Path-D v2-side regen_stage0 bin — mirror v3's regen_bootstrap pattern (sunny-otter-371/#3407 recommendation)

[briansrls]

Expected-Red Callout

This PR's cargo run -p v2-compiler --bin regen_stage0 -- --verify gate is EXPECTED-RED on first run. It documents pre-existing v2 stage0 seed drift that the substrate-fill follow-up will close. Do not make the gate permissive with || true; the honest red result is the mechanism surfacing the drift.

Summary

Adds a v2 regen_stage0 bin mirroring v3's regen_bootstrap pattern: --verify compares committed generated stage0 Rust against a fresh self-compile, while no flag writes the registered generated outputs. The PR also wires CI/scripts/BOOTSTRAP.md to the new mechanism and replaces the broken ignored-test gate path.

The regen driver fail-closes on output-set drift: fresh emitted top-level .rs files must be registered in GENERATED_STAGE0_FILES, registered files must be emitted, and committed generated stage0 files cannot sit outside the registry.

Follow-Up: v2 Stage0 Seed Drift Exposed by regen_stage0

The new gate exposes current v2 seed drift. A fresh self-compile differs in these 16 generated files:

• extdeps_languages_rust_emit.rs: runtime function registry adds Set intrinsics mapping empty_set -> rc_empty_set, set_insert -> rc_set_insert, set_union -> rc_set_union, set_contains -> set_contains.
• std_effects.rs: by-ref signature drift for derive_op_effect method/path plus match formatting; no observed set unresolved errors.
• std_graph.rs: visited sets move from Rc<HashMap<String,bool>>/local set_has to Rc<BTreeSet<String>> plus empty_set/set_insert/set_contains calls; fresh build has missing set helpers.
• std_http_path.rs: optional split handling changes from empty-string fallback to MalformedPathSegment returns; no observed set unresolved errors.
• v2_compiler_complexity.rs: SCC member/assigned sets move from HashMap bool-sets to BTreeSet plus empty_set/set_insert/set_contains; includes UNRESOLVED_TypeVariable in fold accumulator.
• v2_compiler_effect_derivation.rs: import/signature adaptation for derive_op_effect by-ref method/path; no observed set unresolved errors.
• v2_compiler_emit.rs: render_node_type shared_types moves from HashMap bool-set to BTreeSet and set_contains/empty_set.
• v2_compiler_emit_rust.rs: broad Rust emit surface moves recursive/shared/ownership/read-only sets from HashMap bool-sets to BTreeSet; many empty_set/set_contains/set_insert/set_union helper calls plus UNRESOLVED_TypeVariable accumulators.
• v2_compiler_infer.rs: EmitGraphInfo set fields move from HashMap bool-sets to BTreeSet and empty_emit_graph_info uses empty_set; set_contains replaces map_get bool probes.
• v2_compiler_infer_cycle.rs: cycle/name set logic moves from HashMap bool-sets to BTreeSet helpers; fresh build has missing empty_set/set_contains/set_insert and UNRESOLVED_TypeVariable.
• v2_compiler_infer_emit_info.rs: EmitGraphInfo set fields/constructors move to empty_set/set_contains; fresh build has missing empty_set/set_contains.
• v2_compiler_infer_lookup.rs: adds set_element_type_in_env and imports node_is_set_collection; no observed direct helper unresolved errors.
• v2_compiler_infer_method.rs: set_union builtin registry clone fix; builtin names already seeded here.
• v2_compiler_infer_types.rs: adds node_is_set_collection and bare_set_node; no observed direct helper unresolved errors.
• v2_compiler_ownership.rs: ownership/read-only/callable-ref sets move from HashMap bool-sets to BTreeSet helpers; fresh build has missing empty_set/set_insert/set_contains/set_union and UNRESOLVED_TypeVariable.
• v2_compiler_runtime_rust.rs: runtime shim text adds BTreeSet import plus rc_empty_set/rc_set_insert/rc_set_union/set_contains implementations.

Unresolved generated function names observed when checking fresh emission:

File	empty_set	set_contains	set_insert	set_union
std_graph.rs	2	3	2	0
v2_compiler_complexity.rs	4	2	2	0
v2_compiler_emit.rs	1	1	0	0
v2_compiler_emit_rust.rs	16	23	4	1
v2_compiler_infer.rs	9	3	5	0
v2_compiler_infer_cycle.rs	2	2	3	0
v2_compiler_infer_emit_info.rs	6	1	0	0
v2_compiler_ownership.rs	11	1	3	9

compile_error!("UNRESOLVED_TypeVariable") appears in fresh emitted BTreeSet fold accumulator positions in five files:

• v2_compiler_complexity.rs
• v2_compiler_emit_rust.rs
• v2_compiler_infer.rs
• v2_compiler_infer_cycle.rs
• v2_compiler_ownership.rs

Type-mismatch failures are E0308 HashMap-vs-BTreeSet shape conflicts: examples include expected &HashMap<String, _> found &Rc<BTreeSet<String>>, expected Rc<HashMap<String, bool>> found Rc<BTreeSet<String>>, and expected Rc<BTreeSet<String>> found Rc<HashMap<String, bool>>.

Hypothesis: the substrate intent for Set/BTreeSet appears present upstream of the seed (runtime shim text, builtin function registry, and container template/type probes all exist in .dag or generated fresh output), but the committed v2 stage0 seed is still on mixed HashMap-as-bool-set conventions. The failing fresh output is therefore not just a missing import; v2's current emitter/type inference mapping is partly applying the newer Set/BTreeSet model while still leaving call sites/import wiring and accumulator element inference unresolved. That puts the gap at the emitter/type-inference/seed boundary rather than solely in one runtime carrier.

Test Plan

• cargo build -p v2-compiler --release — passes on the restored committed stage0 seed.
• cargo check -p v2-compiler --bin regen_stage0 — passes.
• cargo test -p v2-compiler --bin regen_stage0 — passes.
• cargo fmt --all --check — passes.
• cargo run -p v2-compiler --bin regen_stage0 -- --verify — expected-red: fails with the 16-file stale report above.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 9111ecc7 · Trigger: schedule
• Thinking: 296s wall

---

BLOCKING (1)

Root Cause

• src/v2/stage0/src/bin/regen_stage0.rs The generated-output registry is trusted as complete without reconciling it against compile_sources output → compare the fresh emitted src/*.rs file set to GENERATED_STAGE0_FILES and fail on any unregistered fresh file or stale committed generated file before write/verify.

⚠️ The regen driver needs an output-set completeness check before it can replace the stage0 gate.

[briansrls]

BLOCKING: --verify only compares paths already named in GENERATED_STAGE0_FILES, so a fresh self-compile can emit a new unregistered stage0 file and CI will still pass without committing it, violating INVARIANTS P2/P3 freshness and fail-closed boundary discipline.

[briansrls]

Reviewed dashboard-only approval artifact /api/reviews/15641/artifacts/stdout.log.

No blocking findings to fix. I verified the optional observations against the current diff:

• The inline generated.method_template_projection fixture is intentional for this mechanism PR; the underlying producer/drift disposition remains the operator-held follow-up captured in the drift artifact rather than additional scope here.
• Replacing the old ignored bootstrap_fixed_point CI path with regen_stage0 --verify is the intended Path-D gate change; the dropped idempotence coverage is an operational note, not a requested fix.
• WIP commit history is covered by the repo squash-merge policy.

No code change for this approval artifact. — sent from stern-deer-537