link_section attribute seems to be broken

[ababo]

I compile the code below:

#[link_section = ".multiboot_header"]
pub static MULTIBOOT_HEADER: multiboot::Header = multiboot::Header {
    magic: multiboot::HEADER_MAGIC,
    flags: HEADER_FLAGS,
    checksum: (-((multiboot::HEADER_MAGIC + HEADER_FLAGS) as i32) as u32),
    header_addr: 0,
    load_addr: 0,
    load_end_addr: 0,
    bss_end_addr: 0,
    entry_addr: 0,
    mode_type: 0,
    width: 0,
    height: 0,
    depth: 0
};

using compiler version 1.3.0-nightly (922aef0 2015-07-27) with the following command line:

rustc --crate-type rlib --target x86_64-unknown-linux-gnu -C opt-level=3 -C no-stack-check -Z no-landing-pads --cfg arch_x86_64 --sysroot /dev/null ...

As result I get the following symbol inside a produced rlib:

0000000000000000 g O .rodata._ZN4arch4boot16MULTIBOOT_HEADER20h86b23e12a2a05e14jaaE 0000000000000030 arch:: boot ::MULTIBOOT_HEADER::h86b23e12a2a05e14jaa

[alexcrichton]

How'd you verify this? I tried a few variants myself and it looks like it works.

[ababo]

It's broken in Nightly release. Try the following code in Playground after choosing 'Asm' and 'Nightly'.

#![crate_type="rlib"]

pub struct Struct {
    pub field: u32,
}

#[link_section = ".my_section"]
pub static VAR: Struct = Struct { field: 1 };

[alexcrichton]

Aha, looks like I was using an older nightly.

cc @eefriedman, some informal bisecting turns up 8ebf952 as a suspicious commit

[eefriedman]

Yes, that bisection is correct; PR coming up.