v1.14.0 — Dialect-Aware Transforms, Snowflake 100%, Schema Introspection

Headline themes: dialect-aware transforms, Snowflake at 100% of the QA corpus, ClickHouse significantly expanded (83% of the QA corpus, up from 53%), live schema introspection, SQL transpilation, and first-class integration sub-modules (OpenTelemetry and GORM). Drop-in upgrade from v1.13.0 — no breaking changes.

Added

Dialect-aware formatting (closes #479)

• transform.FormatSQLWithDialect(stmt, keywords.SQLDialect): renders an AST using dialect-specific row-limiting syntax — SELECT TOP n for SQL Server, FETCH FIRST n ROWS ONLY for Oracle, LIMIT n for PostgreSQL/MySQL/SQLite/Snowflake/ClickHouse/MariaDB
• transform.ParseSQLWithDialect(sql, keywords.SQLDialect): dialect-aware parsing convenience wrapper
• FormatOptions.Dialect field threads dialect through the formatter pipeline
• normalizeSelectForDialect() converts Limit/Offset into TopClause (SQL Server) or FetchClause (Oracle) on a shallow copy — original AST never mutated
• Parsed TopClause now renders (previously silently dropped — a long-standing bug)
• SQL Server pagination: OFFSET m ROWS FETCH NEXT n ROWS ONLY when both limit and offset are set
• PR #507

Snowflake dialect (100% QA corpus — 87/87 queries, epic #483)

• MATCH_RECOGNIZE clause (SQL:2016 R010): PARTITION BY, ORDER BY, MEASURES, ONE/ALL ROWS PER MATCH, AFTER MATCH SKIP, PATTERN, DEFINE (PR #506)
• @stage references in FROM clause with optional path and file format (PR #505)
• MINUS as EXCEPT synonym for Snowflake/Oracle (PR #494); fix for MINUS-as-alias edge case (PR #504)
• SAMPLE / TABLESAMPLE clause with BERNOULLI/ROW/SYSTEM/BLOCK methods (PR #501)
• QUALIFY clause for window-function filtering (Snowflake/BigQuery) (PR #490)
• VARIANT colon-path expressions (expr:field.sub[0]) (PR #496)
• Time-travel AT/BEFORE/CHANGES on table references (PR #495)
• LATERAL FLATTEN table function + named arguments (name => expr) (PR #492)
• TRY_CAST + IGNORE NULLS / RESPECT NULLS for window functions (PR #486)
• LIKE ANY/LIKE ALL and ILIKE ANY/ILIKE ALL quantifiers (PR #500)
• USE [WAREHOUSE|DATABASE|SCHEMA|ROLE] and DESCRIBE with object-kind prefixes (PR #491)
• COPY INTO, PUT, GET, LIST, REMOVE parse-only stubs (PR #499)
• CREATE STAGE/STREAM/TASK/PIPE/FILE FORMAT/WAREHOUSE/DATABASE/SCHEMA/ROLE/SEQUENCE/FUNCTION/PROCEDURE parse-only stubs (PR #498)
• CLUSTER BY, COPY GRANTS, and CTAS (CREATE TABLE ... AS SELECT) for CREATE TABLE (PR #504)
• ILIKE + PIVOT/UNPIVOT support for Snowflake dialect (PR #484)

ClickHouse dialect (69/83 QA corpus — 83%, up from 53% in v1.13.0, epic #482)

