Native broker 2C/C3: an erased empty-map type argument defers to the target's inference (std_list, −3 broker errors)

[gunbai-bot]

Native broker wave 2C, class C3: the std_list UNRESOLVED_CompilerError and the generic T erased to ().

What was wrong

In std.list duplicated_by_key, DuplicatedScan { firsts: empty_map(), .. } is a field typed Map<String, T>. It emitted v1_rt::rc_empty_map::<String, compile_error!("UNRESOLVED_CompilerError")>(). The record was then inferred as DuplicatedScan<()>, which produced E0282 and E0308 (Vector<()> vs Vector<T>).

Where the boundary is (DESIGN §6b)

The emitted functions do carry <T>, so this is not lane 1C's "generic fns without ". The turbofish's type also arrives already erased: v1.compiler.infer does not bind a generic record's parameter at an unannotated record literal. No renderer can name that type. A first attempt rendered it in the fn's generic scope, measured no change, and was reverted.

The infer gap is open and tracked, not absorbed. gentle-dove-36 is routing it to lane 2D (loyal-gull-749), whose in-flight fix "a variant literal of a generic owner takes the expected application" (infer_record_lit_structural) looks like the same root. When that lands, the erased argument should stop arriving, and this arm stops firing for it.

The emitter rule

rust_empty_map_init_expr already defers ::<_, _>() to the target's inference when it cannot name a turbofish's formals in the fn's scope. An erased type argument is the same state seen one stage earlier, so it now takes the same arm (type_node_carries_erased_type_argument).

This follows the rule in rust_witness_carrier_from_type_node's note: an erased first source declines so the next authority is consulted, rather than rendering its own refusal. rustc infers T from the field's later map_insert, or refuses loudly if it cannot. Nothing is widened.

Control

Shared label gunbc test //gunbc/instruments:native-emission-controls, cases generic_empty_map_turbofish_cases. They call the real std.list duplicated_by_key:

• ["a","b","a","c","b","a"] must give ["a","b"] (each repeated key's first element, once, in repeat order);
• a list with no repeats must give [].

One remote dispatch ran the label, regenerated, installed and rebuilt the mirrors, then ran the label again:

• Before: E0282/E0308, exit 2.
• After: label exit=0, both cases held.

Broker probe delta

cargo check errors went from 144 on main to 141 (−3). All std_list.rs errors are gone.

The remaining durable_cas_file_store errors are F's and are fixed by #13213. #13202, #13213 and this PR all regenerate v1_compiler_emit_rust.rs: whichever merges later regenerates it rather than text-merging.

🤖 Generated with Claude Code

[briansrls]

LAND on substance. This is an honest interim rather than a substitute for 2L: the emitter does not invent or recover T; it declines an erased empty-map turbofish to the same Rust-inference arm already used when the enclosing function’s formals cannot be named. The generated assignment and later map_insert still constrain the type, and rustc refuses if that context is insufficient. The upstream generic-record-instantiation gap remains named in declared_type_conformance_note; once 2L supplies the application, this compiler-error predicate naturally stops firing for the specimen. The real std.list.duplicated_by_key cases establish the generic body and its behavior, not a fixture-only shape. Non-blocking: update the PR body’s lane owner from 2D to 2L. The exact-head witnesses run was still in progress when I submitted this review, so do not queue it until that run succeeds and the #13164 window closes.

[gunbai-bot]

The failing check is rust-unit-tests, and it timed out: "The action 'Run the v1 unit tests' has timed out after 40 minutes". Every test that ran before the cutoff passed. That is a runtime budget on the lane, not a test this PR breaks. C3's change is one emitter predicate plus its regenerated mirror. Main's own latest witnesses run (a2bf3b9) also failed. I've asked gentle-dove-36 whether rust-unit-tests is now a required lane; if it is, its budget is the issue, not this diff. The other lanes on this head are still running.
— sent from still-raven-321