Salsa Incremental Query Engine

ModelScript's semantic analysis and symbol resolution pipeline is driven by an incremental query computation engine inspired by the open-source Salsa incremental query framework.

The engine lives in @modelscript/runtime and provides memoized, demand-driven evaluation across multi-file engineering workspaces.

---

The Query Life Cycle

Every query executed in ModelScript follows a 4-phase lifecycle to ensure maximum cache re-use and correct invalidation:

Rendering diagram...

The 4 Stages

1. fetch: Checks if the query result exists in the memoization table.
2. deep_verify: If the workspace revision has advanced, transitively checks whether any input queries or files that contributed to this result have actually changed.
3. execute: Re-runs the query definition if inputs have genuinely changed, tracking any newly accessed queries.
4. backdate: If re-executing the query produces an output value identical to the previous result, the query's revision is backdated to prevent cascading invalidation down the dependency tree.

---

Key Compiler Queries

The query engine handles semantic resolution, type checking, and specialization:

Query	Scope	Description
instantiate	Class	Resolves all direct components, inherited elements, and nested declarations of a class.
classInstance	Component	Specializes a generic class definition with specific modifier expressions.
resolveSimpleName	Scope	Performs hierarchical lexical scoping and imports resolution.
variability	Component	Computes effective variability: constant, parameter, discrete, or continuous.
causality	Component	Computes signal causality: input, output, or internal.
effectiveModification	Hierarchy	Resolves and merges class modification clauses across multi-level inheritance graphs.

---

Salsa & Linear CST Synergy

AST wrapper nodes are never stored inside the Salsa cache. Storing heavy AST objects would duplicate memory and defeat garbage collection guarantees.

Instead:

• The Salsa cache stores semantic integer handles and primitive metadata (e.g. SymbolId, resolved TypeInfo, variability enums).
• Concrete CST nodes are retrieved on-demand from linear memory using db.cstNode(symbolId) only when lowering or evaluating equations.