vers = "1.0", author = "who")];

// Make a library ("bin" is the default)
#[crate_type = "lib"];

// Load some modules from other files
mod test;

Module:

pub extern fn foo(a: int) -> int {
    1
}

Thoughts?

[kvanbere]

mod test; -> pub mod test; works as intended. Sorry folks, closing.

[Aatch]

Re-opening because the general idea is still sound. Adding a #[linkage] attribute for more exotic linkage types seems valid to me.

Also, I'm nominating this for production ready, based on the idea that dllimport/dllexport exist for a reason. Though I'm happy to be corrected here.

[emberian]

Visiting for triage; nothing to add

[catamorphism]

Accepted for feature-complete

[pnkfelix]

Assigning P-low. Not 1.0.

[klutzy]

We actually dllexports all public symbols:

// lib.rs
#![crate_type = "dylib"]
#![crate_id = "r"]
#[no_mangle] pub fn func1() {} // exported
#[no_mangle] fn func2() {}
pub fn func3() {} // exported
fn func4() {}

C:\home\stone\tmp>dumpbin /exports r-1437d9a7-0.0.dll
...
File Type: DLL

  Section contains the following exports for r-1437d9a7-0.0.dll

    00000000 characteristics
    538549CB time date stamp Wed May 28 11:28:27 2014
        0.00 version
           1 ordinal base
           4 number of functions
           4 number of names

    ordinal hint RVA      name

          1    0 000015F0 _ZN5func320h830e15bc5d49336cjaa4v0.0E
          2    1 0000162C __morestack
          3    2 000015B0 func1
          4    3 00004000 rust_metadata_r_0.0_ac1193c82740e58b

but just by accident! If we don't explicitly export anything, ld exports everything public.

The "unintended export" will not occur if we add some explicit dllexports in native code:

// c.c
__declspec(dllexport)
void func5(void) {}

$ gcc -c -o c.o c.c
$ ar rcs ext.lib c.o

// lib.rs
#![crate_type = "dylib"]
#![crate_id = "r"]

#[link(name = "ext", kind = "static")]
extern "C" {
    pub fn func5();
}

#[no_mangle] pub fn func1() {} // exported
#[no_mangle] fn func2() {}
fn func3() { unsafe { func5(); } } // use dllexported symbol

$ rustc lib.rs

C:\home\stone\tmp>dumpbin /exports r-1437d9a7-0.0.dll
...
  Section contains the following exports for r-1437d9a7-0.0.dll

    00000000 characteristics
    538552D1 time date stamp Wed May 28 12:06:57 2014
        0.00 version
           1 ordinal base
           1 number of functions
           1 number of names

    ordinal hint RVA      name

          1    0 0000161C func5

// a.rs
extern crate r;

fn main() {
    r::func1();
}

$ rustc lib.rs
$ rustc a.rs -L .
error: linking with `gcc` failed: exit code: 1
...
note: a.o:(.text+0x1a): undefined reference to `func1'

This is bad. We have to export them explicitly in this case.

[steveklabnik]

I'm pulling a massive triage effort to get us ready for 1.0. As part of this, I'm moving stuff that's wishlist-like to the RFCs repo, as that's where major new things should get discussed/prioritized.

This issue has been moved to the RFCs repo: rust-lang/rfcs#826

[klutzy]

(Note that the missing export behavior is certainly a bug, which can be fixed without introducing dllexport syntax support. I'll open new issue for the one.)

[alexcrichton]

Note that for MSVC targets the compiler now places dllexport on all exported functions from a library. It does not, however deal with dllimport at all, and in general the situation is unfortunately a bit of a mess.

I've written up a comment about this, but I'm going to use this bug to track a more proper implementation of using dllimport and dllexport, many of the concerns of which are outlined in that bug. @klutzy also probably knows enough about this to be a good liason to talk to!

I think this is more concretely actionable in this repo than the RFC repo for now, so I'm going to reopen this.

[retep998]

I think we shouldn't indiscriminately put dllexport on all exported functions from a library, because if we statically link the library, then those functions still have dllexport and are thus exported from the resulting executable, even though the main crate doesn't publicly re-export them. I think this might be why link is failing to strip unreferenced functions.