Repository navigation
build(deps)!: require Effect 4.0.0 and Vitest 5 - #22
Merged
Merged
Conversation
Effect moves from 4.0.0-rc.115 to the stable 4.0.0 release, with @effect/platform-node, @effect/platform-bun and @effect/vitest in lockstep. Migrate off the `effect/unstable/*` export paths that 4.0.0 removed (32 imports across 27 files): unstable/http -> effect/http, unstable/httpapi -> effect/http-api, unstable/process -> effect/process and unstable/cli -> effect/cli. Docs, READMEs and the "Effect v4 RC required" notices drop the now-stale RC wording. Vitest moves 4.1.11 -> 5.0.4 with @vitest/coverage-v8 5.0.4 (its peer matches the vitest version exactly) and @cloudflare/vitest-plugin 1.1.13 -> 1.4.0, the first plugin release supporting Vitest 5. That finally satisfies @effect/vitest's `vitest >=5` peer. Vitest 5 enables `fsModuleCache` by default, and plugin 1.4.0 cannot read those host-FS cache entries from workerd, so every workerd test file failed before running. Set `fsModuleCache: false` in the four workerd configs until cloudflare/workers-sdk#16163 ships; the removal follow-up is recorded in CONTRIBUTING.md. BREAKING CHANGE: consumers must migrate to Effect 4.0.0.
@mpsuesser/oxlint-plugin-effect moves 0.4.3 -> 0.6.0 and oxlint 1.83.0 -> 1.87.0. 0.6.0 is the first plugin release built against stable effect@4.0.0, which is the stack the previous commit moved to. Its fixes include detecting Config.String/Config.NonEmptyString secrets and no longer demanding schema type aliases for schema checks, filters, equivalences and formatters. 0.5.0 had also enabled 16 generic "anti-slop" rules ported from dmmulroy/anti-slop, taking both presets from 57 to 73 rules. Those are not adopted here: on this codebase they produce 1410 findings across 139 files, 73% of them require-safety-comment-for-type-assertion. They are disabled in oxlint.config.ts with the rationale, so this upgrade keeps the current lint policy while picking up the 0.6.0 fixes. The 57 pre-existing rules pass clean on 0.6.0 with no changes. deno.lock is regenerated by the deno smoke run.
alchemy moves 2.0.0-beta.79 -> 2.0.0-beta.81 in both cf-workers-alchemy examples, in the catalog, and in the release-age exclusions. beta.80 is the first alchemy line to declare stable-only Effect peer ranges: effect, @effect/vitest, @effect/platform-bun, @effect/platform-node and @effect/sql-* all become ^4.0.0. The rc-era copies of @effect/sql-d1, @effect/sql-sqlite-do and @effect/vitest stop resolving, and nothing resolves 4.0.0-rc.115 any more. beta.81 rather than the latest (beta.83) because beta.82/83 resolve @distilled.cloud/*@1.0.0-rc.15, published 2026-10-12T19:2x -- inside the 24h window the repo's supply-chain policy enforces. Lockfile verification rejects it, and the release-age exclusions do not exempt verification even though pnpm writes them itself, so a committed lockfile containing rc.15 fails 'pnpm install --frozen-lockfile' -- the command CI runs. beta.81 resolves @distilled.cloud/*@1.0.0-rc.13, published 2026-10-02. 'pnpm install --frozen-lockfile' passes and the examples typecheck 12/12. The next sweep can go straight to alchemy's latest once rc.15 ages out.
`runSubcommand` only turned the aggregated per-runtime result into an exit
code inside `if (spec.forceExit)`, and only `integration` sets that flag:
it exists to work around the vitest-plugin/miniflare cleanup hang, not to
signal failure. Every other subcommand recorded `{ ok: false }` into
`results`, printed "x FAIL", and then returned normally, so
`NodeRuntime.runMain` exited 0.
The effect was that `pnpm test:unit` and `pnpm test:smoke` could not fail a
build. A deliberately failing unit test produced "1 failed | 32 passed
(33)" and exit code 0; a smoke run with a broken runtime reported
"4/5 passed" and exit code 0. `pnpm verify` and CI both include those
steps, so those gates were decorative.
