sql-migration-linter（SQL 迁移 linter）

SQL Migration Linter

Rule-based linter for SQL migration files. Catches mistakes that make migrations non-idempotent, destructive, or unsafe under concurrent load. Pure Python stdlib — no dependencies.

Supports dialects: generic, postgres, mysql, sqlite.

Commands

# Lint a single file
python3 scripts/sql_migration_linter.py lint migrations/001_init.sql

# Lint a directory recursively
python3 scripts/sql_migration_linter.py lint migrations/

# Specify dialect (unlocks Postgres-specific rules)
python3 scripts/sql_migration_linter.py lint migrations/ --dialect postgres

# Filter by minimum severity
python3 scripts/sql_migration_linter.py lint migrations/ --min-severity warning

# JSON output for CI
python3 scripts/sql_migration_linter.py lint migrations/ --format json

# Compact summary
python3 scripts/sql_migration_linter.py lint migrations/ --format summary

# List all rules
python3 scripts/sql_migration_linter.py rules

Rules (17 total)

Structure

• missing-trailing-semicolon (error) — file does not end with ;
• mixed-indentation (warning) — tabs and spaces mixed in the same line
• trailing-whitespace (info)
• keyword-case-inconsistent (info) — same keyword appears in mixed case

DDL safety

• drop-without-if-exists (warning) — DROP TABLE/INDEX/... without IF EXISTS
• destructive-drop-table (warning) — DROP TABLE flagged for review
• create-without-if-not-exists (warning) — CREATE TABLE/INDEX/... without IF NOT EXISTS
• create-index-locks-table (warning, postgres) — CREATE INDEX without CONCURRENTLY
• add-column-not-null-no-default (error, postgres) — ADD COLUMN ... NOT NULL without DEFAULT
• reserved-word-identifier (warning) — identifier matches a SQL reserved word (e.g. user, order)

DML safety

• update-without-where (error)
• delete-without-where (error)
• truncate-is-destructive (warning)
• select-star (info) — SELECT * in migrations
• insert-without-conflict-handling (info) — INSERT without ON CONFLICT / ON DUPLICATE KEY

Transactions

• missing-transaction (warning) — 2+ DDL statements without explicit BEGIN/COMMIT
• begin-without-commit (error)

Output formats

• text (default) — grouped by file, line:severity: [rule] message, with totals
• json — array of {file, line, rule, severity, message} objects
• summary — counts per severity + top 10 rules by frequency

Exit codes (CI-friendly)

• 0 — clean (or only info below min-severity)
• 1 — warnings present, no errors
• 2 — errors present

Examples

# Pre-commit hook — fail on any warning or error
python3 scripts/sql_migration_linter.py lint migrations/ --min-severity warning

# CI gate — fail only on errors
python3 scripts/sql_migration_linter.py lint migrations/ --min-severity error

# Postgres-specific audit
python3 scripts/sql_migration_linter.py lint migrations/ --dialect postgres --format json > report.json

Why this exists

Migrations that look fine locally fail in production because:

• They aren't idempotent (re-run fails)
• They lock large tables (Postgres CREATE INDEX, ADD COLUMN NOT NULL)
• They mutate every row (UPDATE / DELETE without WHERE)
• They use reserved words as identifiers and break under different parsers

This linter catches those before the PR gets merged.

Limitations

• Uses regex + statement splitting; not a full SQL parser
• No schema knowledge — cannot check FK targets, column types, etc.
• keyword-case-inconsistent is per-statement, not repo-wide