Static variables are not exported on multiple platforms.

[Jake-Shadle]

It appears that #88092 has regressed in nightly, The last working nightly we used was rustc 1.64.0-nightly (d68e7ebc3 2022-07-20), rustc 1.64.0-nightly (62b272d25 2022-07-21) was the first nightly that stopped working correctly.

I tried this code:

// lib.rs
#[no_mangle]
#[used]
pub static FOO: u64 = 4242;

I expected to see this happen:

cargo build --release --target wasm32-unknown-unknown && wasm2wat target/wasm32-unknown-unknown/release/missing_symbols.wasm

In rustc 1.62.0 (a8314ef7d 2022-06-27), this properly exports the FOO static.

(module
  (table (;0;) 1 1 funcref)
  (memory (;0;) 17)
  (global $__stack_pointer (mut i32) (i32.const 1048576))
  (global (;1;) i32 (i32.const 1048576))
  (global (;2;) i32 (i32.const 1048584))
  (global (;3;) i32 (i32.const 1048592))
  (export "memory" (memory 0))
  (export "FOO" (global 1))
  (export "__data_end" (global 2))
  (export "__heap_base" (global 3))
  (data $.rodata (i32.const 1048576) "\92\10\00\00\00\00\00\00"))

Instead, this happened:

cargo +nightly build --release --target wasm32-unknown-unknown && wasm2wat target/wasm32-unknown-unknown/release/missing_symbols.wasm

In rustc 1.64.0-nightly (62b272d25 2022-07-21), FOO is not exported.

(module
  (table (;0;) 1 1 funcref)
  (memory (;0;) 16)
  (global $__stack_pointer (mut i32) (i32.const 1048576))
  (global (;1;) i32 (i32.const 1048576))
  (global (;2;) i32 (i32.const 1048576))
  (export "memory" (memory 0))
  (export "__data_end" (global 1))
  (export "__heap_base" (global 2)))

Meta

rustc --version --verbose:

rustc 1.64.0-nightly (62b272d25 2022-07-21)
binary: rustc
commit-hash: 62b272d25c5bb8b6bb8ac73797d82b8b9a1eabda
commit-date: 2022-07-21
host: x86_64-unknown-linux-gnu
release: 1.64.0-nightly
LLVM version: 14.0.6

[bjorn3]

Regression range: d68e7eb...62b272d I expect this to be caused by #93718 cc @thomcc

[robbert-vdh]

This also seems to affect at least x86_64-unknown-linux-gnu, x86_64-pc-windows-gnu, x86_64-apple-darwin, and aarch64-apple-darwin. Since about three days ago #[no_mangle]+#[used] static symbols no longer appear in the data section (/other platform's equivalents). With an explicit #[used(linker)] it works, but that feature is of course still unstable.

[thomcc]

Hm, interesting. That doesn't make much sense at I suspect this is an issue with how we compile #[used(linker)]? That patch just selects between #[used(linker)] and #[used(compiler)] based on some settings. (No, the original code was badly broken) CC @nikic, who might have more of an idea about why it may be broken.

I don't have time to fix this right now, and it's very non-obvious to me why this would happen, so if it's as broken as @robbert-vdh suggests, the PR should probably be reverted.

(Removing O-wasm based on the previous comment)

[thomcc]

The first PR should definitely fix this. The second fixes the bug. I don't really have the ability to dig into this at the moment, but this should unblock either option.

That said, some evidence in favor of reverting is that this may have been fixed between when I wrote that PR and when it landed (which was quite a long time), perhaps by #95604 ? Not sure.

[thomcc]

@Jake-Shadle @robbert-vdh Can you confirm that this is fixed in rustc 1.64.0-nightly (6dbae3ad1 2022-07-25)? (Or later)

[Jake-Shadle]

Confirmed, nightly build went green again, I'll mark this issue as fixed.