Skip to content

Claude Code. Any Model. The most powerful AI coding agent now speaks every language.

test(recovery): restore the suite to baseline after the Tier-1 ladder

Jack Rudenko4b6e56caa92e
diff --git a/packages/cli/src/handlers/composed-handler-connection-error.test.ts b/packages/cli/src/handlers/composed-handler-connection-error.test.ts
index 90f6689e..b29d3f50 100644
--- a/packages/cli/src/handlers/composed-handler-connection-error.test.ts
+++ b/packages/cli/src/handlers/composed-handler-connection-error.test.ts
@@ -107,7 +107,19 @@ interface CapturedResponse {
function makeContext(): { c: Context; captured: CapturedResponse } {
const captured: CapturedResponse = {};
const c = {
- req: { header: () => ({}) },
+ // These tests exercise CLASSIFICATION and STATUS, not the retry ladder —
+ // see the file header: "These tests add NO retry machinery and assert
+ // none." They opt out of the ladder exactly the way `probe-live.ts:157`
+ // does, with the header `composed-handler.ts:441` reads.
+ //
+ // Without this, a provoked connect failure enters the ladder and every
+ // assertion below times out at the test runner's limit instead of failing
+ // honestly — 12 tests reporting ~5000ms, which reads as a hang rather than
+ // as the deliberate behaviour change it is.
+ req: {
+ header: (name?: string) =>
+ name === undefined ? {} : name === "x-claudish-no-recovery" ? "1" : undefined,
+ },
header: () => {},
body: (body: BodyInit | null, init?: ResponseInit) => new Response(body, init),
json: (body: unknown, status?: number) => {
diff --git a/packages/cli/src/handlers/composed-handler-error.test.ts b/packages/cli/src/handlers/composed-handler-error.test.ts
index e2f57bdc..898f538c 100644
--- a/packages/cli/src/handlers/composed-handler-error.test.ts
+++ b/packages/cli/src/handlers/composed-handler-error.test.ts
@@ -112,7 +112,15 @@ interface CapturedResponse {
function makeContext(): { c: Context; captured: CapturedResponse } {
const captured: CapturedResponse = {};
const c = {
- req: { header: () => ({}) },
+ // Opt out of the Tier-1 retry ladder. These tests assert how an error is
+ // SURFACED, not whether it is retried, so they take the same escape hatch
+ // `probe-live.ts:157` uses — the header read at `composed-handler.ts:441`.
+ // Without it a provoked connect failure retries for the full deadline and
+ // the assertions time out rather than failing.
+ req: {
+ header: (name?: string) =>
+ name === undefined ? {} : name === "x-claudish-no-recovery" ? "1" : undefined,
+ },
header: () => {},
body: (body: BodyInit | null, init?: ResponseInit) => new Response(body, init),
json: (body: unknown, status?: number) => {
diff --git a/packages/cli/src/providers/transport/connect-evidence.test.ts b/packages/cli/src/providers/transport/connect-evidence.test.ts
index dc500888..0e67f0d7 100644
--- a/packages/cli/src/providers/transport/connect-evidence.test.ts
+++ b/packages/cli/src/providers/transport/connect-evidence.test.ts
@@ -129,37 +129,39 @@ describe("LocalTransport.refreshAuth — the evidence survives the health check"
});
/**
- * THE PROBE-COUNT TRIPWIRE.
+ * THE PROBE COUNT — a retried `refreshAuth()` must actually re-probe.
*
- * `refreshAuth()` opens with `if (this.healthChecked) return;`, and
- * `checkHealth()` sets `healthChecked = true` on FAILURE as well as on
- * success. The measured consequence, pinned below: the SECOND `refreshAuth()`
- * touches no network AND DOES NOT THROW — it returns successfully for a
- * server that is still dead.
+ * This was a TRIPWIRE pinning a bug, and the bug is now fixed; the test is
+ * flipped rather than deleted, because the reason it existed is the reason it
+ * still matters.
*
- * That is harmless today, because nothing calls `refreshAuth()` twice within
- * a request. It stops being harmless the moment a retry ladder does: attempt
- * 2 would return `ok` instantly, the episode would close as "recovered", and
- * a recovery record would be written for an outage that never ended.
+ * `refreshAuth()` opens with `if (this.healthChecked) return;`. `checkHealth()`
+ * used to set `healthChecked = true` on FAILURE as well as on success, so the
+ * SECOND `refreshAuth()` touched no network AND DID NOT THROW — it returned
+ * successfully for a server that was still dead.
*
- * This test therefore pins the CURRENT behaviour deliberately, as a tripwire:
- * whoever makes the latch success-only will see this test go red and must
- * read this comment before changing the numbers. It is not an endorsement.
+ * Harmless while nothing called it twice inside a request. The retry ladder
+ * calls it twice. Attempt 2 would have resolved instantly, the episode would
+ * have closed as "recovered", and a recovery record would have been written
+ * for an outage that never ended — while the real failure reappeared
+ * milliseconds later from the fetch path looking like a separate incident.
+ *
+ * The latch now records success only (`local.ts:271`). So: both calls probe,
+ * and both throw.
*/
- test("TRIPWIRE: the health-check latch currently suppresses the second probe entirely", async () => {
+ test("a retried refreshAuth re-probes and still throws while the server is down", async () => {
const probes = stubFailingProbes();
const transport = new LocalTransport(OLLAMA_CONFIG, "llama3.2");
await expect(transport.refreshAuth()).rejects.toThrow("Cannot connect to Ollama");
const afterFirst = probes.count;
- // Second call: no throw, no probe.
- await transport.refreshAuth();
+ // Second call: probes again, and still reports the server as unreachable.
+ // A resolve here is the "recovered from an outage that never ended" bug.
+ await expect(transport.refreshAuth()).rejects.toThrow("Cannot connect to Ollama");
expect(afterFirst).toBe(2); // /api/tags then /v1/models
- expect(probes.count).toBe(afterFirst); // ← flip to `afterFirst + 2` when the
- // latch becomes success-only. If this line is the only one you change, also
- // change the `.resolves` above to `.rejects`.
+ expect(probes.count).toBe(afterFirst + 2); // and again, rather than latching
});
});