map_get fork: the declared projection renames to map_get_checked

[gunbai-bot]

Defect (RFM one_spelling_names_a_primitive_and_a_declaration_with_different_contracts)

The spelling map_get named two contracts:

• the interpreter PRIMITIVE map_get (Optional — literally outcome_accepted(map_lookup(...)), interpreter:14297 registry), and
• the v2.std.collection DECLARED projection map_get (Outcome<Optional>), emitted as its own body.

The authoritative meaning is the primitive's: it is the runtime the floor roster already treats as the bare-provider path, and the declared body was the divergent second meaning. The declared projection is renamed to map_get_checked at its owner (src/v2/std/collection.dag) and every consumer follows.

What changed

• Declaration renamed at src/v2/std/collection.dag; 27+ src/v2 consumer files (imports, qualified calls, bare calls under declared imports).
• dag/std/primitive_projection.dag: the projection row's decl_name becomes map_get_checked; the primitive slug primitive.map_get keeps its spelling (the primitive is not renamed).
• Probe strings updated where they exercise the DECLARED route (resolved_call_emission_identity, callable_candidate_ambiguity, self_host_emitted_call_target_realization); primitive-route probe strings with Present/Absent arms are intentionally untouched.
• src/v1/stage0/src/declaration_index.rs: synthesized test fixture updated; the v1 mechanism never keyed on the declared spelling (method_call.map_get is the kernel method).
• RFM row kept, REPAIR RECEIPT added.
• std_primitive_projection.rs regenerated (required-regen: self-compile clean, adjudicated 163, declared_divergent=1 — the projection row follows the rename).
• Merges origin/main@39d1e2e6db (carries Delete the v1 Rust tests that take over 100ms #13452/Delete ct_self_parse_all_test at its authority (main off regen fixed point after #13452) #13456, restoring the regen fixed point).

Reds and positive control

• RED w_projection_census_declares_no_divergent_shared_spelling: folds primitive_projection_roster(); failed on main (red-confirmed 2026-10-05 19:11Z) while any row's decl_name equals its primitive's runtime name at DivergentProjection fidelity; passes after the rename.
• POSITIVE CONTROL w_projection_census_keeps_the_outcome_lookup_projection_under_its_own_name: the outcome-lookup projection keeps its own name (a rename that fixed the fork by mis-renaming a distinct projection would fail it).
• Emission identity: w_divergent_map_get_checked_keeps_declared_outcome_identity pins the declared route's emitted call target crate::v2_std_collection::map_get_checked().

Witness evidence (remote claim_batch, 2026-10-06)

• primitive_projection_authority entry: 24 witnesses PASS (4 listed FAILs are arg-taking helper fns invoked by claim_batch without args — composition building blocks, not claims).
• resolved_call_emission_identity: 7/7 PASS.
• callable_candidate_ambiguity: 5 witnesses PASS (3 helpers are arg-taking).
• value_null_split (primitive route untouched): 7/7 PASS.
• Full floor run on this head: zero non-wet failures; the only refusals are 19 wet-lane host-premise entries (the remote runner pool lacks the browser toolchain — remedy the RUNNER, not the subject; CI classifies infra/environment as not a verdict about the diff).

Consumer census

Found by identity across src/v2, dag, src/v1: all declared-form call sites renamed (incl. the compiler passes 02_parse/03_ingest/03_name_resolve/03_resolve/emit_, lens/, std/, workflow/); floor_unimported_bare_provider_debt_roster rows stay (primitive-route sites); calm-koi-257's #13376 floor_route_gap.dag calls only map_contains_key/empty_map/map_insert (unrenamed) — coordinated, no interaction.

[briansrls]

Approved at a1d0d22.

I audited the rename at contract grain rather than treating it as a textual sweep.

1. Every executable call renamed to map_get_checked was already a consumer of the declared Outcome<Optional<V>> projection. The sites either match Accepted/Rejected directly or feed the value through an Outcome combinator before matching the inner Present/Absent. None of the renamed calls used the primitive's direct Optional<V> contract. The declaration body itself is unchanged apart from the name: it still projects total map_lookup through outcome_accepted.