• Remaining ClickHouse QA gaps (tracked for v1.15): ARRAY JOIN / LEFT ARRAY JOIN, LIMIT n,m BY, named window definitions, scalar CTE subqueries, CREATE MATERIALIZED VIEW / DISTRIBUTED TABLE, SAMPLE with decimal and OFFSET, standalone SELECT ... SETTINGS without preceding FROM/WHERE
• Reserved-word identifiers: table, partition, tables, databases now valid as column/identifier names (closes #480, PR #481)
• Nested column types: Nullable(T), Array(T), Map(K,V), LowCardinality(T) in CREATE TABLE (PR #488)
• Engine clauses: MergeTree(), ORDER BY, PARTITION BY, PRIMARY KEY, SAMPLE BY, TTL, SETTINGS (PR #488)
• Parametric aggregates: quantile(0.95)(duration) double-paren syntax (PR #487)
• Bare-bracket array literals: [1, 2, 3] without ARRAY keyword (PR #485)
• ORDER BY ... WITH FILL with FROM/TO/STEP (PR #493)
• CODEC(...) column compression option (PR #497)
• WITH TOTALS in GROUP BY, LIMIT ... BY clause, ANY/ALL JOIN strictness prefix, DEFAULT as identifier (PR #503)
• SETTINGS clause on SELECT and in CREATE TABLE, TTL column clause, INSERT ... FORMAT (JSONEachRow, CSV, Parquet, etc.) (PR #489)

MariaDB dialect

• New DialectMariaDB extending MySQL (mariadb) (PR #431)
• SEQUENCE DDL: CREATE/DROP/ALTER SEQUENCE with full option set (START WITH, INCREMENT BY, MINVALUE/MAXVALUE, CYCLE, CACHE, NOCACHE)
• Temporal tables: FOR SYSTEM_TIME, WITH SYSTEM VERSIONING, PERIOD FOR
• Hierarchical queries: CONNECT BY with PRIOR, START WITH, NOCYCLE
• Added to playground and WASM dialect map (PR #432)

SQL Server enhancements

• PIVOT / UNPIVOT clause parsing (T-SQL) (PR #477)

SQL Transpilation (pkg/transpiler)

• Transpile(sql, from, to keywords.SQLDialect): composable rewrite-rule pipeline for cross-dialect conversion (PR #449)
  • MySQL → PostgreSQL: AUTO_INCREMENT → SERIAL/BIGSERIAL, TINYINT(1) → BOOLEAN
  • PostgreSQL → MySQL: SERIAL → INT AUTO_INCREMENT, ILIKE → LOWER() LIKE LOWER()
  • PostgreSQL → SQLite: SERIAL/BIGSERIAL → INTEGER, array types → TEXT
• gosqlx.Transpile() top-level convenience wrapper
• gosqlx transpile --from <dialect> --to <dialect> CLI subcommand

Live schema introspection (pkg/schema/db)

• Loader interface with DatabaseSchema/Table/Column/Index/ForeignKey types (PR #448)
• PostgreSQL loader (pkg/schema/postgres): introspects tables, columns (with primary/unique flags), indexes, and foreign keys via information_schema and pg_catalog
• MySQL loader (pkg/schema/mysql): introspects tables, columns, indexes, and foreign keys via information_schema
• SQLite loader (pkg/schema/sqlite): introspects via PRAGMA commands (pure Go, no cgo)
• gosqlx.LoadSchema() top-level dialect-agnostic convenience wrapper
• Integration tests using testcontainers-go v0.32.0 for PostgreSQL and MySQL loaders

DML Transform API (pkg/transform)

• AddSetClause, SetClause, RemoveSetClause, ReplaceSetClause for UPDATE statements (PR #446)
• AddReturning, RemoveReturning for INSERT/UPDATE/DELETE (PR #446)

Linter — expanded from 10 to 30 rules (PR #445)

• New rule categories: safety, performance, naming (alongside existing style and whitespace)
• Rules cover SQL injection patterns, missing WHERE clauses on UPDATE/DELETE, implicit type conversions, inconsistent identifier casing, table aliasing conventions, and more
• See docs/LINTING_RULES.md for full reference

Optimization Advisor (pkg/advisor)

• 12 new optimization rules: OPT-009 through OPT-020 (PR #464)
• gosqlx optimize CLI subcommand runs the full advisor pipeline

Fingerprinting (pkg/fingerprint)

