LLVM assert when defining same native function with different types

[brson]

mod a {
    type rust_task = uint;
    native mod rustrt {
        fn rust_task_is_unwinding(rt: *rust_task) -> bool;
    }
}

mod b {
    type rust_task = bool;
    native mod rustrt {
        fn rust_task_is_unwinding(rt: *rust_task) -> bool;
    }
}

fn main() { }

rustc: /home/banderson/Dev/rust2/src/llvm/include/llvm/Support/Casting.h:194: typename llvm::cast_retty<To, From>::ret_type llvm::cast(const Y&) [with X = llvm::Function, Y = llvm::Value*, typename llvm::cast_retty<To, From>::ret_type = llvm::Function*]: Assertion `isa<X>(Val) && "cast<Ty>() argument of incompatible type!"' failed.

[lht]

llvm Module::getOrInsertFunction:
http://llvm.org/docs/doxygen/html/Module_8cpp_source.html#l00139

If a function with the name is found but type is different, llvm bitcasts the found function type to the requested type and returns it. Later, when we set calling convention, the given function's type isn't consistent with the one (before conversion) in llvm symbol table.

After all, without mangling type information, we cannot have functions with the same name, right?

[catamorphism]

Still a bug as of 6/21/2012.

[mitsuhiko]

I have a very similar case:

mod llt {
    pub enum SDL_Window {}
}

#[link_name = "SDL2"]
extern mod ll {
    use llt::*;
    pub fn SDL_CreateWindow(title: *c_char, x: c_int, y: c_int,
                            w: c_int, h: c_int, flags: uint32_t) -> *SDL_Window;
}

This causes:

Assertion failed: (isa<X>(Val) && "cast<Ty>() argument of
  incompatible type!"), function cast, file
  ~/Development/rust/src/llvm/include/llvm/Support/Casting.h, line 197.

If I remove the -> *SDL_Window and replace it with an c_int it works.

[emberian]

Still a bug as of 6/22/2013 (good timing, past @catamorphism).

In test suite, xfailed, as issue-1866.rs

[brson]

nominating production ready

[graydon]

accepted for production-ready milestone

[nikomatsakis]

I believe that I fixed this bug as part of the PR for #3678 (which has not yet
landed). My fix was to cast the type of native fns where necessary.

[metajack]

This is indeed fixed. I submitted #8986 to remove the xfail comment on the test, but the test was already moved to run-pass some time ago.