2. Primitive consumers were correctly left on map_get. Representative production and witness sites use the direct Optional shape (Present/Absent, or return the bare lookup as Optional) and remain untouched. The original transitive incident source is deliberately unchanged: after the declared projection vacates the spelling, its bare map_get resolves without ambiguity to the primitive route whose contract its arms already expect.

3. Probe strings moved only when they assert the declared call target. Their expected target is now crate::v2_std_collection::map_get_checked, while they explicitly reject v1_rt::map_get; primitive-route specimens retain map_get.

4. The hand-written v1 change is fixture-only. declaration_index.rs changes only inside #[cfg(test)] mod import_binding_authority_tests: the synthetic declared candidate is renamed, and the permanent regression arm requires bare map_get never to become ambiguous again. No production declaration-index mechanism changes. The only other v1 file is the generated std_primitive_projection.rs mirror, whose sole change is the declared projection name; the primitive identity and divergent-contract record remain intact.

5. The repair controls are discriminating: the projection roster refuses any divergent projection that reuses its primitive's runtime name, a positive control retains the primitive.map_get -> v2.std.collection::map_get_checked projection fact, and the original incident source now requires zero ambiguous references.

All five exact-head jobs passed. No blockers.

[gunbai-bot]

Both findings addressed in 61849f0:

1. Markers — rewritten. The block now reads std.optional and names map_get_checked; the asserted emitted target crate::v2_std_collection::map_get_checked( is unchanged. The committed-marker class (a_verification_reads_its_own_label_not_its_output) was caused by my stash dance during the Optional de-fork step 2: re-home v2.std.optional to std.optional under the dag root #13388 conflict fix; the corrective commit 0125350 dropped the two files that rode along, and this one cleans the block itself.

2. The Rust arm's red — containment, not retirement. I re-read declared_candidate_rivals_the_builtin (v1_compiler_infer_lookup.rs:179): it is strictly roster-based — primitive_projection_row_for_declaration(decl_ref(owner, name)), DivergentProjection => true. The only DivergentProjection row in the committed roster is (v2.std.collection, map_get_checked) (retained deliberately as the receipt-in-the-roster recording the closed divergence), and map_get_checked is not in the primitive registry, so builtin_callable_candidates is empty for it. There is no spelling in the current tree that is both a registered primitive and a divergent projected declaration — that absence is the repair working. Re-enrolling the minting red in this synthetic fixture would require either (a) a fixture-side forgery of the roster (the env never reads the real corpus, as you say), or (b) adding a false DivergentProjection row to the dag ledger so the generated mirror routes it. Both falsify the executing evidence rather than enrolling it.

The class's enrolled executing evidence therefore lives in the .dag tree, against real data: w_projection_census_declares_no_divergent_shared_spelling (folds the real roster; red the moment any row's decl_name equals its primitive's runtime name) and a_listed_import_does_not_exclude_a_transitively_reached_homonym (the live transitive-reach refusal at the int_max boundary homonym). The Rust arm holds the resolved consequence and pins it against the live registry + live roster; if either live surface changes to re-admit the collision, the tree witnesses go red. Labeled containment: the causal chain stops short of a Rust-injectable projection roster.

— sent from vivid-lynx-377

[gunbai-bot]

Review 77428, finding 1 — addressed at d3b5791. You are right that the comment was false and that the rival-join composition had lost its executing red; the fix splits along the seam the code actually draws.

Why the suggested fixture cannot mint the rival (verified against this head's code): parent_closure_callable_candidates takes the declared identity's decl_name from the LOOKUP KEY, not from sig.name (v1_compiler_infer_sigs.rs:339), so a fixture that declares "map_get" mints the identity (v2.std.collection, map_get). The compiled roster's only DivergentProjection row answers to (v2.std.collection, map_get_checked) (std_primitive_projection.rs:423-425), and the primitive registry keys the builtin as map_get (v1_compiler_infer_method.rs:719). Ambiguity needs ONE lookup name carrying both a rival declaration and a builtin co-candidate; after the rename no such name exists — the rival row's declaration and the registry key no longer share a spelling. That absence is the repair landing, not a hole.

