Files
learnharness/tests/byok-refusal.test.ts
T
amosblomqvist 3b97056d01 test: infer the session send return type in refusal coverage
Avoid hardcoding Promise<void> so the test also typechecks with PR154's Promise<boolean> admission result. Type-only correction; no product behavior changes.
2026-10-27 19:17:27 +02:00

228 lines
11 KiB
TypeScript

import { describe, it } from "node:test";
import assert from "node:assert/strict";
import { parseModelsRefusal, parseProviderKeyRefusal, retryFailedResponse } from "../components/chat.tsx";
import { SessionModel } from "../components/session-core.ts";
import { fauxAssistantMessage } from "@earendil-works/pi-ai";
import { newUserMessage, type LhMessage } from "../src/core/native-messages.ts";
import { prefillFromSlots, prefillLevelFromPin, setupStage } from "../components/setup-dialog.tsx";
import {
isModelsRefusal,
isProviderKeyRefusal,
modelsNotConfigured,
providerKeyMissing,
PROVIDER_KEY_MISSING_CODE,
} from "../src/core/settings.ts";
import type { ModelCatalogEntry, ModelSlotName } from "../src/core/index.ts";
/**
* The BYOK universal gate rendered honestly (owner, 2026-09-22): the server
* refuses ALL turns with 503 {error, code: "models_not_configured", missing}
* when a slot tier is unset. These pin the client's pure seams: the refusal
* detector (the transport throws the 503 body verbatim as the error text),
* the job-record refusal matcher, and the SetupDialog's prefill.
*/
const REFUSAL = JSON.stringify(modelsNotConfigured(["strong"]));
describe("parseModelsRefusal — the send refusal detector", () => {
it("reads the missing tiers from the envelope text", () => {
assert.deepEqual(parseModelsRefusal(REFUSAL), ["strong"]);
assert.deepEqual(parseModelsRefusal(JSON.stringify(modelsNotConfigured(["strong", "vision"]))), [
"strong",
"vision",
]);
});
it("anything else is a real error — null, the honest banner stays", () => {
assert.equal(parseModelsRefusal(undefined), null);
assert.equal(parseModelsRefusal("Something went wrong"), null); // not JSON
assert.equal(parseModelsRefusal('{"error":"A turn is already running"}'), null); // JSON, wrong code
assert.equal(parseModelsRefusal('{"code":"models_not_configured","missing":"strong"}'), null); // missing not an array
assert.equal(parseModelsRefusal('{"code":"models_not_configured","missing":[42]}'), null); // tiers are strings
});
});
describe("Retry owns no composer recovery draft", () => {
for (const envelope of [providerKeyMissing("openrouter", ["strong"]), modelsNotConfigured(["strong"])]) {
it(`accepted A / failed model / unsent B / Retry ${envelope.code} retires only A's fallback`, async () => {
const durable: LhMessage[] = [
newUserMessage("Accepted text A"),
{ ...fauxAssistantMessage("Partial reply", { stopReason: "error", errorMessage: "Model failed" }), id: "failed-A" },
];
const submittedDraft: { current: string | null } = { current: "Accepted text A" };
let composer = "Unsent text B";
const body = JSON.stringify(envelope);
const model = new SessionModel({
chatId: "fixture", initialMessages: durable,
fetch: async (_url, init) => {
assert.equal(submittedDraft.current, null, "retire A before Retry can reach a refusal");
assert.deepEqual(JSON.parse(String(init?.body)).input, { type: "retry" });
return new Response(body, { status: 503 });
},
});
let request!: ReturnType<SessionModel["send"]>;
retryFailedResponse((input) => request = model.send(input), submittedDraft);
await request;
assert.equal(model.getState().error?.message, body, "503 stays raw for existing setup/picker routing");
assert.ok(parseModelsRefusal(body) || parseProviderKeyRefusal(body));
// The existing refusal effect restores only an owned submitted
// draft. Retry must leave this slot empty, not replace B with A.
