tesserakit-sql 0.4.0

SQL job pack for Tessera: lint SQL files/migrations into a statement and table catalog.

tesserakit-sql

Lint SQL files and migrations into a statement and table catalog.

tessera-sql parses .sql files with lightweight heuristics (no database connection, no execution), builds a catalog of statements and declared tables, and flags high-signal migration-safety issues.

Lint SQL

tessera sql lint --input migrations/ --output ./out/sql_pack
tessera sql lint --input schema.sql --output ./out/sql_pack

Artifacts written:

statements.jsonl         one SqlStatement per parsed statement (kind, target, flags)
tables.jsonl             one SqlTable per CREATE TABLE (columns, primary-key flag)
index.md                 statement catalog
validation_report.md     safety findings
coverage_report.md       statement-kind distribution
tables.md                table catalog with columns and PK status

Lint rules

Query safety:

• delete_without_where (error) — DELETE with no WHERE removes every row
• update_without_where (warning) — UPDATE with no WHERE writes every row
• select_star (info) — SELECT * couples the query to column shape

Migration safety (the costly, easy-to-miss class):

• add_not_null_without_default (error) — ALTER TABLE ... ADD COLUMN ... NOT NULL with no DEFAULT rewrites the table and fails on existing rows
• truncate_table (warning) — TRUNCATE wipes all rows and is often non-transactional / irreversible
• drop_column (warning) — dropping a column is destructive and irreversible
• rename_breaks_compatibility (warning) — RENAME breaks code referencing the old name; prefer add-new + backfill + drop-old
• drop_without_if_exists (warning) — DROP without IF EXISTS fails if the object is absent
• create_table_without_if_not_exists (info) — non-idempotent if the migration re-runs

Schema:

• table_without_primary_key (warning) — a CREATE TABLE declares no PRIMARY KEY
• no_statements — nothing parsed

Limitations (v0.1)

Parsing is heuristic: comments are stripped, statements are split on top-level semicolons (quote-aware), and classification is keyword/regex based. It is tuned for migration and schema files, not for validating arbitrary vendor SQL dialects.