Files
ARCADIA-Studio/harness/release-port-core.md
T

47 lines
7.4 KiB
Markdown

# Typed port core milestone handoff
Implemented in `D:/CodexWorktrees/Capella2-typed-ports`, branch `codex/typed-port-contracts`, based on `c20a207`. This is the core portion of [the adopted port contract](release-port-contract.md). It must be integrated with the migration/source worker's companion changes; desktop/native acceptance remains the root task's work.
## Delivered behavior
- Schema 3 persists `PortContract::{FunctionInput, FunctionOutput, Component, Legacy}`, preserved `LegacyPortContract`, Standard/Flow component kind and separate orientation, mandatory FE port UUIDs, `BoundaryExchange.oriented`, and identified `PortAllocation` relations. Legacy association data is preserved through typed resolution and remains reference-aware without becoming an invented provided/required declaration.
- Typed function/component ownership, direct item declarations, empty-contract handling, actual work allocation, endpoint direction, same-layer references, unique PortAllocation pairs, explicit interface roles, and exchange/port allocation correspondence are validated in `ports.rs` through the normal candidate validator. Standard-to-Flow component exchanges are structural errors, reflecting the root's later native observation. Legacy compatibility remains explicitly warned.
- Public `plan_functional_exchange`, `plan_component_exchange`, `plan_legacy_port_resolution`, and `plan_legacy_work_port_split` implement the contract request types and return concrete existing commands. Public `plan_function_port_items` supplies explicit item synchronization. Planners validate complete candidate batches; intermediate generation does not modify the original project.
- FE creation reuses exactly one compatible typed endpoint or creates deterministic new endpoints. Explicit incompatible reuse fails; ambiguous Auto creates a fresh port. Component generation creates/reuses both component ports, two PortAllocations, the boundary exchange, and ExchangeAllocation together. It requires actual work allocations and never invents interfaces or provider/requirement roles. Repeated generation with an existing boundary reuses facts and identities.
- Split preserves the original legacy record, primary UUID/alias/prose, artifacts, constraints, and traces; it creates one opposite typed port and rewires source/target FE references. Resolution retains original interface metadata and legacy role identities. API/source can represent preserved legacy roles on resolved ports, with clear warnings; normal typed function ports use item contracts.
- Reidentify visits typed item declarations, required FE endpoints, legacy interface associations, and allocation relations. Deleting a function port cascades incident FEs; deleting an FE retains independent ports. Deleting a work Allocation cascades dependent PortAllocations, and deleting a required PortAllocation cascades its dependent ExchangeAllocation. Interface deletion clears the preserved optional association and removes role relations while retaining items/ports.
- OA-to-SA Interaction creation creates typed SA endpoints through the FE builder. SA-to-OA FE mapping remains portless, and Required scope does not copy even Legacy SA endpoint ports solely because an FE used them. Typed SA ports, boundary implementation, and dependent role/allocation records have explicit omission warnings. Original SA objects remain intact. Existing counterpart updates retain their endpoint identities.
- The demo now uses typed ports, explicit function/component port allocations, and port-level interface roles. Projection anchors derive from visible owners without requiring port placements; orientation is independent of anchor side and actual exchange endpoint IDs are retained.
## Integration APIs and changed rendering fields
All port helpers and request types are re-exported through `oa_core::*`; the bounded implementation module is `oa_core::ports`.
- `make_port(project, id, owner, name, contract) -> Result<Element, ModelError>` builds an explicit port element for a caller's transaction. It does not adopt a model or independently validate unrelated candidate state.
- `ElementBody::new(Kind::Port, ..)` now returns None: contextual authoring must choose a typed contract and owner. Do not restore a generic InOut default to make old forms compile.
- `RelationBody::new(FunctionalExchange, ..)` is an explicitly unbound draft with nil endpoint IDs; it cannot pass candidate validation. Normal GUI/API authoring should use the public planner or a concrete complete Batch. `Command::Connect(FunctionalExchange, ..)` and `Project::relate(FunctionalExchange, ..)` now create/reuse real endpoint ports. The latter remains the existing unchecked Project builder; Session validates its resulting transaction.
- `ProjectedPort.direction` is replaced by `side: PortAnchorSide::{Left, Right, Bottom}`. `projection::port_anchor_side(&Element)` is public. Glyph semantics must inspect the element's actual PortContract; Standard/Unset and InOut Flow are not interchangeable.
- `ProjectedEdge` adds `oriented: bool`; BoundaryExchange uses its persisted value and other edges use true. The root's newer main-only `summary: bool`, routing, and layout changes must be merged alongside this field, not overwritten by this older checkpoint's projection body.
- `item_satisfies` is identity equality until R6 specialization lands. `items_satisfy`, `actual_performer`, `component_allows`, `has_port_allocation`, and `port_contract` provide the small shared validation/authoring seams.
Preview binding to the current revision/fingerprint and UI display of non-FE references during split belong to the existing desktop preview/usage surface. Port planners return concrete commands; they do not introduce a second preview transaction engine. The UI must present preserved legacy associations distinctly and expose source/form/connection paths using these helpers.
## Verification
Build output is isolated at `D:/CodexBuilds/Capella2-typed-ports-core`.
- `cargo test -p oa-core`: all 124 tests passed after core/source/migration integration, including unchanged legacy admission/location tests. One additional directed-legacy-resolution regression was then added to the focused port suite.
- `cargo test -p oa-core --test release_ports`: all 9 focused tests passed, covering atomic creation/reuse/collisions, actual allocation/generation, provider/requirement independence, Standard/Flow rejection, Legacy split and resolution, preserved attachments and references, deletion/remap/undo, portless transitions, and contract-aware geometry.
- `cargo clippy -p oa-core --all-targets -- -D warnings`: passed after the final core/test changes.
- Existing scenario14, diagram3, and workflow28 suites pass, including remapping/deleting every demo identity and checking the remaining candidate. Source/migration tests and server checks are owned/reported by the companion worker.
No CUA, desktop/native acceptance, deployment, future R6 item specialization, or R3 semantic baseline engine is claimed. Current R3 name/description-only drift remains the separately documented release seam; this milestone does not disguise it as completed semantic review.
## Owned commit scope
Core: `model.rs`, `ports.rs`, `lib.rs`, `validation.rs`, `references.rs`, `session.rs`, `transitions.rs`, `demo.rs`, `projection.rs`.
Tests: new `release_ports.rs`; typed fixture adaptations in `workflows.rs` and `release_scenarios.rs`.
Handoff: this file only. Migration/source/server/documentation files and their tests belong to the companion worker and are intentionally excluded from the core commit.