Skip to content

sql-parser-candidate-review.md

Latest commit

 

History

History
48 lines (41 loc) · 4.16 KB

sql-parser-candidate-review.md

File metadata and controls

48 lines (41 loc) · 4.16 KB
Raw

SQL parser candidate source review

TASK-SQLC-R12 / REQ-SQLC-001/006/009 under Accepted ADR-065. Read-only primary-source research, 2026-10-03. Parser selection, package installation, syntax execution and native client qualification remain pending. This review complements the183-command inventory; it does not prove it passes.

Candidate / inspected source Finding and remaining evidence
SqlParserCS, source e1929977a96290f937e9a2a67e3c823b7dc25cd2 Managed C# syntax AST, multiple dialects; maintainers explicitly describe incomplete dialect support and no comprehensive conformance assessment. Source targets net10, but stable package0.6.5 equivalence to this source is unverified. No typed execution or demonstrated bounded PG18 conformance.
libpg_query18.0.0, source 204fbdbd3ed5f8691ab358e49f1fc5397b4679e2 Tagged PostgreSQL18-derived native C parser. Strongest inspected version-specific syntax reference, an integration-owner inference from primary source. Parse structures alone do not bind KeyLoad schemas/types, authorize plans or execute operations. Native packaging/lifetimes/resource limits require a separate accepted boundary contract.
pgsqlparser, source cf04a7e14fc0adf51e6ac2d834a91d72ecb1d14a .NET wrapper with bundled native libraries and parse-tree API. Inspected README example reports parser170005; exact published package/native-source PG18 provenance is unverified. Async APIs alone do not prove interruptible native parsing.
GrepSQL, source db641f1f017f33b6815038b094e9e779d53d4b15 Native wrapper/AST tools. Inspected README example reports170004; RID declarations are not executed deployment evidence. No verified PG18 package binding.
PostgresQuery0.1.4 Syntax wrapper; exact source/native parser revision is unverified. Package documentation explicitly leaves Windows/macOS native binaries untested. No KeyLoad interoperability proof.

Root retains the PostgreSQL18 working target without selecting a parser. Investigate the tagged upstream parser and exact wrapper/native provenance first; evaluate a managed alternative only against a maintained differential corpus. Neither a broad dialect list nor accepting a command completes full SQL support. No exact missing-construct matrix or comparable measurements were established.

Before implementation freeze SQL byte/token/depth/AST/node limits, allocation and CPU budgets, thread safety, ABI/RIDs, allocator/free ownership, native failure containment and cancellation/drain semantics. Cancelling a Task does not establish that native work stopped. Parsing cannot bypass end-to-end admission or error contracts. KeyLoad still owns typed binding, null/type/coercion semantics, persisted authority, read cuts, bounded plans and canonical RF3 mutations.

flowchart LR
    Input[Bounded SQL input] --> Parser[Version pinned syntax parser]
    Parser --> AST[Bounded syntax structure]
    AST --> Binder[KeyLoad typed schema and authority binding]
    Binder --> Plan[Bounded effect aware plan]
    Plan --> Execute[Canonical request and RF3 execution]
Loading

Qualification requires real PG18 differential syntax/execution cases plus negative, malformed, nested, oversized, cancellation and resource fixtures. Public RF3 SQL and native driver flows must independently qualify every advertised family. Parser performance needs actual comparable GitHub measurements; no tests, runtime/native calls, packages or dependency changes were performed by this review.

Research packet SHA256 6128727b4fd8eda9bbb964fe94b5b79d8b6042205a24ac0e4af08f2f0284737f; machine-readable source inventory SHA256 0622f3a6469da40262ffc9009847f3055aa2fd06cdf6246999ec87206e6a7b9e. Pinned upstream links remain the durable source references; private research is manual architecture evidence, not product qualification or uploaded CI artifacts.