Make MockHTTPFetcher registration atomic and fix drained-queue reporting

[germ-mark]

Suggestions into #32 — targets feat/testing-mocks, so the diff here is one commit. Merge or cherry-pick as you like.

Scoped to GermConvenienceMocks only. The core-library changes that were previously here are split out into #38, which targets main independently.

on() / enqueue() was not atomic

on() parked currentUrl/currentMethod on the actor and enqueue() read them back. Those are two separate actor messages, so a second chain could overwrite the cursor in between — responses land under the wrong URL, or a URL ends up with an empty queue and throws unmockedRequest later, far from the cause.

It is not theoretical. Measured against the current implementation:

concurrency	misattribution
2 chains, idle machine	~10 / 2000
2 chains, loaded machine	482 / 2000
8 chains	~970 / 1000

Load-dependent flakiness inside a testing library is the worst kind, because it gets blamed on the code under test.

on() now returns a Registration value carrying the url and method, and enqueue() hangs off that. The cursor is gone, so the race is unrepresentable rather than merely unlikely — and Errors.missingOnUrl becomes impossible and is removed. Call sites are unchanged apart from dropping try, since enqueue no longer throws:

let mock = await MockHTTPFetcher()
    .on(tokenUrl, method: .post).enqueue(.success(postResponse))
    .on(tokenUrl).enqueue(.success(anyResponse))

A drained exact queue reported "never mocked"

Because the exact-to-any fallback switches keys as soon as an exact queue empties, requestCount > handlerCount was structurally unreachable for an exact key. So over-requesting an exact-only mock threw unmockedRequest — indistinguishable from a URL you never registered — and tooManyRequests only ever fired for .any handlers. The more specific registration produced the vaguer error.

Registered-but-drained now reports tooManyRequests for either key. The exact-to-any fallback itself is unchanged — exhaustedExactHandlerFallsBackToAny shows it is deliberate, so this only covers the case where there is no .any handler to fall back to.

Smaller

• on(origin) never matched. Lookup keys off request.request.url, which HTTPRequest has round-tripped through its pseudo-header fields — that rewrites a bare origin's empty path to /. Registration now goes through the same round trip, so on(URL("https://as.example")) matches a request to that origin instead of keying a url no request can produce.
• Requests that matched no handler are now logged, matching how tooManyRequests rejections already were, so requests(for:) is a complete record of what the code under test sent. Shout if you would rather the log only contain matched requests — the asymmetry was the actual problem, either direction fixes it.
• The two implicit returns from your review comments.
• The changeset said GermConvenienceTesting after the target was renamed to GermConvenienceMocks; that would have shipped into the CHANGELOG.
• Documented: concurrent same-url enqueue order is undefined, and requests(for:) includes rejected requests while allRequested counts only those that consumed a response.

Test notes

19 tests, clean build, no warnings. All 8 of your existing tests pass unmodified except for dropping the redundant try. Added: the drained-exact case, unmocked-request logging, bare-origin registration, over-request-then-enqueue, and a 16-task concurrent configuration test that ran 20/20 clean.

One coordination note: #38 moves extension BundledHTTPRequest: Equatable onto the type itself. Whichever lands second, this PR's copy in MockHTTPFetcher.swift should be deleted — it is marked with a comment.

🤖 Generated with Claude Code

[changeset-bot]

⚠️ No Changeset found

Latest commit: 6672ac8

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes changesets to release 1 package

Name	Type
@germ-network/germ-convenience	Patch

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR