nocter-workspace-analysis

Responsibility

Turn accepted document revisions into one frozen workspace topology, compilation demand, and set of immutable analysis generations.

Contract

The crate consumes canonical workspace roots, installation facts, and linear WorkspaceSourceRevision values. It prepares package, toolchain-standard, or single-file scopes, retains the shared compiler-computation owner across revisions, supplies its reusable source syntax to discovery, invokes one closed analysis demand per affected scope, and publishes normalized workspace outcomes to the language server. Each accepted transition exposes both the generations it recomputed and one complete immutable view of every active generation after the transition.

Internal Responsibilities

• revision causality and owner-sequence validation
• package-root catalog construction and topology freeze
• complete module-root compilation input construction
• scope caching, invalidation, and generation publication
• atomic publication of the complete post-transition scope set
• accepted workspace revision publication into the shared compiler-computation owner
• source declaration-surface and module-surface query composition
• discovery through the shared computed syntax provider
• transport of only the final complete-or-incomplete unit product to session

Invariants

• Every document has exactly one selected or rejected topology result per revision.
• Topology is computed once and cannot vary between documents in the same revision.
• Compilation receives the exact package-root catalog that topology used; it cannot repeat root selection against the overlay.
• Ambiguous shared-source contexts are rejected rather than ordered.
• Changed/closed files may invalidate demand but cannot create demand.
• Foreign, cloned, or non-increasing revisions cannot mutate latest state.
• Consumers of URI-global state receive the complete post-transition generation set; they never reconstruct current ownership by replaying scope invalidation instructions.
• Overlay membership, each open source's bytes, and the filesystem epoch are separate inputs; editing one open source cannot dirty unrelated source-text queries.
• Reused parse products are rebound through the syntax-owned text/identity contract; workspace code cannot rewrite syntax identities itself.
• Speculative mutation validation owns an isolated computation database and cannot read or mutate the accepted revision's query state.
• A module surface depends on each member source's canonical declaration syntax. A body-only edit may reevaluate that source boundary, but its unchanged fingerprint prevents module recomputation.
• Compiler computation publishes semantic and exact-current fingerprints from the same shared discovery snapshot. Workspace orchestration cannot inspect declaration storage, body recipes, fingerprint construction, or invalidation rules.
• For source-complete input, workspace passes only a closed semantic success/rejection product to session. Missing declaration, preparation, or finalization products are integrity errors; they cannot select a session fallback that reruns compiler stages.
• Workspace demands one top-level unit-analysis query rather than selecting complete versus incomplete syntax or demanding declaration, preparation, lexical, typed, finalization, and recovery products separately.
• Workspace neither enumerates body queries nor transports intermediate lexical or typed sets. Semantic computation owns complete-set demand and canonical finalization order.