5.3 KiB
Constructor history correction
Scope: oa-core only, based on HOST f6a59de85b815fbea5ec578bb79d0a15819c42b4. No desktop, source grammar, API shape, or OA port semantics changed by the keyed closure correction.
Reproduction and decision
The original independent Boolean-to-Enumeration probe passes OA to SA on an unmodified HOST core. Extending the same fixture to SA to OA reproduces the duplicate Literal(incoming,incoming) conflict. New production-core regressions reproduce both the literal and Event.Signal member failures before the implementation change; constructor-closure-red.log records two failures. This is directional catalog projection, not UUID ordering: a live SA-only selector survives alongside the newly paired selector after a constructor copy.
Constructor closure now removes that redundant transient one-sided projection when a pair covers it. Established historical one-sided selectors remain, so changing their correspondence still requires an explicit reset. Both-sided historical pairs and removed branch observations retain their existing rules.
Suppressing the duplicate exposed another reverse-direction defect: target-local literals were lost because their implicit canonical key looked paired even when the source had no member. Preservation now uses observed membership (true or an optional row containing a value), with the existing prior source-membership guard preventing resurrection of reviewed removals. This implements the existing language.rst requirement to preserve target-only members when a historical switch reaches the target's current constructor; no normative contract was relaxed.
The intermediate normalization-only run (constructor-closure-normalized.log) passes Signal pairing but fails reverse literal preservation. Three final regressions cover both directions, no-history paired Signal insertion, established one-sided history requiring reset, local literal retention, and previously corresponding removed members staying removed. A separate read-only review checked selector normalization and presence semantics; its deletion-guard test recommendation was implemented.
Verification
Dedicated target: D:/CodexBuilds/Capella2-scenario-target.
$env:CARGO_TARGET_DIR='D:/CodexBuilds/Capella2-scenario-target'
cargo test --locked --offline -p oa-core --test r3_semantic
cargo test --locked --offline -p oa-core
cargo clippy --locked --offline -p oa-core --all-targets -- -D warnings
cargo fmt --all -- --check
cargo run --offline --manifest-path C:/Users/sorok/AppData/Local/Temp/capella2-constructor-literal-host-probe/Cargo.toml
All 31 semantic tests and all 356 Windows core tests pass. Strict core all-target Clippy and workspace formatting pass. The independently adapted probe passes both directions against the final implementation (constructor-independent-green.log). Full logs: constructor-core-test.log, constructor-core-clippy.log, constructor-r3-semantic-green.log in the dedicated target. Actual GUI/API acceptance and Linux integration remain with the root engineer; this core-only milestone makes no native acceptance claim.
Separate follow-up: review an existing OA Legacy port
The shared transition planner previously removed every typed SA port before resolving explicit existing-target mappings. Consequently an existing trace could not Keep or supply a Legacy Candidate in SA-to-OA semantic review: it failed with Review source is outside selected transition mappings. Two focused regressions reproduced that boundary on the committed keyed correction (existing-port-red.log).
The language contract permits same-kind Port traces and reviewed updates from either layer, while retaining OA Legacy-only compatibility. The planner now retains only explicitly mapped SA ports whose selected existing OA target is Legacy. It does not synthesize a target, change a port contract, or un-omit dependent SA implementation records. Structural validation still rejects typed OA Take/Candidate proposals. Automatic/new-counterpart omissions remain unchanged.
Three regressions exercise Keep and Candidate retaining Legacy, unchanged SA source, typed history source/JSON round trips, typed OA proposals failing without mutation, and an explicit new Port counterpart remaining an SA-only omission. A separate read-only review found no scope/invariant defect. Focused red-to-green result: all three pass (existing-port-green.log). This change uses the common planner consumed by desktop and REST; it adds no endpoint or transport-specific behavior.
$env:CARGO_TARGET_DIR='D:/CodexBuilds/Capella2-scenario-target'
cargo test --locked --offline -p oa-core --test r3_semantic reverse_ -- --nocapture
cargo test --locked --offline -p oa-core
cargo clippy --locked --offline -p oa-core --all-targets -- -D warnings
cargo fmt --all -- --check
All 359 Windows core tests pass, including all 34 semantic tests and existing portless-OA transition coverage. Strict core all-target Clippy and workspace formatting pass. Logs: existing-port-core-test.log and existing-port-core-clippy.log in the dedicated target. REST preview uses the corrected core planner at oa-server/src/routes.rs and commit uses the shared core prepare path; those handlers were inspected, not changed. GUI constructor tests and actual API/native acceptance remain with root integration; no native input or desktop code was changed here.