Fix message-subscription races in callCommand and expect/intercept

[nathanfallet]

Summary

Two hangs that share one root cause: code acts first and only then subscribes to hear the result. Because messages flow through a replay = 0 shared flow, a message arriving in that window is dropped.

1. callCommand response race

It sent the request and only afterwards subscribed (responses.first { it.id == … }) to await the reply. A reply that arrived before the collector subscribed was lost → callCommand hangs forever (no timeout).

Fix: register an id -> CompletableDeferred<Message.Response> in a pendingRequests map (mutex-guarded) before sending; the receive loop completes the matching deferred, so the reply is captured regardless of await timing. The map is also a keyed registry — O(1) routing instead of N broadcast collectors, and a foundation for failing in-flight requests on disconnect / per-request timeouts (follow-ups).

2. expect / intercept subscribe-after-enable race

BaseRequestExpectation / BaseFetchInterception enabled the domain and only then launched the event collectors (lazily), so an event fired in that window was missed → getRequestEvent() hangs.

Fix: launch the collectors with CoroutineStart.UNDISPATCHED and before enable(), so the subscription is established before any event can fire.

Testability seam

The WebSocket is extracted behind a WebSocketTransport interface (default KtorWebSocketTransport), injected via a protected open fun createTransport() on DefaultConnection, so the message plumbing can be exercised without a real browser.

Verification (deterministic red → green)

Both tests use UnconfinedTestDispatcher + a fake transport that delivers the reply/event in the exact race window, so they fail against the old code and pass against the fix (not just guard tests):

• CallCommandResponseRaceTest — delivers the response inside send(): RED (timeout) on responses.first, GREEN with the waiter map.
• ExpectationSubscribeRaceTest — fires requestWillBeSent while handling Network.enable: RED (timeout) on enable-then-subscribe, GREEN with subscribe-before-enable.

Full :core:jvmTest (real headless Chrome — incl. RequestExpectationTest, FetchInterceptionTest, BrowserTest) + :opentelemetry:jvmTest pass. macOS native tests pass; Linux/MinGW targets compile.

Out of scope (tracked separately)

• Failing in-flight callCommand deferreds on disconnect, and closing the WS HttpClient, are pre-existing limitations (the old code hung/leaked the same way). The new registry makes the disconnect fix clean — slated for the connection-lifecycle PR.

Test plan

• ./gradlew :core:jvmTest --tests "*.CallCommandResponseRaceTest" --tests "*.ExpectationSubscribeRaceTest" — red on old code, green after
• ./gradlew :core:jvmTest (real Chrome) — all pass
• ./gradlew :opentelemetry:jvmTest — all pass
• ./gradlew :core:macosArm64Test + Linux/MinGW compile — pass

[codecov]

Codecov Report

❌ Patch coverage is 77.58621% with 13 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
.../kdriver/core/connection/KtorWebSocketTransport.kt	62.06%	3 Missing and 8 partials ⚠️
...n/dev/kdriver/core/connection/DefaultConnection.kt	90.47%	0 Missing and 2 partials ⚠️

📢 Thoughts on this report? Let us know!