gudusoft.gsqlparser.dlineage.dynamicsql

Class RoutineCatalog

Object

gudusoft.gsqlparser.dlineage.dynamicsql.RoutineCatalog

public final class RoutineCatalog
extends Object

Analysis-scoped index of routine DEFINITIONS and top-level candidate CALL-SITE statements across the units of a multi-file analysis (plan §5.6, Phase 5 slice 3 — the seed of the shared routine/signature index).

Lazily built and memoized per unit text: a unit is parsed (through the shared, caller-owned parse cache) and its statement list walked at most ONCE per analysis, no matter how many procedures consult the catalog. Callers keep their own cheap contains-prefilter, so units that never mention a routine of interest are still never parsed at all — cost semantics are identical to the previous per-procedure scans, minus the repeated walking.

The catalog only INDEXES; every matching, vetting, and ambiguity decision stays in DynamicSqlLineageResolver so this refactor is behavior- preserving by construction.

How this differs from TSQLEnv / TSQLProcedure

The sqlenv package already keeps a procedure registry (gudusoft.gsqlparser.sqlenv.TSQLProcedure under TSQLSchema, with typed parameters and even the definition source string). The two serve different layers and are NOT redundant:

• AST working set vs metadata registry. TSQLEnv stores metadata — names, parameter TYPES, the definition as a STRING — and serves NAME RESOLUTION (resolver2/dlineage resolving a.b.c against catalog objects). This catalog stores live AST nodes plus each statement's unit provenance and effective USE database/schema, because the dynamic-SQL binding machinery must hand the callee's AST to the evaluator/interpreter and extract literal arguments from caller ASTs. An env definition string would have to be re-parsed anyway — which is exactly what this catalog's parse cache does, while also indexing what the env has no concept of at all: call sites.
• Duplicate visibility vs registry merging. An env is keyed by qualified name, so re-registering a same-named routine merges or overwrites. The binding rules ("never guess") need the opposite: every definition OCCURRENCE with its origin unit must stay visible so a same-named duplicate can be detected and refused. The information a registry naturally collapses is precisely what the ambiguity verdicts here require.
• No environment assumed. dlineage frequently runs with no TSQLEnv attached (and env construction is option-gated); this catalog depends only on the analysis unit texts themselves.
• Planned convergence. Plan §5.6 promotes one routine/signature index shared by production dlineage and Analyzer v2. This class starts analysis-side because behavior preservation was the gate; unifying it with env registration and v2's BoundProgram is later Phase 5/6 work.

When metadata already contains stored-procedure definitions

• Metadata carrying routine SOURCE (sqlflow JSON sourceCode, grabit exports, DDL dumps) already flows in with no extra wiring: those definitions become ordinary analysis units, so this catalog indexes them like any other file while SQLEnvParser independently registers them for name resolution.
• Metadata carrying only SIGNATURES (name + parameter types, no body — the catalog-input-interface direction) cannot produce lineage THROUGH a body, but enables correct IN/OUT binding at call sites, overload disambiguation, and a "known external routine, effects unknown" diagnostic instead of a bare unresolved.
• Future slice ("env-backed routine bodies"): when an attached env holds a TSQLRoutine definition string for a routine that is NOT among the analysis units, this catalog could parse that definition as a virtual unit — covering the "caller file plus metadata-only definition" scenario.

Constructor Summary

Constructors

Constructor and Description
RoutineCatalog(EDbVendor vendor, List<String> unitTexts, Map<String,TStatementList> parseCache)

Method Summary

Modifier and Type	Method and Description
static String	numberedGroupAfter(TObjectName name) The ;N group of a deprecated numbered procedure, scanned from the source tokens right after the name — the parser drops it from the AST name entirely.

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

RoutineCatalog

public RoutineCatalog(EDbVendor vendor,
                      List<String> unitTexts,
                      Map<String,TStatementList> parseCache)

Method Detail

numberedGroupAfter

public static String numberedGroupAfter(TObjectName name)

The ;N group of a deprecated numbered procedure, scanned from the source tokens right after the name — the parser drops it from the AST name entirely. ;1 normalizes to null (p and p;1 are one identity in T-SQL).