[vitest-pool-workers] `inject()` in pool options no longer infers types or checks keys

[petebacondarwin]

Split out of #15123 so that PR can land. Raised by Devin: #15123 (comment)

Problem

WorkerPoolOptionsContext.inject was changed from typeof inject (vitest's) to a widened signature in packages/vitest-pool-workers/src/pool/plugin.ts:

inject: <T = unknown>(key: string) => T;

The change was deliberate, and the reason is documented in place: binding to typeof inject from this package's copy of vitest makes keyof ProvidedContext resolve to never, because a consumer's declare module "vitest" augmentation only applies to the consumer's own copy, and pnpm can resolve two separate virtual-store instances. So the previous typing was broken in its own way.

The cost is that inference and key checking are both gone:

• The return type now defaults to unknown, so anywhere the value is passed to a typed parameter or assigned to a typed variable, consumers need an explicit type argument. Our own fixture had to become inject<number>("echoServerPort") in fixtures/vitest-pool-workers-examples/hyperdrive/vitest.config.ts.
• key is string, so a misspelled key is no longer a type error.

There is also a semver question: this is a change to a public type that can require consumers to edit their config, which is not obviously a patch.

Possible direction

Keep an overload that preserves inference for known keys while still accepting arbitrary strings, e.g.:

inject: {
  <K extends keyof ProvidedContext>(key: K): ProvidedContext[K];
  <T = unknown>(key: string): T;
};

This needs checking against the actual failure it was working around — if keyof ProvidedContext resolves to never in the cross-copy case, the first overload may simply never match and the fallback would still apply, but that wants verifying with a real pnpm-resolved consumer rather than assumed.

Acceptance

• Consumers using a ProvidedContext augmentation get inference and key checking back, without reintroducing the never resolution problem.
• If the public type ends up changing in a way consumers must react to, the changeset is classified accordingly.

[workers-devprod]

Note

This is an automated, advisory triage report generated by workers-devprod. It is not an official maintainer response — a maintainer will follow up.

🤖 Automated triage report

Issue Triage: #15137

Summary

Tracking item to restore type inference and key checking for WorkerPoolOptionsContext.inject in vitest-pool-workers, which was widened to <T = unknown>(key: string) => T to work around a cross-copy ProvidedContext augmentation problem.

Classification

• Type: Bug / Tech-debt tracking item
• Component: packages/vitest-pool-workers
• Severity: Low
• Has Reproduction: N/A (internal follow-up with a documented cause and pointer to the originating change)
• Quality: Complete

Findings

• Created: not available (metadata truncated in context.json)
• Author: petebacondarwin (Pete Bacon Darwin — maintainer)
• Version: not specified
• Labels: not available (metadata truncated in context.json)
• Comments: not available (metadata truncated in context.json)

Analysis

• Filed by a maintainer as a deliberate split-out of PR Bump the workerd-and-workers-types group across 1 directory with 2 updates #15123 (raised by Devin in review) so that PR could land — this is a follow-up tracking item, not an end-user bug report.
• The affected code is explicitly identified: WorkerPoolOptionsContext.inject in packages/vitest-pool-workers/src/pool/plugin.ts. The component is unambiguously vitest-pool-workers.
• The regression is a developer-experience / typing loss: inject() now returns unknown (requires explicit type args, e.g. inject<number>("echoServerPort")) and no longer type-checks the key string. There is no runtime impact, hence Low severity.
• The issue includes a concrete proposed direction (an overload preserving inference for known ProvidedContext keys while accepting arbitrary strings) and flags an open semver question (the public type change may not be a patch), which needs verification against the original pnpm cross-copy failure.
• Not closeable under any triage category: it is a valid, well-specified, maintainer-owned item with a clear next step. It has sufficient information, so it does not need more info.
• Note: the pre-fetched context.json is a single-line JSON whose body exceeds the tool's 2000-char read limit, so trailing metadata (labels, dates, comments, state) could not be read. Triage is based on the fully-readable issue body and author.

Recommendation

Status: KEEP OPEN

Reasoning: This is a valid, clearly-scoped technical tracking item filed by a maintainer with a documented root cause and a proposed solution. It does not match any closure category and has enough detail to act on.

Action: Keep open as a tracking item for the vitest-pool-workers team. Verify whether the proposed overload (<K extends keyof ProvidedContext>(key: K): ProvidedContext[K] with an <T = unknown>(key: string): T fallback) actually restores inference under the real pnpm cross-copy resolution that the widening was working around, and resolve the semver question (likely a minor if it changes the public type, or confirm the original widening's release channel).

Suggested Labels: (none — repository label set could not be confirmed from the truncated context.json; omitting rather than guessing)

Suggested Comment

(Omitted — this is an internal maintainer-filed tracking item that should stay open; no reply to a reporter is needed.)

Structured summary (JSON)

{
  "issueNumber": 15137,
  "title": "Restore type inference and key checking for vitest-pool-workers inject()",
  "githubUrl": "https://github.com/cloudflare/workers-sdk/issues/15137",
  "recommendation": "KEEP OPEN",
  "difficulty": "medium",
  "reasoning": "Valid, well-scoped tech-debt tracking item filed by a maintainer with a documented root cause and a proposed overload-based fix. It does not match any closure category and has sufficient detail to act on.",
  "suggestedAction": "Keep open for the vitest-pool-workers team. Verify the proposed inject() overload restores inference under the real pnpm cross-copy resolution, and resolve the semver question for the public type change.",
  "hasSuggestedComment": false,
  "suggestedLabels": []
}

[workers-devprod]

• Reproduction: ✅ Successfully reproduced. I checked out the branch containing the widened inject: <T = unknown>(key: string) => T signature (commit 1de4bd5a4) and ran a minimal TypeScript reproduction against it. tsc reports TS18046: 'port' is of type 'unknown' when inject("echoServerPort") is called without an explicit type argument, and a misspelled key like inject("echoServerPrt") is silently accepted instead of being rejected. By contrast, using the original typeof inject signature correctly infers number for the same key and rejects the misspelled key with TS2345. The issue is also visible in the repo's own hyperdrive fixture, which had to be changed to inject<number>("echoServerPort") on that branch.