Retire cssl parse_closure_mods onto the emitted_population manifest: the last live Rust text-parse on the seed-linked path recovers CLOSURE MEMBERSHIP by scraping the emitted lib.rs pub-mod lines, and the manifest in the same out_dir already declares that population with no Rust grammar involved

[briansrls]

The last live Rust text-parse on the seed-linked path recovered CLOSURE MEMBERSHIP by scraping pub mod <name>; lines out of the emitted lib.rs. The emitter already declares that population in the same out_dir: v1.compiler.emit_rust emit_emitted_population_manifest renders one line per produced path into src/emitted_population.rs, under a grammar both ends resolve from one authority (gunbc.stage0_emitted_population_manifest). Reading it is a split and a strip — no Rust grammar on either side, which is the property that made the artifact a file of // lines in the first place.

The swap is a strengthening, not a rename

lib.rs's mod list is rendered FROM the same module file list the writer then writes out (emit_lib_rs_from_files(all_mod_files) → write_output_files), so the retired check compared one emit artifact against another and could only ever agree with itself on the real path. Its only reachable red came from a hand-authored lib.rs.

The manifest declares Cargo.toml, lib.rs, main.rs, every module, every test file and itself, so assembly now asks the question that matters: is every path the emitter DECLARED it produced materialized in out_dir. Wider coverage obtained by deleting a parser, not by adding a check.

What goes with it

BOOTSTRAP_INLINE_MODS (NonEmptyVec, NonEmptyBTreeSet) is deleted. It existed because a hand-authored shim lib.rs can carry pub mod lines for wrappers the emitter inlines as bare structs. The manifest never declares them — they are not files — so the state that skip tolerated is not representable in the population assembly now reads (DESIGN §4b top rung, not the one below).

Typed refusals, each with a discriminating RED enrolled

• RefusedDeclaredMember — a declared path is not materialized. The fixture deliberately keeps the missing member OUT of lib.rs, so it went green under the retired parse: the refusal is evidence the manifest is the authority now, not merely that some refusal survived the change.
• MissingPopulationManifest — no fall back to reading lib.rs. The fixture is an out_dir with a perfectly good lib.rs and no manifest, exactly the state the retired parse accepted.
• EmptyPopulationManifest — a file at the manifest path carrying no line under the grammar is ignorance, not an empty population; answering "nothing declared, so nothing missing" would be the absorbing fallback read in the narrowing direction (DESIGN §5).

Fixtures build the manifest through the same grammar authority (write_population_manifest) rather than hand-spelling // and \n, so a change to the grammar moves both ends at once.

Incidental, and a real hazard

temp_fixture_root keyed the fixture root on the pid alone while cargo ran the cases concurrently, so every case's remove_dir_all raced every sibling's writes at the same <root>/out. It passed by timing rather than by construction; keyed on the case name now.

Authority note tools.self_host_curated_seed_linked_harness cssl_closure_assembly_note updated to state the manifest rule (it read "dep manifest = gunbc-emitted src/lib.rs pub mod list").

Measurement

cargo test -p v1-compiler --bin cssl_assemble (remote): 10 passed, 1 failed. The failure is the pre-existing environment one — normalize_stale_narrow_lib_without_namespace_graft_refuses_cargo panics with release bins or shim dir missing (gunbc=false, assemble=false, shims=true) because the remote runner has no target/release bins. Same outcome before and after this change.