Propagate failures for every subcommand, keeping `forceExit` as the hang
workaround. Verified: a failing unit run now exits 1 while a passing run
still exits 0, and smoke exits 1 when a runtime fails.
The catalog pinned effect, @effect/platform-node, @effect/platform-bun and @effect/vitest to exactly 4.0.0. Because 'catalog:' feeds both the devDependency and the peerDependency, every package published an exact 'effect: "4.0.0"' peer -- an RC-era habit, from when the 4.0.0-rc.* line was still moving. It forces consumers onto a single patch release. Widen the catalog to ^4.0.0 so consumers can take any 4.x patch. The lockfile resolves the family to 4.0.3, the newest release older than the 24h window the repo's supply-chain policy enforces. 4.0.4 exists but was published ~5h ago, and resolving it would make 'pnpm install --frozen-lockfile' -- the command CI runs -- fail with ERR_PNPM_MINIMUM_RELEASE_AGE_VIOLATION until it ages out. 4.0.3 also satisfies the @effect/sql-* packages alchemy resolves, which want ^4.0.3, clearing the last effect-related peer warning. The published peer stays "4.0.0 or above"; only the tested version moves. A later in-range update can float the family to 4.0.4 once it clears the policy window.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Effect 4.0.0 (stable) + Vitest 5, plus tooling and a CI-gate fix
Effect
effect,@effect/platform-node,@effect/platform-bunand@effect/vitestmove to the stable 4.0.0 line.effect/unstable/*export paths that 4.0.0 removed:http,http-api,process,cli.Peer range (consumer-visible)
4.0.0; becausecatalog:feeds both the devDependency and the peerDependency, every package publishedeffect: "4.0.0"exactly. Widened to^4.0.0so consumers can take any 4.x patch.pnpm install --frozen-lockfilefail withERR_PNPM_MINIMUM_RELEASE_AGE_VIOLATION.Vitest 5
vitest4.1.11 → 5.0.4,@vitest/coverage-v8→ 5.0.4,@cloudflare/vitest-plugin1.1.13 → 1.4.0 (first plugin release supporting Vitest 5). This finally satisfies@effect/vitest'svitest >=5peer.fsModuleCache: falsein the four workerd configs — Vitest 5 enables the filesystem module cache by default and plugin 1.4.0 cannot read those host-FS entries from workerd, so every workerd test file failed before running. Revert once [vitest-plugin] Fix filesystem module caching cloudflare/workers-sdk#16163 ships (tracked in CONTRIBUTING.md).fix(test-runner) — worth reviewing closely
runSubcommandonly converted the aggregated per-runtime result into an exit code insideif (spec.forceExit), and onlyintegrationsets that flag. Every other subcommand recorded{ ok: false }, printedx FAIL, then returned normally, soNodeRuntime.runMainexited 0. A deliberately failing unit test produced1 failed | 32 passed (33)and exit code 0; a smoke run with a broken runtime reported4/5 passedand exit 0. CI'spnpm test:unitstep was one of them. Failures now propagate for every subcommand.Tooling
2.0.0-beta.81(beta.80+ is the first line with stable-only^4.0.0Effect peer ranges; beta.82/83 resolve@distilled.cloud/*@1.0.0-rc.15, published <24h ago, which the release-age policy rejects during lockfile verification).@mpsuesser/oxlint-plugin-effect→ 0.6.0 (first release built against stableeffect@4.0.0) and oxlint → 1.87.0. The 16 generic anti-slop rules 0.5.0 enabled are deliberately not adopted — they produce 1410 findings across 139 files; they are disabled inoxlint.config.tswith the rationale.Verification
pnpm install --frozen-lockfile— passes (this was failing on an earlier draft of the alchemy bump)pnpm verify— green: CDP codegen freshness x3, check (lint/fmt/fallow/tsgo/package-versions), build, unit 444, smoke 5/5 (node, workerd, workerd-nocompat, bun, deno), examples 12/12, docs 90 blocks / 653 linksNote on the changelog
The peer-range widening is a root-only
build(deps)commit; release-please attributes entries by package path, so it may not appear. Flagging it here since it changes the published peer contract.