• Normalize(sql) canonicalizes literals (WHERE id = 123 → WHERE id = ?) (PR #463)
• Fingerprint(sql) returns SHA-256 hash of normalized query for deduplication and query caches
• Integration with advisor and linter for aggregated findings

Integrations (sub-modules)

• integrations/opentelemetry/: InstrumentedParse() wraps gosqlx.Parse() with OpenTelemetry spans including db.system, db.statement.type, db.sql.tables, db.sql.columns attributes (PR #451)
• integrations/gorm/: GORM plugin that records executed query metadata (tables, columns, statement type) via GoSQLX parsing with GORM SQL normalization (backtick identifiers, ? placeholders); exposes Stats() and Reset() APIs (PR #452)
• CI workflow for integration sub-modules (.github/workflows/integrations.yml)

Parser / AST additions

• DDL formatter, CONNECT BY hierarchical queries, SAMPLE clause (PR #472)
• JSON function parsing improvements (PR #460)
• gosqlx stats CLI subcommand exposes object pool utilization (PR #459)
• gosqlx watch CLI subcommand for continuous validation (PR #458)
• gosqlx action GitHub Actions integration with inline annotations (PR #443)

C binding

• Coverage hardened from 18% to 93% via comprehensive test suite (PR #447)

Docs, website, and CI

• "Who's Using GoSQLX" section in README (PR #475)
• OpenSSF Scorecard security analysis workflow (PR #443)
• Sentry → GitHub Issues automation (PR #438)
• Website: mobile responsiveness improvements (PR #441), comprehensive a11y/UX audit (PR #440)
• MariaDB + Snowflake added to playground dialect dropdown (PR #432)

Changed

• OpenTelemetry SDK bumped from v1.42.0 to v1.43.0 to address CVE-2026-39883 (PR #502)
• Linter rule count exposed in glama.json, docs, and CLI help (10 → 30)
• Docs: SQL compatibility matrix updated to reflect Snowflake and ClickHouse 100% pass rate

Fixed

• SQL Server TOP rendering: parsed TopClause values now correctly render in formatted output (PR #507). Round-trippers that parse SELECT TOP 10 * FROM users and format again will see TOP 10 preserved — previously it was silently dropped.
• Snowflake ILIKE + PIVOT/UNPIVOT parse-time fix (PR #484)
• MINUS consumed as select-list alias: SELECT 1 MINUS SELECT 2 now correctly treats MINUS as a set operator (PR #494)
• ClickHouse production playground WASM 404: committed gosqlx.wasm to git, removed auto-commit step blocked by branch protection (PRs #423, #424)
• Playground React error #310: useState/useCallback now above early returns (PR #429)
• Website: Vercel Analytics, Speed Insights, and CSP fixes (PR #433); Sentry hydration mismatch (PRs #434, #437, #439)
• CI: testcontainers skip on Windows; .trivyignore entries for unfixable transitive CVEs; graceful handling of glama-sync on non-GA runners

Deprecated

Carried over from v1.13.0 (no new deprecations in v1.14.0):

• parser.Parse([]token.Token) — use ParseFromModelTokens instead
• ParseFromModelTokensWithPositions — consolidated into ParseFromModelTokens
• ConversionResult.PositionMapping — always nil, will be removed in v2

Security

• CVE-2026-39883 (HIGH, go.opentelemetry.io/otel/sdk): resolved by upgrading to v1.43.0 (PR #502)
• .trivyignore entries audited and documented — all remaining ignores are transitive test-only (Docker via testcontainers) or npm-only (website build-time dependencies, not shipped in Go binaries)

Companion releases

• VS Code extension: bumped to 1.14.0 — tracks library version, no behavioral changes
• MCP server: bumped to 1.14.0 — glama.json updated with MariaDB and ClickHouse in dialect list
• pygosqlx Python bindings: bumped to 0.2.0 — Python bindings follow an independent semver track (alpha) since they have not yet received the same QA sweep as the core library