if (submittedDraft.current !== null) composer = submittedDraft.current;
assert.equal(composer, "Unsent text B");
assert.deepEqual(model.getState().nativeMessages, durable, "Retry adds no optimistic learner message");
});
}
it("does not clear a later ordinary send's recovery draft when Retry settles", async () => {
const submittedDraft: { current: string | null } = { current: "Accepted text A" };
let finish!: () => void;
const pending = new Promise<void>((resolve) => { finish = resolve; });
retryFailedResponse(async () => pending, submittedDraft);
assert.equal(submittedDraft.current, null);
submittedDraft.current = "Later ordinary text C";
finish();
await pending;
await Promise.resolve();
assert.equal(submittedDraft.current, "Later ordinary text C", "no asynchronous cleanup may retire a newer owner");
});
it("ordinary text refusals still keep their submitted draft for recovery", async () => {
const submittedDraft: { current: string | null } = { current: "Ordinary text C" };
const body = JSON.stringify(providerKeyMissing("openrouter", ["strong"]));
const model = new SessionModel({
chatId: "fixture", initialMessages: [],
fetch: async () => new Response(body, { status: 503 }),
});
await model.send({ type: "text", text: submittedDraft.current! });
assert.equal(submittedDraft.current, "Ordinary text C");
assert.equal(model.getState().error?.message, body);
assert.equal(parseProviderKeyRefusal(model.getState().error?.message), true);
});
});
describe("isModelsRefusal — the job-record matcher", () => {
it("matches the runner's composed wording, not real failures", () => {
assert.equal(isModelsRefusal(`${modelsNotConfigured(["vision"]).error} (missing: vision)`), true);
assert.equal(isModelsRefusal(modelsNotConfigured(["cheap"]).error), true);
assert.equal(isModelsRefusal(undefined), false);
assert.equal(isModelsRefusal("FAILED: headless crashed"), false);
});
});
describe("providerKeyMissing — the envelope factory (provider-key interlude, 2026-09-25)", () => {
it("names the provider from the curated label and carries the exact machine code", () => {
const envelope = providerKeyMissing("openrouter", ["strong"]);
assert.deepEqual(envelope, {
error: "No OpenRouter key is set — add one in Settings → Keys",
code: PROVIDER_KEY_MISSING_CODE,
provider: "openrouter",
slots: ["strong"],
});
assert.equal(envelope.code, "provider_key_missing");
});
it("an unknown provider falls back to its raw id as the label", () => {
const envelope = providerKeyMissing("weird", ["cheap"]);
assert.equal(envelope.error, "No weird key is set — add one in Settings → Keys");
});
});
describe("parseProviderKeyRefusal — the send refusal detector's twin (zero-keys wave)", () => {
it("reads the envelope's own code + invariant tail — the wording can never drift client-side", () => {
const REFUSED = JSON.stringify(providerKeyMissing("openrouter", ["strong"]));
assert.equal(parseProviderKeyRefusal(REFUSED), true);
assert.equal(parseProviderKeyRefusal(JSON.stringify(providerKeyMissing("anthropic", ["strong", "vision"]))), true);
});
it("anything else is a real error — false, the honest banner stays", () => {
assert.equal(parseProviderKeyRefusal(REFUSAL), false); // the models refusal is NOT this one
assert.equal(parseProviderKeyRefusal("FAILED: headless crashed"), false);
assert.equal(parseProviderKeyRefusal(undefined), false);
assert.equal(parseProviderKeyRefusal("{not json"), false);
});
});
describe("isProviderKeyRefusal — the job-record matcher (provider-key interlude)", () => {
it("matches the envelope wording regardless of provider label, not real failures", () => {
