improve(tests): narrow shared runner resets (#118455)

* test(agents): reuse shared runner module graph

* test(agents): narrow shared runner resets

* test(agents): keep broad reset private

---------

Co-authored-by: Peter Steinberger <steipete@mac-studio-sf2.local>
This commit is contained in:
Peter Steinberger
2026-08-02 22:07:48 -07:00
committed by GitHub
parent 708c0acf16
commit 578c190579
13 changed files with 106 additions and 28 deletions
@@ -11,7 +11,7 @@ import {
mockedRunEmbeddedAttempt,
mockedSleepWithAbort,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
useOpenAIPlatformAuthFixture,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
@@ -78,7 +78,7 @@ describe("runEmbeddedAgent before_agent_finalize", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
useOpenAIPlatformAuthFixture();
mockedGlobalHookRunner.hasHooks.mockImplementation(
(hookName: string) => hookName === "before_agent_finalize",
@@ -6,7 +6,7 @@ import {
mockedGlobalHookRunner,
mockedRunEmbeddedAttempt,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
@@ -54,7 +54,7 @@ describe("runEmbeddedAgent before_agent_reply seam", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
});
it("lets before_agent_reply claim cron runs before the embedded attempt starts", async () => {
@@ -9,7 +9,7 @@ import {
mockedMarkAuthProfileFailure,
mockedRunEmbeddedAttempt,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
import { hasCodexAppServerRecoveryRetryBudget } from "./run/codex-app-server-recovery.js";
@@ -98,7 +98,7 @@ describe("runEmbeddedAgent Codex app-server recovery", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
mockedClassifyFailoverReason.mockReturnValue(null);
});
@@ -11,7 +11,7 @@ import {
mockedIsFailoverAssistantError,
mockedRunEmbeddedAttempt,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
useOpenAIPlatformAuthFixture,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
@@ -24,7 +24,7 @@ describe("runEmbeddedAgent Codex server_error fallback handoff", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
useOpenAIPlatformAuthFixture();
mockedGlobalHookRunner.hasHooks.mockImplementation(() => false);
});
@@ -25,7 +25,7 @@ import {
mockedIsCompactionFailureError,
mockedIsLikelyContextOverflowError,
mockedRunEmbeddedAttempt,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
@@ -111,7 +111,7 @@ describe("post-compaction loop guard wired into runEmbeddedAgent", () => {
beforeEach(() => {
liveToolCallSeq = 0;
diagnosticSessionStates.clear();
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
mockedIsCompactionFailureError.mockImplementation((msg?: string) => {
if (!msg) {
return false;
@@ -15,7 +15,7 @@ import {
mockedRunEmbeddedAttempt,
mockedResolveAuthProfileOrder,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
warmRunOverflowCompactionHarness,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
@@ -174,7 +174,7 @@ describe("runEmbeddedAgent cross-provider fallback error handling", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
useCrossProviderAuthFixture();
mockedGlobalHookRunner.hasHooks.mockImplementation(() => false);
});
@@ -10,7 +10,7 @@ import {
mockedGlobalHookRunner,
mockedRunEmbeddedAttempt,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
import { buildEmbeddedRunPayloads as realBuildEmbeddedRunPayloads } from "./run/payloads.js";
@@ -67,7 +67,7 @@ describe("runEmbeddedAgent silent-error retry", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
mockedGlobalHookRunner.hasHooks.mockImplementation(() => false);
mockedClassifyFailoverReason.mockReturnValue(null);
});
@@ -11,7 +11,7 @@ import {
mockedGlobalHookRunner,
mockedRunEmbeddedAttempt,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
import type { EmbeddedRunAttemptResult } from "./run/types.js";
@@ -57,7 +57,7 @@ describe("runEmbeddedAgent fast auto progress", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
mockedGlobalHookRunner.hasHooks.mockImplementation(() => false);
mockedClassifyFailoverReason.mockReturnValue(null);
});
@@ -447,10 +447,7 @@ export const overflowBaseRunParams = {
runId: "run-1",
} as const;
/** Reset every mocked runner dependency to the default successful no-op state. */
export function resetRunOverflowCompactionHarnessMocks(): void {
vi.unstubAllEnvs();
resetCommandQueueStateForTest();
function resetMockAgentHarness(): void {
clearAgentHarnesses();
registerAgentHarness({
id: "codex",
@@ -472,6 +469,13 @@ export function resetRunOverflowCompactionHarnessMocks(): void {
return { assistant, ...(result.attemptUsage ? { usage: result.attemptUsage } : {}) };
},
});
}
/** Reset every mocked runner dependency to the default successful no-op state. */
function resetRunOverflowCompactionHarnessMocks(): void {
vi.unstubAllEnvs();
resetCommandQueueStateForTest();
resetMockAgentHarness();
mockedGlobalHookRunner.hasHooks.mockReset();
mockedGlobalHookRunner.hasHooks.mockReturnValue(false);
mockedGlobalHookRunner.runBeforeAgentReply.mockReset();
@@ -663,6 +667,80 @@ export function resetRunOverflowCompactionHarnessMocks(): void {
