v0.3.5

Full Changelog: v0.3.4...v0.3.5

v0.3.4

Full Changelog: v0.3.3...v0.3.4

v0.3.3

[0.3.3] - 2026-05-19

Added

• man/sqlt.1 system man page shipped with the crate tarball. After brew install codedeviate/cli/sqlt (or any other install path that drops the file onto $MANPATH) man sqlt renders the full reference: NAME, SYNOPSIS, DESCRIPTION, COMMANDS, DIALECTS, ENCODINGS, LINT RULE CATEGORIES, TRANSLATION WARNINGS, EXAMPLES, EXIT STATUS, FILES, ENVIRONMENT, SEE ALSO, BUGS, AUTHOR.
• Colorized --examples output across every subcommand, matching the recon-style scheme (yellow-bold section headers, bold sub-headings, cyan commands, green shell comments, dimmed dividers and notes). Respects NO_COLOR / CLICOLOR_FORCE.
• README.md header now ships a badge row (GitHub, latest release, crates.io, Homebrew tap, Rust edition / MSRV, MIT license).

Changed

• Default SIGPIPE behaviour is restored at startup on Unix, so sqlt man | less (or any other pager / head / grep) closing early ends the process quietly instead of panicking on the next println!.

v0.3.2

[0.3.2] - 2026-05-08

Added

• LICENSE file containing the full MIT License text. Previously the repo declared a license in Cargo.toml but shipped no LICENSE file.
• Cargo.toml metadata required for cargo publish to crates.io: authors, readme, homepage, documentation, keywords (sql, parser, translator, dialect, lint), categories (command-line-utilities, database, parser-implementations), and an exclude list (.github/**, .idea/**, target/**, /.gitignore, OUT-OF-SCOPE.md, CLAUDE.md, BREW.md) that keeps maintainer-facing files out of the published tarball.
• [profile.release] settings (lto = true, codegen-units = 1, strip = "symbols") so binaries built by Homebrew (and anyone running cargo install --path .) are smaller. Trade-off: longer release build (~3 minutes vs. seconds).
• README.md install section now documents Homebrew (brew install codedeviate/cli/sqlt) and crates.io (cargo install sqlt) alongside the existing cargo install --path . source build.

Changed

• Homebrew distribution now lives in the consolidated codedeviate/homebrew-cli tap; the in-repo Formula/sqlt.rb template was retired in favour of the tap's authoritative copy. The crate tarball no longer carries a Homebrew formula.
• Cargo.toml license field narrowed from MIT OR Apache-2.0 to MIT to match the single LICENSE file in the repository.
• Cargo.toml repository field corrected from https://github.com/thomasbjork/sqlt to https://github.com/codedeviate/sqlt (the actual remote).
• LICENSE copyright holder set to codedeviate to match the publishing identity used on crates.io and Homebrew.
• README.md License section updated from the stale "Dual-licensed under MIT or Apache-2.0" text to a single MIT note that links to LICENSE.

v0.3.1

[0.3.1] - 2026-05-03

Added

• Top-level sqlt --examples flag. Prints a cross-cutting overview with example invocations of every subcommand. Complements the existing per-subcommand sqlt <COMMAND> --examples.

Changed

• sqlt build-schema --help now renders the full long-form description (use cases, what's tracked, version handling) via a BUILD_SCHEMA_LONG_ABOUT constant. Previously only the one-line summary was visible from --help, requiring users to guess the syntax.
• Per-flag descriptions on BuildSchemaArgs (every flag) describe accepted values, defaults, and interactions in line with the help-text policy in CLAUDE.md.
• sqlt build-schema --examples rewritten to a four-section format: Flag reference (every flag, with defaults), Common workflows (covering each commonly-used flag combination), What gets tracked, Versioning, Exit codes. Includes process-substitution and jq-piping recipes.
• CLAUDE.md "Help & examples maintenance rule" expanded with concrete standards: every subcommand variant must have a long_about (not just a one-line ///), every flag mentioned in --help must appear in --examples, both sqlt --examples and sqlt <COMMAND> --examples must work. Listed as mandatory for every change that affects how the utility is used.

