regression: duplicate symbol definition

[Mark-Simulacrum]

https://crater-reports.s3.amazonaws.com/beta-1.86-3/beta-2025-02-28/gh/ns6251.spin-cookie-token-sample/log.txt

Possibly expected / not actually a problem, but probably good to figure out where this came from.

[INFO] [stdout]   = note: /usr/bin/ld: /opt/rustwide/target/debug/deps/libwit_bindgen_rust-7b7d20e188069fd3.rlib(wit_bindgen_rust-7b7d20e188069fd3.wit_bindgen_rust.712e9c6990f0d8a7-cgu.0.rcgu.o): in function `canonical_abi_realloc':
[INFO] [stdout]           /opt/rustwide/cargo-home/git/checkouts/wit-bindgen-e92a7091d4f51398/2f46ce4/crates/rust-wasm/src/lib.rs:134: multiple definition of `canonical_abi_realloc'; /opt/rustwide/target/debug/deps/libwit_bindgen_rust-742545eb826378e5.rlib(wit_bindgen_rust-742545eb826378e5.wit_bindgen_rust.ceb23070018b5742-cgu.0.rcgu.o):/opt/rustwide/cargo-home/git/checkouts/wit-bindgen-e92a7091d4f51398/e9c7c0a/crates/rust-wasm/src/lib.rs:134: first defined here
[INFO] [stdout]           /usr/bin/ld: /opt/rustwide/target/debug/deps/libwit_bindgen_rust-7b7d20e188069fd3.rlib(wit_bindgen_rust-7b7d20e188069fd3.wit_bindgen_rust.712e9c6990f0d8a7-cgu.0.rcgu.o): in function `canonical_abi_free':
[INFO] [stdout]           /opt/rustwide/cargo-home/git/checkouts/wit-bindgen-e92a7091d4f51398/2f46ce4/crates/rust-wasm/src/lib.rs:158: multiple definition of `canonical_abi_free'; /opt/rustwide/target/debug/deps/libwit_bindgen_rust-742545eb826378e5.rlib(wit_bindgen_rust-742545eb826378e5.wit_bindgen_rust.ceb23070018b5742-cgu.0.rcgu.o):/opt/rustwide/cargo-home/git/checkouts/wit-bindgen-e92a7091d4f51398/e9c7c0a/crates/rust-wasm/src/lib.rs:158: first defined here

[compiler-errors]

This may have been #137035 -- @Mark-Simulacrum, does the beta you're triaging include that PR or not? Not sure how the timing worked out.

[saethlin]

I got some weird results when trying to bisect this. Looks like the build still fails on nightly so I don't think that PR is it.

[Mark-Simulacrum]

Yeah, 137035 is included in beta (commit 42212a5 is what was tested here).

[apiraino]

Assigning priority (discussion on Zulip).

@rustbot label -I-prioritize +P-high

[saethlin]

The crate in question uses a proc macro, and with that expanded, I can minimize the crate to this:

use spin_sdk;

#[export_name = "handle-http-request"]
unsafe extern "C" fn __wit_bindgen_handle_http_request() {
    let layout = core::alloc::Layout::from_size_align_unchecked(1, 1);
    let result = std::alloc::alloc(layout);
    if result.is_null() {}

    wit_bindgen_rust::rt::as_i32(0);
    std::hint::black_box(RET_AREA);
}
static RET_AREA: u8 = 0;

with

[lib]
crate-type = [ "cdylib" ]

[dependencies]
spin-sdk = { git = "https://github.com/fermyon/spin", tag = "v0.3.0" }
wit-bindgen-rust = { git = "https://github.com/bytecodealliance/wit-bindgen", rev = "e9c7c0a3405845cecd3fe06f3c20ab413302fc73" }

Linkage minimizations are always trippy, so bear with me. We have a few layers to go.

[saethlin]

Ah-ha! I tried a bunch of codegen-units settings and it looks like the reproducer is highly sensitive to the number of codegen-units used in building dependencies. That's probably why this error comes and goes sporadically.

[saethlin]

Minimized a bit farther: (note that I turned spin-sdk into two of its transitive dependencies)

use spin_sdk;

#[export_name = "handle-http-request"]
pub unsafe extern "C" fn __wit_bindgen_handle_http_request() {
    let layout = core::alloc::Layout::from_size_align_unchecked(1, 1);
    let result = std::alloc::alloc(layout);
    if result.is_null() {}

    wit_bindgen_rust::rt::as_i32(0);
}

[lib]
crate-type = [ "cdylib" ]

[dependencies]
anyhow = "1" # required ???
wit-bindgen-rust = { git = "https://github.com/bytecodealliance/wit-bindgen", rev = "e9c7c0a3405845cecd3fe06f3c20ab413302fc73" }
spin-sdk = { git = "https://github.com/bytecodealliance/wit-bindgen", rev = "2f46ce4cc072107153da0cefe15bdc69aa5b84d0", package = "wit-bindgen-rust" }

[saethlin]

The regressed crate is depending on multiple versions of wit-bindgen-rust, which has an rt module, which exports the canonical_abi_realloc and canonical_abi_free symbols.

So the bigger question here is how does this crate ever build?

[saethlin]

I'm going to close this because the crate in question is simply buggy. In an ideal world we would be issuing a warning for this kind of situation; this is now the second time I've root-caused a crater link error regression to a crate which is mistakenly depending on two versions of a library that exports an unmangled symbol.

[apiraino]

wow thanks @saethlin for the deep dive 👍

Should this crate be excluded from our crater runs?

@rustbot label -P-high

[saethlin]

Yeah. Posted rust-lang/crater#766