mockedSleepWithAbort.mockResolvedValue(undefined);
}
/** Reset only the seams mutated by the shared public-entry integration suites. */
export function resetSharedRunIntegrationHarnessMocks(): void {
vi.unstubAllEnvs();
resetCommandQueueStateForTest();
resetMockAgentHarness();
mockedBuildEmbeddedRunPayloads.mockReset();
mockedBuildEmbeddedRunPayloads.mockReturnValue([]);
mockedClassifyFailoverReason.mockReset();
mockedClassifyFailoverReason.mockReturnValue(null);
mockedClassifyAssistantFailoverReason.mockReset();
mockedClassifyAssistantFailoverReason.mockImplementation(
(assistant?: { errorMessage?: string | null }): FailoverReason | null =>
mockedClassifyFailoverReason(assistant?.errorMessage ?? ""),
);
mockedCompactDirect.mockReset();
mockedCompactDirect.mockResolvedValue({
ok: false,
compacted: false,
reason: "nothing to compact",
});
mockedEnsureAuthProfileStore.mockReset();
mockedEnsureAuthProfileStore.mockReturnValue({ version: 1, profiles: {} });
mockedEnsureAuthProfileStoreWithoutExternalProfiles.mockReset();
mockedEnsureAuthProfileStoreWithoutExternalProfiles.mockReturnValue({
version: 1,
profiles: {},
});
mockedFormatAssistantErrorText.mockReset();
mockedFormatAssistantErrorText.mockReturnValue("");
mockedGetApiKeyForModel.mockReset();
mockedGetApiKeyForModel.mockImplementation(
async ({ profileId }: MockGetApiKeyForModelParams = {}) => ({
apiKey: "test-key",
profileId: profileId ?? "test-profile",
source: "test",
mode: "api-key",
}),
);
mockedGlobalHookRunner.hasHooks.mockReset();
mockedGlobalHookRunner.hasHooks.mockReturnValue(false);
mockedGlobalHookRunner.runBeforeAgentReply.mockReset();
mockedGlobalHookRunner.runBeforeAgentReply.mockResolvedValue(undefined);
mockedIsCompactionFailureError.mockReset();
mockedIsCompactionFailureError.mockReturnValue(false);
mockedIsFailoverAssistantError.mockReset();
mockedIsFailoverAssistantError.mockReturnValue(false);
mockedIsLikelyContextOverflowError.mockReset();
mockedIsLikelyContextOverflowError.mockImplementation((msg?: string) => {
const lower = normalizeLowercaseStringOrEmpty(msg ?? "");
return (
lower.includes("request_too_large") ||
lower.includes("context window exceeded") ||
(lower.includes("context window") && lower.includes("ran out of room")) ||
lower.includes("prompt is too long")
);
});
mockedIsRateLimitAssistantError.mockReset();
mockedIsRateLimitAssistantError.mockReturnValue(false);
mockedResolveAuthProfileOrder.mockReset();
mockedResolveAuthProfileOrder.mockReturnValue([]);
mockedResolveModelAsync.mockReset();
mockedResolveModelAsync.mockImplementation(
async (provider?: string, modelId?: string, _agentDir?: string, cfg?: unknown) =>
createMockResolvedModel(provider, modelId, cfg),
);
mockedRunEmbeddedAttempt.mockReset();
mockedAcquireAgentRunPreparedModelRuntime.mockClear();
mockedLog.warn.mockClear();
mockedMarkAuthProfileFailure.mockClear();
mockedSleepWithAbort.mockClear();
}
/** Install module mocks, import the runner, and return the mocked entrypoint. */
export async function loadRunOverflowCompactionHarness(): Promise<{
runEmbeddedAgent: typeof import("./run.js").runEmbeddedAgent;
@@ -7,7 +7,7 @@ import {
mockedClassifyFailoverReason,
mockedRunEmbeddedAttempt,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
useOpenAIPlatformAuthFixture,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
@@ -20,7 +20,7 @@ describe("runEmbeddedAgent prompt timeout fallback handoff", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
useOpenAIPlatformAuthFixture();
});
@@ -7,7 +7,7 @@ import {
mockedResolveAuthProfileOrder,
mockedRunEmbeddedAttempt,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
@@ -34,7 +34,7 @@ describe("runEmbeddedAgent timeout recovery composition", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
});
it("adopts a compacted transcript and retries with the complete runtime context", async () => {
@@ -10,7 +10,7 @@ import {
mockedGlobalHookRunner,
mockedRunEmbeddedAttempt,
overflowBaseRunParams,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
import { isEmbeddedAgentRunActive, queueEmbeddedAgentMessageWithOutcome } from "./runs.js";
@@ -23,7 +23,7 @@ describe("sessions_yield orchestration", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
mockedGlobalHookRunner.hasHooks.mockImplementation(() => false);
});
@@ -7,7 +7,7 @@ import {
mockedAcquireAgentRunPreparedModelRuntime,
mockedResolveModelAsync,
mockedRunEmbeddedAttempt,
resetRunOverflowCompactionHarnessMocks,
resetSharedRunIntegrationHarnessMocks,
} from "./run.overflow-compaction.harness.js";
import { loadSharedRunIntegrationHarness } from "./run.shared-integration-harness.test-support.js";
import type { EmbeddedRunAttemptResult } from "./run/types.js";
@@ -48,7 +48,7 @@ describe("runEmbeddedAgent usage reporting", () => {
});
beforeEach(() => {
resetRunOverflowCompactionHarnessMocks();
resetSharedRunIntegrationHarnessMocks();
});
it("bootstraps runtime plugins with the resolved workspace before running", async () => {