assert.equal(isProviderKeyRefusal(providerKeyMissing("together", ["cheap"]).error), true);
assert.equal(isProviderKeyRefusal(providerKeyMissing("openai", ["strong"]).error), true);
assert.equal(isProviderKeyRefusal(""), false);
assert.equal(isProviderKeyRefusal(undefined), false);
assert.equal(isProviderKeyRefusal("FAILED: headless crashed"), false);
assert.equal(isProviderKeyRefusal(modelsNotConfigured(["strong"]).error), false, "the models refusal is a different failure");
});
});
describe("prefillFromSlots — the SetupDialog prefill (BYOK wave)", () => {
const entry = (id: string, provider: string): ModelCatalogEntry => ({
id,
name: id,
provider,
reasoning: false,
input: ["text"],
supportedLevels: ["low"],
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
contextWindow: 8192,
maxTokens: 1024,
});
const catalog = [entry("glm-5.3-flash", "z-ai"), entry("qwen3.8-flash", "qwen")];
const pins = (strong: boolean, vision: boolean) => ({
strong: strong ? { model: { provider: "z-ai", id: "glm-5.3-flash" } } : null,
cheap: null,
vision: vision ? { model: { provider: "qwen", id: "qwen3.8-flash" } } : null,
});
it("prefills exactly the pinned tiers, matched by provider AND id", () => {
const pre = prefillFromSlots(catalog, pins(true, true));
assert.equal(pre.strong?.id, "glm-5.3-flash");
assert.equal(pre.vision?.id, "qwen3.8-flash");
assert.equal(pre.cheap, undefined);
});
it("nothing pinned (first-run) → nothing prefilled — all three marked needed", () => {
const pre = prefillFromSlots(catalog, pins(false, false));
assert.deepEqual(pre, {});
});
it("a disconnected saved identity still prefills without selectable metadata", () => {
const pre = prefillFromSlots(catalog, { strong: { model: { provider: "gone", id: "vanished" } }, cheap: null, vision: null });
assert.deepEqual(pre, { strong: { provider: "gone", id: "vanished" } });
});
it("provider collisions never cross-pollute (same id, different provider)", () => {
const both = [entry("shared-id", "openai"), entry("shared-id", "openrouter")];
const pre = prefillFromSlots(both, { strong: { model: { provider: "openrouter", id: "shared-id" } }, cheap: null, vision: null });
assert.equal(pre.strong?.provider, "openrouter");
});
it("absent slots payload (older shape) prefills nothing", () => {
const pre = prefillFromSlots(catalog, undefined);
assert.deepEqual(pre, {});
});
});
describe("prefillLevelFromPin — the setup dialog's level prefill (chunk 2)", () => {
const entry = (levels: string[]) =>
({ id: "m", name: "M", provider: "p", reasoning: true, input: ["text"], supportedLevels: levels, cost: { input: 1, output: 1, cacheRead: 0, cacheWrite: 0 }, contextWindow: 1000, maxTokens: 1000 }) as never;
it("a stored level the pinned model still supports prefills as-is", () => {
assert.equal(prefillLevelFromPin(entry(["low", "high"]), { thinkingLevel: "high" }), "high");
});
it("a stored level the model LOST rides as-is — Save clamps server-side (bd68486)", () => {
assert.equal(prefillLevelFromPin(entry(["minimal", "medium"]), { thinkingLevel: "xhigh" }), "xhigh");
});
it("a legacy pin with no stored level prefills the ordering's level", () => {
assert.equal(prefillLevelFromPin(entry(["low", "medium"]), {}), "medium");
assert.equal(prefillLevelFromPin(entry(["low", "medium"]), null), "medium");
});
});
describe("setupStage — unknown is not keyless", () => {
it("keys first (owner ordering), pickers when a provider is keyed", () => {
assert.equal(setupStage(null), "loading");
assert.equal(setupStage([{ id: "x", keyPresent: false }]), "keys");
assert.equal(setupStage([{ id: "x", keyPresent: true }]), "pickers");
});
});