[dinwwwh]

Requested by dinwwwh · project thread

What

Simplifies #315, following the request for the simplest approach with minimal logic.

A record keyed by an enum becomes one property per name. #315 made enumNames walk a chain of propertyNames $refs, one annotated $ref at a time with a visited set, so that it stopped at a $ref with a narrowing keyword beside it. That set was allocated for every schema, which the review of #315 measured at about 6% on the schema benchmarks.

enumNames now follows one $ref beside only annotations, as it did before #315, but with resolvePointer instead of resolve. A target that is a $ref again fails the existing hasOnly check, so it holds no names. The record then loses its key names and counts as loose, the same as any other propertyNames it cannot read.

Tests

The #315 test now expects a record keyed through an alias of an alias to lose its key names (additionalProperties stays as written), and a not over a record keyed through a narrowed $ref still becomes {}. pnpm check, pnpm test and coverage (100%) pass.

Tradeoffs

• Output. A record keyed through a chain of two or more $refs, such as a described alias of a reusable enum, no longer gets one property per name. 3.0 then allows any key there. A single $ref, as Zod and Valibot write for a reusable enum, still converts as before.
• Speed. No allocation per schema any more.

---

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 (3c7ab4b) with main (24d36dc)}

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

This reverses the visited-guarded chain walk that #315 added to enumNames, following a propertyNames $ref through at most one hop so no per-schema set is allocated.

• enumNames collapses to a single $ref hop — packages/downgrader/src/v3.1-to-v3.0.ts drops the loop for one getRef + isAnnotatedRef + resolvePointer step. A target that is a $ref again now falls through hasOnly (a $ref key is neither an annotation, enum, const, nor type) and yields undefined, so the record loses its key names and counts as loose.
• Test reworked to pin the new contract — packages/downgrader/tests/v3.1-to-v3.0.test.ts:2433 now expects a record keyed through an alias-of-an-alias (ByAlias) to lose key names while NotEnglish still collapses to {}.

I traced the behavior against the previous code: for depth ≤ 1 (a direct $ref, an inline enum/const, a $ref with a non-annotation sibling, and a $ref with an undefined-valued sibling, which hasOnly treats as missing) the new code is identical to the old; the only divergence is a two-or-more-hop chain of annotated $refs, which is the tradeoff described in the PR. enumNames is consumed by losesRestriction('propertyNames'), finishSchema, and isIntoReplaced, all keyed off the same predicate, so they stay consistent. The README already documents only "a $ref to one", so no doc drift. Ran pnpm test (357 passed, including the 130 in the touched test file).

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