refactor: reorganize_definitions: use the fallback/C ABI instead of Rust

[ahomescu]

Both sites now use Abi::FALLBACK (= Abi::C { unwind: false }).

Regression test: test_reorganize_implicit_extern (tests/snapshots/reorganize_implicit_extern.rs); on pre-fix code its implicit-externdeclaration fails to merge with theextern "C"duplicate, producing anextern "Rust"block and a spuriouscompute_1` rename.

reorganize_definitions.rs:843-847 and :1779-1782 both do:

let abi = m.abi.and_then(|abi| abi::lookup(&abi.symbol.as_str())).unwrap_or(Abi::Rust);

Rust's default ABI for extern {} blocks is "C" (rustc's own lowering uses
Abi::FALLBACK, which is Abi::C). Consequences:

• An implicit extern { fn foo(); } block never dedups against an
  extern "C" { fn foo(); } declaration — the ABI-mismatch continue at
  :2144 in find_foreign_item skips it.
• Worse, its items are re-emitted by into_items via
  mk().extern_(Abi::Rust) (:1990), silently changing the declared ABI of
  foreign functions.

Latent only because transpiled code usually writes extern "C" explicitly.
Fix: fall back to Abi::C { unwind: false } (or Abi::FALLBACK).

[ahomescu]

According to https://doc.rust-lang.org/reference/items/external-blocks.html#r-items.extern.abi.default, "C" is indeed the default.

[thedataking]

The transpiler always emits an explicit .extern_("C") (c2rust-transpile/src/translator/mod.rs:1231, :1459, :2009, :2236, :2781), so an implicit extern block can't come from transpiled code.

[ahomescu]

Right, which is why we probably didn't catch this. Do we want to be defensive here, or punt this PR?

[thedataking]

From offline discussion: we don't have a C pattern that is not UB and would trigger this pattern. We assume that the transpiler operates on valid C and that the refactorer operates on whatever Rust the transpiler outputs.

[ahomescu]

From offline discussion: we don't have a C pattern that is not UB and would trigger this pattern. We assume that the transpiler operates on valid C and that the refactorer operates on whatever Rust the transpiler outputs.

Did you mean to reply to this PR or #1935 ? Edit: if you meant 1935, could you approve 1924 so we can land it, since it's a legitimate bug fix?

[thedataking]

From offline discussion: we don't have a C pattern that is not UB and would trigger this pattern. We assume that the transpiler operates on valid C and that the refactorer operates on whatever Rust the transpiler outputs.

Did you mean to reply to this PR or #1935 ? Edit: if you meant 1935, could you approve 1924 so we can land it, since it's a legitimate bug fix?

My bad. Reviewing this one.

[thedataking]

From offline discussion:

• we shouldn't add tests here since it gives the impression that this bug can be triggered
• would be better to expect(...) here so this fails loudly.
• alternatively, just fix the Abi to FALLBACK, don't add comments or tests.