What is now enrolled (d3b5791):

1. The seam arm's comment is rewritten to state exactly what the arm can and cannot see: it is the fork's permanent regression control (it reds if the fork returns), and the rival-join is no longer reachable at this seam by design, held at tree grain by w_projection_census_declares_no_divergent_shared_spelling.
2. New Rust control declared_candidate_rivals_the_builtin_discriminates_by_row_fidelity: the predicate is checked against the compiled roster's actual rows — the divergent declaration must rival, a modeled projection (empty_map) must not, an unprojected declaration must not, a builtin is never a rival to itself. A row that loses DivergentProjection fidelity, a predicate that stops consulting the roster, or a roster that drops the divergent specimen all fail here. cargo test -p v1-compiler --lib declaration_index: 8 passed, 0 failed, including both this control and the seam arm.

The composition red (builtin joins rival → FuncSigAmbiguous) is unrepresentable at this seam without manufacturing a shared production spelling, which is exactly what this PR exists to remove; the predicate-level control plus the census wall are the honest enrolled evidence. If you read the composition as still requiring a seam-level red, the mechanism that would carry it is a fixture-local roster injection, which the stage0 crate does not have — flag it and I'll take that up as its own change rather than widen this one.

[briansrls]

Approved at d3b5791c65eba1852b5bff68d54f12a067cece31.

The delta is coherent and remains fixture-only on the v1 side.

• The seam comment now states the actual reach. parent_closure_callable_candidates derives a declared identity's decl_name from the lookup key. The compiled projection roster's divergent declaration is (v2.std.collection, map_get_checked), while the builtin registry key is map_get. Therefore no current lookup spelling can mint both that divergent declared candidate and the builtin candidate. Recreating the old composition red inside this seam would require either a false shared production spelling or injectable/fabricated roster state; the absence is the rename's result, not missing coverage.
• The new declared_candidate_rivals_the_builtin_discriminates_by_row_fidelity test invokes the production predicate, whose implementation consults the generated primitive_projection_row_for_declaration. It pins the current divergent row true, the current modeled empty_map row false, an unprojected declaration false, and a builtin candidate false. Dropping the divergent specimen or changing its fidelity makes the positive arm fail.
• Commit d3b5791c65 changes only src/v1/stage0/src/declaration_index.rs. The other post-approval movement is main integration, including the std.optional re-home/re-point; the current PR file set contains no unrelated generated-registration residue.
• All five exact-head jobs in run 37574044032 passed.

Nonblocking precision: the comment's phrase that any predicate that stops consulting the roster must fail is stronger than this finite black-box control—a future implementation hard-coded to the current divergent identity could satisfy it. The test does establish the current real-row/fidelity behavior and the roster specimen's continued presence, which is sufficient for this delta; a future genericity proof could fold the compiled roster itself.

[briansrls]

Approved at 0edf0e5. Delta from approved d3b5791 is confined to the origin/main merge plus the two-file repair commit. floor_pure_producer_share's five new declared-route calls all move with the Outcome<Optional> authority to map_get_checked; outer Rejected/Accepted handling remains explicit through share_map_lookup_causes and the value readers still distinguish Accepted Present, Accepted Absent, and Rejected. map_lookup_dual_dispatch_witness rewrites only the two optional-vs-required comparisons to the already-authoritative Present/Absent match form; it avoids the constructed-Present equality realization defect and preserves the exact hit assertions. No other post-approval PR-owned change found. Exact-head run 37740552960 is green.

[gunbai-bot]

Superseded by #13641 at 634453d: this PR's head is an ancestor of integration/v1-closeout. The source branch is kept for archaeology; this PR is no longer an independent merge authority. — sent from neat-wolf-604