[dinwwwh]

Requested by dinwwwh · project thread

Summary

3.2 sends a parameter value as written, without percent-encoding, for style: "cookie" and for allowReserved: true in a path or cookie. The downgrade already removes both (3.1 has no cookie style and applies allowReserved only to query parameters), so 3.1 falls back to form or simple, which percent-decode the value. A value such as a%3D that 3.2 hands to the schema as a%3D then reaches a 3.1 schema as a=, so a schema like pattern: '^[^=]*$' rejects a value the 3.2 document accepts.

The fix: for such a parameter (with schema, not content), the schema becomes {}. Headers are left alone, since neither version percent-encodes them. The README gets one table row.

Tradeoffs

• Code generators lose the parameter's type for these parameters and treat the value as untyped. This only hits 3.2 documents using style: "cookie" or allowReserved outside query, which oRPC does not write, so oRPC output is unchanged.

Test plan

• New test: cookie with style: "cookie", cookie and path with allowReserved, and a header with allowReserved that keeps its schema.
• vitest run packages/downgrader (302 passed) and pnpm check pass locally.

---

Generated by Claude Code

[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

1 warning

⚠️ 1 benchmark is dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

✅ 29 untouched benchmarks
⏩ 2 skipped benchmarks¹

---

_{Comparing claude/project-thread-h3q238 (261d141) with main (bcb4125)}

Footnotes

1. 2 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[pullfrog]

✅ No new issues found.

Reviewed changes

• finishParameter widens the schema for parameters 3.2 writes as written (packages/downgrader/src/v3.2-to-v3.1.ts:246): a parameter beside schema (not content) whose style is "cookie", or whose allowReserved is true with in: "path"/"cookie", now converts to schema: {}. 3.1 drops that style and falls back to form/simple, which percent-decode, so the original constraint could reject a value 3.2 accepts. Headers are correctly left alone.
• Test coverage (packages/downgrader/tests/v3.2-to-v3.1.test.ts:233) pins the cookie-style, cookie/path-allowReserved, and header-keeps-schema cases.
• README gains one row documenting the schema: {} rule and the re-decoding it compensates for.

This is a sound, conservative fix — it only widens the schema, consistent with the downgrader's "loses detail, never meaning" contract. I checked the trigger set against the version docs (v3.2.ts:540-554 vs v3.1.ts:590-591, and the cookie-style note at v3.2.ts:435-437): the cases where 3.2 writes as-written but 3.1 re-decodes are exactly {style: "cookie"} ∪ {allowReserved on path/cookie}, and query/headermatch across versions, so no case is missed. Verifiedvitest run packages/downgrader(302 passed) anduncheck` on the three files, all green.

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