v0.3.0

[0.3.0] - 2026-05-02

Added

• --schema <file> flag on sqlt lint (repeatable). Parses each file as schema input — CREATE TABLE, ALTER TABLE, DROP TABLE, CREATE INDEX, foreign-key constraints, CREATE DATABASE, USE. Replays them in CLI order so the schema reflects its current state, not its initial CREATE. Schema files are NOT linted (only feed the model). Statements that don't affect the schema (INSERT, GRANT, DELIMITER, stored-procedure bodies, …) emit note: skipping <kind> at <file>:<line> on stderr. Files with .json extension are loaded as a previously built artifact (see build-schema).
• sqlt build-schema subcommand. Compiles one or more --schema files into a JSON artifact that can be consumed by --schema schema.json later. Use case: a long migration history that takes time to parse + replay can be compiled once, checked into the repo, and re-used by every CI lint run. Mixing .json + .sql is supported: sqlt lint --schema base.json --schema migrations/late.sql query.sql.
• Per-database namespacing. Schemas now track CREATE DATABASE and USE; tables live in per-database namespaces. Queries can use 3-part db.table.col references; the USE cursor persists across --schema files in CLI order. Same-named tables in different databases no longer collide (e.g. shop_db.orders and global_db.orders).
• Full DDL replay in Schema::apply_statement: CREATE/ALTER/DROP TABLE, ADD/DROP/MODIFY/CHANGE/RENAME COLUMN, RENAME TO (within and across DB), CREATE [UNIQUE] INDEX, ALTER TABLE ADD INDEX/UNIQUE/PRIMARY KEY/FULLTEXT/SPATIAL, ADD CONSTRAINT FOREIGN KEY, DROP INDEX/CONSTRAINT/FOREIGN KEY/PRIMARY KEY, CREATE TABLE LIKE (column copy from referenced table), CREATE OR REPLACE TABLE. IF EXISTS/IF NOT EXISTS guards are honored. Missing-target ops emit a stderr note: but never error.
• Schema is now serializable via serde for the JSON artifact path. Indexes and foreign keys are tracked in the model (no rule consumes them yet, but the data is available for future SQLT0503-style refinements).
• ~17 new schema unit tests covering each DDL shape, plus 6 new CLI integration tests for --schema, multi-DB namespacing, late-migration mixing, and the build-schema round-trip.

Added (earlier this cycle, still in 0.3.0)

• Schema-aware lint pass. lint::lint now builds a Schema model from every CREATE TABLE statement in the input before running rules and passes it to each rule via LintCtx::schema. Lookups are case-insensitive; NOT NULL and PRIMARY KEY are tracked per column.
• New rule SQLT0900 unknown-column (Schema category, error, default-on). Fires when a table.col reference resolves to a CREATE TABLE in the same input but the column doesn't exist on that table. Also catches typos in the INSERT INTO Foo (col1, col2) column list. Conservative — never warns about CTEs, derived-table aliases, or tables defined elsewhere.
• New rule category Schema (SQLT09xx namespace) for rules that consult the schema model.

Changed

• SQLT0505 count-of-nullable-column is now schema-aware: when a CREATE TABLE for the referenced column is present and the column is declared NOT NULL, the warning is suppressed (since COUNT(col) and COUNT(*) are guaranteed equivalent).
• SQLT0400 not-in-subquery-null-pitfall is now schema-aware: when the inner subquery projects a single column that resolves to a known NOT NULL column, the warning is suppressed (the NULL pitfall cannot trigger).

Fixed

• MariaDB parsing now uses MySqlDialect directly instead of a forwarding wrapper. The wrapper failed dialect_of!(MySqlDialect) downcast checks scattered through sqlparser, which silently disabled MySQL-superset features that real MariaDB grammar relies on (ON UPDATE timestamp column option, table hints, LIMIT a, b, LOCK/UNLOCK TABLES, etc.). Real-world mariadb-dump output now parses to typed AST instead of falling back to raw passthrough — verified on a 73 KB production schema where 49 raw fragments collapsed to 17 (only legitimately-tricky DELIMITER directives and stored-program bodies remain).
• MariaDB inputs are pre-processed to inject a space after bare -- at end-of-line. mariadb-dump emits -- on a line by itself; sqlparser tokenized that as two minus operators because MySqlDialect::requires_single_line_comment_whitespace is true. The preprocessor tracks string-literal / quoted-identifier / block-comment / line-comment state so the substitution never corrupts data.
• RawStatement carries a start_line so SQLT0001 (raw-passthrough) lint diagnostics report the actual source line of each fragment instead of all firing at 1:1. The field is metadata only — PartialEq for RawStatement ignores it so round-trip parse → emit → parse still produces equal ASTs.
• Raw classifier recognises delimiter and stored_program_body (CREATE TRIGGER/FUNCTION/PROCEDURE) reasons so the diagnostic message is more actionable.
• Raw classifier additionally recognises optimization_hint (ALTER TABLE … DISABLE/ENABLE KEYS, the standard mariadb-dump wrapper around per-table INSERT blocks), definer_clause (CREATE DEFINER=…), and create_event. The classifier peeks through a leading conditional comment (/*!NNN …*/, /*M!NNN …*/) before pattern-matching, so wrapped statements are recognised by their inner content.
• MariaDB/MySQL conditional comments (/*!NNN … */, /*M!NNN … */, also their unversioned forms) are unwrapped during MariaDB preprocessing so the inner SQL parses to typed AST. Marker characters are replaced with spaces of equal length so source line/column positions are preserved end-to-end.
• Per-fragment AST line numbers are now rebased to the original file's line space. The MariaDB fallback path prepends start_line - 1 newlines to each fragment before re-parsing, so every Location.line in the resulting AST refers to the actual source file (e.g. lint findings on a CREATE TABLE at file line 17981 now report 17981, not 1).
• DDL rules (SQLT0801 float-for-money, SQLT0802 varchar-without-length) now attach diagnostics to the column's identifier span (col.name.span) rather than the statement span. Statement::CreateTable has no Spanned impl upstream, so the previous span fell back to 1:1; the column-name span is always populated.

