Lane A Family-A: map-literal String-key coercion (type-derived) — clears HashMap<&str> E0308 family

[briansrls]

Family-A: map-literal String-key coercion (type-derived)

The single biggest E0308 lever in the Route-A re-baseline. A data NAME: Map<String,V> = { k: v, ... } declaration emits as a thread_local! { static CACHED = { let mut __m = HashMap::new(); __m.insert(<key>, <val>); ... } }. The key was emitted via the host→dag text seam without .to_string(), so the literal inferred HashMap<&str, V> and mismatched the declared HashMap<String, V>.

Fix (single authority: src/v1/05_emit_rust.dag)

• New emit_rust_map_literal_key(field_name, key_is_string, corpus_repr) — appends .to_string() to the key literal only when the declared key type is String-like, then routes through the existing host→dag seam.
• New map_literal_key_is_string(map_type_node, source_indices) — reads the map's declared key type (first type arg, via children |> first with the __applied_type_args fallback, mirroring the proven rust_host_text_carrier_elem_name idiom) and tests is_rust_string_like. The coercion is DERIVED from the declared key type, not a blanket keys → to_string — a non-String key is left bare.
• emit_data_def_body computes key_is_string once per map literal and threads it to the key emitter.

The committed Route-A seed (src/v1/stage0/src/v1_compiler_emit_rust.rs) is the hand-synced mirror; the corrected map_literal_key_is_string body is spliced in to match.

Proof by execution

• (a) String key coerces + clears the family. dag_keywords (HashMap<String,String>) now emits __m.insert("true".to_string(), …). Fresh regen_stage0 --emit-fresh build: the HashMap<&str, V> family went 24 → 0.
• (b) Non-String key stays bare (not blanket). Witness non_string_key_is_left_as_literal (key_is_string=false → key emitted without .to_string()). No live corpus map literal has a non-String key (0 non-string-key inserts emitted), and no new "expected non-String key" errors appeared — confirming the coercion is per-slot type-derived, not blanket.
• 4 witnesses green (map_literal_string_key_test.rs): host-repr coerce, faithful-repr seam routing, non-String control, and an end-to-end Map<String,Int> source fixture (String key coerced, Int value left bare).
• Committed seed builds green.

Out of scope (documented, not regressions)

The wider --emit-fresh wiring unmasks a separate pre-existing emitter family: |> first emits as .first().cloned() (Option) where consumers expect the unwrapped element (633 expected String/Rc, found Option). Confirmed pre-existing — the committed seed carries the identical a.clone().first().cloned() bug in extdeps_version_semver.rs:81 but leaves that module unwired in lib.rs. Family B (Box/Rc auto-wrap, ~7) and E0560 measure-tower also remain. These are tracked for follow-on lanes, not this PR.

— sent from deep-wolf-695

[gunbai-bot]

Superseded by #5716. #5716 (base=main) is a strict superset of this branch (head e7d837f is an ancestor; the Family-A map-key commits and both Family-A witnesses are byte-identical inside it) and additionally carries the first-Option renderer fix + residual-3 fail-closed unwraps. #5716 is read-ready and CI-green, so the whole Route-A E0308-clearing delta lands to main via #5716. Closing this to keep a single clean merge path. — sent from deep-wolf-695