[dinwwwh]

Requested by dinwwwh · project thread

Test-only. Closes gaps that Stryker found in the downgrader. No source changes.

3.1 to 3.0, "a schema that lost a restriction" (3 new cases):

• A lost restriction carries over to the schema around it, through allOf, anyOf, properties, not and oneOf.
• propertyNames with a pattern counts as lost (as zod emits for z.record with a key pattern); propertyNames: true doesn't.
• unevaluatedProperties beside allOf counts as lost; beside additionalProperties it doesn't.

Extra expects in existing tests:

• { enum: [null, 'a'] } keeps 'a'.
• contentEncoding: 'binary' gets format: 'binary'.
• An anyOf branch with contentMediaType keeps the anyOf.
• { type: 'number', const: 5 } (Zod z.literal(5)) keeps its written type: { type: 'number', enum: [5] }.
• { examples: [] } becomes exactly {}.
• A oneOf branch with an undefined-valued keyword (if: undefined) keeps the oneOf.
• 3.2 to 3.1: a response $ref stays as written next to the summary handling.

expectValidDowngrade now also checks the output holds only what JSON keeps (toStrictEqual(JSON.parse(JSON.stringify(output)))), so a key holding undefined (e.g. a tag, info or example written as undefined) fails.

All 265 tests and pnpm run check pass.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EjMyAKtd1J94kr1PfQeQza

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will not alter performance

✅ 21 untouched benchmarks

---

_{Comparing claude/project-thread-kczaq1 (57c9f18) with main (6a906e9)}

[pullfrog]

✅ No new issues found.

Reviewed changes

Test-only PR covering gaps Stryker found in the downgrader; no source changes.

• v3.1-to-v3.0.test.ts — lost-restriction coverage: { enum: [null, 'a'] } with no type stays as written; contentEncoding: 'binary' maps to format: 'binary'; an anyOf with a non-media-type-only contentMediaType branch keeps its anyOf; and new cases assert propertyNames with a pattern counts as lost while propertyNames: true does not, unevaluatedProperties beside an in-place applicator (allOf) counts as lost while beside additionalProperties it does not, and looseness propagates through allOf/anyOf/properties/not/oneOf.
• v3.2-to-v3.1.test.ts — response $ref: a response reference stays as written alongside the summary→description handling.

Verified both files pass plus uncheck (oxlint/oxfmt/tsc). To confirm the new cases aren't theater, I mutated losesRestriction's propertyNames branch to return false — three tests failed as expected, then restored the source (tree clean). Assertions match the converter's documented contract.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

Re-review of the delta since the previous pullfrog review (d9d6459); all new content is test-only, no source changes.

• Added a typed-const case: { type: 'number', const: 5 } keeps its written type, yielding { type: 'number', enum: [5] } rather than falling back to the integer type valueType would infer.
• Added { examples: [] } → exactly {}, asserted with toStrictEqual so a surviving examples: undefined key would fail.
• Added an undefined-valued keyword case: oneOf: [{ type: 'object', if: undefined }, { type: 'string' }] keeps its oneOf, pinning the schema[key] !== undefined guard that stops if: undefined from marking the branch loose.
• expectValidDowngrade now asserts output equals JSON.parse(JSON.stringify(output)), so the helper rejects any key holding undefined and any non-JSON value.

Verified each new assertion against shared.ts and v3.1-to-v3.0.ts; the full suite passes (265 tests, 5 files). The prior review left no open threads.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}