Statics should not be able to take the address of a thread_local static

[alexcrichton]

Not sure if this is a codegen bug or just something that's not supported by LLVM

#![feature(thread_local)]

#[thread_local]
static FOO: uint = 0;
static BAR: &'static uint = &FOO;

fn main() {
    println!("{}", *BAR);
}

$ rustc foo.rs && ./foo
282584257676671
$ rustc foo.rs --opt-level=3 && ./foo
0

[steveklabnik]

I'm now getting zero even with -O, so this is fixed, I think.

[alexcrichton]

Ah I believe this is still an issue, the results here should be deterministic. The right answer is happening in the optimized case I believe due to LLVM being able to inline everything.

[steveklabnik]

Oh, I was getting zero in all cases no matter what. how do you trigger the bug?

[alexcrichton]

Interesting, if I compile without optimizations then I get the bogus answer above. That being said, this is semantically just not a correct operation as there is no statically known global value for the address of a thread local (it varies per thread!)

[steveklabnik]

Triage: updated code:

#![feature(thread_local)]

#[thread_local]
static FOO: usize = 0;
static BAR: &'static usize = &FOO;

fn main() {
    println!("{}", *BAR);
}

I still get zero in both cases here.

[Mark-Simulacrum]

This bug is still present. Updated title to correspond to the fact that statics can reference thread locals, but they shouldn't be able to.

[alexcrichton]

This has now been fixed on nightly