[youngwinds]

💡 Background and solution

Expose reusable, parameterized DSL compilation through compileQuery(dsl, { table, columns, dialect }), defaulting to MySQL. MySQL and DuckDB strategies own their SQL differences; dataset queries explicitly keep using DuckDB.

• Replace convertDSLToSQL with convertDSLToMysql and convertDSLToDuckDB; export the compiler and option/result types from both entrypoints.
• Align aggregation, date buckets, string comparisons/LIKE, NULL ordering, median and discrete quantiles. MySQL computes median/quantile with window expressions, including grouped and HAVING-only statistics.
• Maintain one supermarket-schema JSON catalog containing 394 DSL cases and one snapshot suite covering both dialects. Update the existing example generator, regression tests and test workflow docs without database test infrastructure.
• Default Node and browser storage to independent in-memory adapters through separate Node and browser strategy entrypoints. Export VQueryOptions; new VQuery({ enableIndexedDB: true }) explicitly selects IndexedDB persistence. Queries continue to run in memory.
• Add a changeset declaring @visactor/vquery: patch.

Callers of the removed convertDSLToSQL helper should select the explicit dialect helper, or use compileQuery for { sql, parameters } output.

🧪 Validation

• pnpm --filter @visactor/vquery run g: generated output reviewed; no new SQL snapshot differences.
• pnpm --filter @visactor/vquery run test:coverage: 595 tests passed; statements, branches, functions and lines all 100%.
• Fresh unit-only coverage before/after the storage change: 144 → 154 tests passed; all four metrics remain 100% → 100%. Both use Istanbul, src/** and the existing thresholds. Both independent strategy constructors are fully covered; neither runtime entrypoint is excluded from coverage.
• Root lint:check, format:check, typecheck --force and separate vquery test typechecking: passed; 24 typecheck/build tasks executed with no cache hits.
• changeset status --since=origin/main: passed.
• Browser smoke at http://localhost:3000/VBI/, the vquery storage documentation and aggregation examples: average 6,400 rendered correctly; no console warnings/errors observed.
• Real-browser storage acceptance with the built browser entry: default queries work in a sandbox that denies IndexedDB (total 30); opt-in IndexedDB correctly reports the sandbox restriction; real IndexedDB restores schemas/sources across instances; default memory storage remains isolated from persisted data. All four checks passed, and the temporary acceptance dataset was removed.
• Separate MySQL audit against 9,959 supplied CSV rows: all 394 DSL cases executed as parameterized and inline SQL (788 successful queries), with results matching an independent CSV/DSL evaluator. This audit did not add database tests or fixtures to the repository; 29 cases returned empty results on this dataset.

☑️ Self-Check before Merge

• Documentation and generated examples reviewed
• TypeScript definitions exported and checked
• Patch changeset provided
• Changes limited to vquery and its related tests/documentation