feat(effect-bun-test): carry bun's each, re-export Mock, refuse an Effect from the bare it — release 0.5.0

[jfspencer]

What

Three changes to @packages/effect-bun-test, each found and measured against a real consumer: a 615-file uniform-import sweep in the ultravisor repo, moving every test file off bun:test and onto this harness.

1. it.each was missing

The harness API object carried skip, only, skipIf, runIf, fails and for — but no each. Any consumer moving off bun:test silently lost every table-driven test.

The failure mode is the dangerous kind. bun raises the missing registrar as an "unhandled error between tests", which it does not count as a failure. Measured on the sweep: 106 tests vanished while one package still printed 0 fail. Exit codes and failure names both said the change was clean.

It is now delegated to bun's own each rather than reimplemented, so %s/%p name formatting, array-case spreading and the .skip/.only/.todo/.failing chain are bun's by construction and cannot drift. The delegation must be bound — an unbound test.each reference throws Expected this to be instanceof ScopeFunctions.

2. export type { Mock }

The values were re-exported; the type was not. That stranded any file typing a mock() result.

3. The bare it now refuses an Effect

TestCollectorCallable typed the callback's return as unknown. An Effect is not thenable, so one handed to the bare it is never run: the case passes with every assertion skipped and the test count does not move.

bun's own types reject this (void | Promise<unknown>). This harness's widening accepted it — so adopting the harness silently deleted a compile-time guard. It now matches bun.

Two facts worth keeping, both measured, both easy to undo by accident:

• The union is the guard. Splitting void | Promise<unknown> into separate void and Promise<unknown> overloads re-opens the hole, because a bare void return type triggers TypeScript's void-assignability rule and accepts a callback returning anything. Only inside a union does that rule not apply. I shipped the overload form first and measured it failing to catch the probe.
• Biome's noConfusingVoidType fires on exactly that union, so it carries a suppression. That suppression is the reason the guard exists — it is not tidy-up-able.

Verification

• __tests__/bun-registrars.test.ts — red-on-repro proven: 0 pass / 1 fail / 1 error against the pre-change harness, 8 pass after.
• __tests__/bare-it-rejects-effect.types.test.ts — a @ts-expect-error fixture, so a regression fails bun run tsc rather than going quiet.
• bun run dod: 114 tests in effect-bun-test, 0 tsc errors, 0 biome findings.
• Against the consumer: with this build installed, all 615 converted files typecheck at 0 errors (was 19, all Property 'each' does not exist on type 'Methods<never>'), and lose no test and no executed expect() call.

Also in this PR

packages/ddd-meta/__tests__/write-totality.test.ts carried two missedPipeableOpportunity errors at error severity. Those fail bun run lint:effect and therefore bun run dod — the exact command the release workflow gates on — so the release could not run while they were red. Converted to the pipe form the linter prescribes. Unrelated to the harness change.

One thing I could not verify locally

bun run dod is red on my machine in packages/ddd-meta and packages/ddd-path, with three failures that all pin bun 1.3.14's TOML parser message ("Unexpected =") against bun 1.4.0's ("TOML Parse error: Expected a value but found '='"). This repo pins bun@1.3.14 in packageManager and both workflows install 1.3.14, so CI exercises the pinned parser and this local red should not reproduce here. Everything bun-version-independent passes locally: tsc, lint:effect, check:ci, and all 114 effect-bun-test tests.

Please confirm CI is green before tagging v0.5.0 — the release workflow runs bun run dod, so a red tag produces no release.

Release

All seven manifests are at 0.5.0 and bun scripts/assert-tag-version.ts v0.5.0 passes. Tagging v0.5.0 on this commit once merged will trigger the release workflow.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BoLJY2Y4UejKpsN8ogVaip