QQL 0.2.0

[0.2.0]

📦 Packaging (VS Code only)

• VS Code extension 0.2.1 — Marketplace packaging bump (immutable 0.2.0 slot). Ships the same QQL 0.2.0 WASM parser; crate / SDK versions stay at 0.2.0. Fixes analysis feedback loops (CodeLens / host thrash), UTF-8 statement slicing for explain, shared analysis cache, Output channel for plans/routes, and Biome + typecheck scripts. VSIX binaries are no longer committed; build via npm run package (@vscode/vsce).

🚀 Added

• QQL documentation website — a new docs site at qql.veristamp.in (Astro + Starlight) replaces the legacy playground. 40+ pages across Start, Language, Guides, Edge, SDKs, Tools, Reference, and Contributing, including a dedicated Edge runtime section (9 pages), a Filter injection security guide, an Error codes reference, and editors documentation for the VS Code extension. All language, CLI, SDK, and reference pages were rewritten and deepened.
• Interactive playground — integrated into the site and backed by a freshly built qql-wasm bundle. Every documented example is extracted and parsed at build time, so the docs cannot drift from the parser.
• Grammar as the single source of truth — qql-grammar-gen now derives five artifacts from language/v1/grammar.pest: the generated pest grammar, the VS Code TextMate grammar, the VS Code and playground keyword tables, and the Rust KEYWORDS map (keywords.generated.rs) used by the qql-core lexer.
• Language & conformance — new conformance fixtures for CROSS RERANK and formula division defaults, grammar support for USING MULTI/MULTIVECTOR/IMAGE embedding specs, and a documented contract map in language/v1/README.md. The conformance suite now stands at 35 valid files (249 statements) / 53 invalid cases / 35 AST snapshots.
• Python typing — .pyi stubs and py.typed markers for pyqql and pyqql-edge.
• Executable grammar contract — qql-conformance now compiles language/v1/grammar.pest in test-only code and checks the canonical grammar against the fixture corpus without adding Pest to the qql-core runtime parser.
• VS Code 0.2.0 — live diagnostics, hover plans, CTE go-to-definition, document symbols, folding, CodeLens, commands, status bar state, contextual completions, QQL snippets, and Markdown fenced-block support.
• Updated examples — WASM examples now use the checked-in generated package, and the edge example documents the required opt-in CLI feature.

🏗️ Architecture

• Lexer/grammar lockstep — the TokenKind enum stays hand-written, but its keyword map is generated and guarded by bi-directional drift tests (grammar → keywords, and parser keyword vocabulary → grammar), and qql-grammar-gen check rejects grammar rules unreachable from the entry productions. A dead sharding_method_val rule and its unused TokenKind variants were removed.
• Parser-generation roadmap — docs/parser_generation_design.md evaluates pest/LALRPOP/custom generators against the no_std, zero-dependency core and lays out a phased migration.
• Shared editor analysis — the extension now performs one debounced WASM analysis per document and shares the result across diagnostics, hover, symbols, CodeLens, and completion providers.

🔒 Changed & Scoped

• Edge fail-loud hardening — PARAMS (timeout) and PARAMS (consistency) are now rejected on the edge backend with QQL-EDGE-UNSUPPORTED-TIMEOUT / QQL-EDGE-UNSUPPORTED-CONSISTENCY instead of being silently ignored.
• Fail-closed set_shard_key in the Python, Node, and WASM bindings — assigning a shard key to a statement type that does not support routing now raises an error.
• Installer & release tooling — install scripts validate released platform targets (with a build-from-source hint otherwise) and RELEASING.md was aligned to 0.1.5 (edge verification notes the --features edge requirement).
• Versioning — language/v1/spec/versioning.md documents the post-0.1.5 contract corrections and hardening.
• Language version 1.3 — the specification now documents multivector and image embedding directives, strict COUNT and shard-key configuration, and the current 35 valid / 249 statement / 53 invalid / 35 snapshot conformance corpus.
• Fallible planning surface — removed the panicking qql_plan::routing::route() helper and the shard-key-dropping top_level_filter_with_shard() helper; use try_route() or compile_statement().

🐛 Fixed

• Parser/grammar alignment — fixed triple-quoted string preservation, four-quote SQL strings, malformed numeric literals, non-finite floats, COUNT clause ordering, index-type validation, rerank inputs, feedback strategy validation, raw-string prefix handling, and identifier segment validation.
• Planner and execution safety — replaced unsupported PREFETCH panics, integer overflow, and silent prefetch degradation with structured errors; preserved CTE-backed RERANK prefetches.
• Backend correctness — fixed DELETE PAYLOAD batching, edge reads that created missing collections, gRPC/edge point-lookup envelopes, gRPC RRF parameters, and lossy gRPC filter conversion.
• SDK and CLI correctness — aligned Python and Node declarations with native APIs, fixed Node edge HTTP embedding and model forwarding, corrected CLI REPL ANSI output, and rebuilt the bundled editor WASM as a release artifact.

📚 Documentation

• Edge timeout/consistency behavior, error codes, backend compatibility, and API surfaces updated to match the hardened runtime.
• Documentation validation now rejects unwrapped QQL/SQL fences, parse-and-plan checks examples and fixtures, and verifies same-site documentation anchors.

🔗 Major release work

• Core language and execution hardening — PR #25 — grammar/runtime conformance, parser and AST fixes, planner safety, backend correctness, binding alignment, and structured error handling.
• VS Code packaging cleanup — PR #26 — removed the obsolete packaging path, corrected completion output, and added editor source checks.
• Website validation and documentation — PR #27 — executable documentation checks, raw-fence validation, anchor checking, and updated reference content.
• VS Code language intelligence — PR #28 — full editor-host analysis, providers, commands, snippets, release WASM, and the 0.2.0 VSIX.