Changed

• Text-format help: lines are deduplicated by default. The same suggestion is emitted only once per (rule_id, suggestion) pair within a render — for SQLT0001 with hundreds of identical raw-passthrough findings the help shows once at the first occurrence. Output volume on a 7 MB MariaDB dump dropped from 1371 lines to 688 (~50%) with no information lost.

Added

• --help-mode auto|always|never flag on sqlt lint controlling text-format help rendering. auto (default) deduplicates per (rule_id, suggestion). always restores the per-finding rendering. never suppresses help entirely.
• --no-help flag as a shorthand for --help-mode never.
• --verbose / -v flag on sqlt lint. SQLT0001 (raw-passthrough) is now off by default — real mariadb-dump output emits hundreds of fragments sqlparser can't parse (DISABLE/ENABLE KEYS, DELIMITER directives, CREATE DEFINER prefixes), and the flood of identical warnings drowned the typed-AST rule findings. --verbose enables SQLT0001 (equivalent to --rule SQLT0001).
• --examples flag on every subcommand (parse, emit, translate, lint). Prints in-depth examples and exits 0 without parsing any input. Examples cover the basic invocation, every commonly-used flag, file-vs-stdin handling, encoding handling, and (for lint) at least one workflow per rule category. Source lives in src/cli/examples.rs.
• --list-rules flag on sqlt lint. Prints a sortable table of every registered rule with id, slug, category, default severity, default-enabled state, and one-line summary, plus a footer counting total / on / off. Exits 0 without parsing input.
• Greatly expanded long-help text on every subcommand and flag. Each flag now documents its accepted values, default, and interaction with other flags. The terse short-help (-h) is unchanged; the full long-help (--help) is the new in-depth surface.

Removed

• src/dialect/mariadb.rs (the MariaDbDialect wrapper struct). It was the source of the silent feature-flag disablement above. MariaDB-specific parser tweaks now live in src/parse/mod.rs::preprocess_mariadb.

v0.2.0

[0.2.0] - 2026-05-02

Added

• --encoding/-e flag on parse, emit, and translate for non-UTF-8 input/output. Supported values: utf-8 (default), iso-8859-1 (alias latin1), windows-1252 (alias cp1252). Decoding is strict — invalid byte sequences are rejected with exit code 1 rather than substituted with U+FFFD. JSON I/O is always UTF-8 (per spec); the flag governs SQL bytes only.
• Encoding type in src/encoding.rs wrapping encoding_rs with strict decode/encode semantics and aliases.
• CLI tests for Latin-1 round-trip (high-bit byte preservation through translate), default-mode rejection of non-UTF-8 input, and unknown-encoding exit code.
• sqlt lint subcommand with 38 active rules across 8 categories: raw passthrough, dialect cross-contamination, translation pre-flight (driven by dialect/caps.rs when --to is set), join hygiene, subquery improvements, performance pitfalls, correctness pitfalls, style/readability, and DDL hygiene. Every rule has a stable id (SQLT0500), a slug (select-star), inline summary + long-form explanation accessible via sqlt lint --explain, and per-rule fixtures under tests/fixtures/lint/.
• Four lint output formats: text (default, single-line grep-friendly), pretty (grouped per file with snippet pointer and inline rule explanation on first occurrence), json, and SARIF 2.1.0 for GitHub code-scanning integration.
• Lint CLI flags: --rule/--no-rule (repeatable, accept full id / short numeric / slug), --severity (output filter), --exit-on (exit-code threshold, default error), --explain <id> (print rule docs and exit 0).
• Lint architecture: Rule trait whose implementors register only the AST-shape callbacks they need; a shared driver in src/lint/walk.rs does one traversal per statement (sqlparser's Visitor doesn't fire on Select, so the driver does manual descent into SetExpr::Select/TableWithJoins). check_query receives a depth parameter so rules like SQLT0403 order-by-in-subquery-without-limit only fire on nested queries.
• tests/lint.rs per-rule fixture walker plus insta snapshot tests for json/sarif output.

