Parse the any deparsed or pretty-formatted SQL back and verify the AST matches the original. This ensures the formatting doesn't break the SQL semantics.
Please review the test utilities — note that exact SQL string equality is not required. The focus is on comparing the resulting ASTs.
Refer to expectAstMatch to understand how correctness is validated.
The pipeline is: parse(sql1) → ast → deparse(ast) → sql2 While sql2 !== sql1 (textually), a correct round-trip means: parse(sql1) === parse(sql2) (AST-level equality).
You can see expectAstMatch here: packages/deparser/test-utils/index.ts
Our systematic approach to fixing deparser issues:
- One test at a time: Focus on individual failing tests using
yarn test --testNamePattern="specific-test" - Always check for regressions: After each fix, run full
yarn testto ensure no previously passing tests broke - Track progress: Update this file with current pass/fail counts after each significant change
- Build before testing: Always run
yarn buildafter code changes before testing - Clean commits: Stage files explicitly with
git add <file>, never usegit add . - Tight feedback loops: Use isolated debug scripts for complex issues, but don't commit them
Workflow: Make changes → yarn test --testNamePattern="target-test" → yarn test (check regressions) → Update this file → Commit & push
When committing to TESTS.md, always run all tests — do not use testNamePattern, only yarn test