v0.1.0

0.1.0 - 2026-05-02

Added

• Initial project scaffolding: Cargo manifest, module layout, conventions documentation.

• sqlt parse --from <dialect> [--pretty] [file|-] subcommand. Reads SQL from a file or stdin and emits a JSON envelope { sqlt_version, dialect, statements } using the upstream sqlparser AST's serde representation.

• Dialects supported by parse: mysql, postgres (aliases postgresql, pg), mssql (aliases tsql, sqlserver), sqlite, generic.

• Smoke parse fixtures under tests/fixtures/<dialect>/*.sql covering SELECT, INSERT, RETURNING, ON CONFLICT, TOP, and bracketed identifiers.

• sqlt emit --to <dialect> [file|-] subcommand. Reads a JSON envelope and emits SQL using the upstream sqlparser Display impls (per-dialect overrides land later as round-trip tests find infidelities).

• Round-trip integration suite (tests/roundtrip.rs) asserting parse → emit → parse produces an identical AST for every fixture across mysql/postgres/mssql/sqlite, plus JSON serde round-trip equivalence.

• MariaDB dialect (mariadb, alias maria) as a first-class target. MariaDbDialect wraps the upstream MySqlDialect and the parser falls back to a raw-passthrough representation (SqltStatement::Raw { sqlt_raw, reason }) for MariaDB-specific syntax with no upstream AST node — WITH SYSTEM VERSIONING, FOR SYSTEM_TIME, CREATE PACKAGE, MariaDB sequence option ordering, vector types. Same-dialect round-trip (parse → emit) preserves the original SQL verbatim for these.

• SqltStatement enum (Std(Box<Statement>) | Raw(RawStatement)) with #[serde(untagged)] so the JSON wire format for typed statements is unchanged.

• Heuristic statement splitter (parse::split) used by the MariaDB fallback path. Respects single/double quotes, backticks, and line/block comments.

• sqlt translate --from <src> --to <dst> [--strict] [file|-] subcommand. Parses the input, rewrites the AST against the target dialect's capability table, and emits SQL. Warnings are printed to stderr; --strict makes any warning a non-zero exit (code 3).

• Per-dialect capability tables (dialect/caps.rs) covering RETURNING on INSERT/UPDATE/DELETE, CREATE SEQUENCE, ON DUPLICATE KEY UPDATE, ON CONFLICT, and MariaDB raw fallback support.

• Translation rewriter (translate/rewrite.rs) that drops RETURNING when the target lacks it, warns when CREATE SEQUENCE cannot be represented, and warns when a MariaDB raw fragment passes through to a non-MariaDB target.

• WarnCode enum (RETURNING_DROPPED, SEQUENCE_DROPPED, ON_DUPLICATE_KEY_UNSUPPORTED, RAW_PASSTHROUGH) with WarnSink trait, StderrSink for the CLI, and CollectingSink for golden tests.

• Golden translation test harness (tests/translate.rs) walking tests/fixtures/translations/<src>__<dst>/<case>.{in.sql,expected.sql,expected.warn}. Covers mariadb→mysql (RETURNING dropped), mariadb→postgres (RETURNING through, system-versioning passes raw with warning), postgres→mariadb (RETURNING through cleanly).

• End-to-end CLI integration suite (tests/cli.rs) exercising the built sqlt binary: parse→emit pipe round-trip, translate warning emission, --strict exit code 3, parse error exit code 1, unknown dialect exit code 2, and multi-statement input parsing.

Changed

• The statements field of the JSON envelope is now Vec<SqltStatement> instead of Vec<Statement>. For typed statements the on-the-wire shape is unchanged thanks to #[serde(untagged)]; only raw fallback fragments introduce a new shape.