Files
OmniRoute/tests/unit/provider-models-route-codex.test.ts
Xiangzhe c2df757610 fix(tests): drain base-red cluster from 2026-08-23 merges (#9985)
Drain the unit-shard reds the 2026-08-23 merge wave left on
release/v3.8.50, each discriminated as stale-test (contract moved
intentionally, test aligned) vs real bug (fixed in code/messages):

1. check-deps 6A.8 allowlist — #11224 added @testing-library/dom and
   @testing-library/user-event to package.json without the gate
   allowlist. Both are legitimate: @testing-library/dom is a required
   peer of @testing-library/react v16, and user-event is the official
   companion for UI tests. Fix: allowlist entries with a justification
   note referencing #11224/#9985.

2. search-route 400-vs-fallback — #11097 intentionally changed the
   zero-credential /v1/search contract: instead of returning 400 it
   promotes the fallback-only duckduckgo-free provider so out-of-the-box
   search works. The test pinned the OLD 400 contract. Fix (contract
   alignment): the test now pins the new fallback contract — 200,
   provider duckduckgo-free, DuckDuckGo lite endpoint called, results
   parsed from lite HTML.

3. codex catalog token limits (4 named reds + 2 sibling sweeps) —
   #11179 raised GPT_5_6_CODEX_CAPABILITIES from the 272K pricing tier
   to the real usable 872K window (live evidence: 390K served past
   272K with HTTP 200) and updated codex-gpt56-catalog.test.ts but
   missed the sibling pins. Stale tests aligned: models-catalog-combo-
   metadata (max_input_tokens now clamps to min(872000, 500000 override)
   = 500000), vscode-token-routes x3 (872000), and two more found in the
   sibling sweep: vscode-token-routes-gpt56 and provider-models-route-
   codex (conservative merge semantics unchanged: pinned 872000 < live
   999999 still wins).

4. CLI catalog counts (3 reds) — #11166 added prime-agent (agent
   category) without the cardinality pins: EXPECTED_AGENT_COUNT 8->9,
   total 34->35, D15 agent list + prime-agent, cli-tools-schema id list.
   Also added the missing English/Vietnamese cliTools descriptions for
   prime-agent (cli-catalog-display-contract) and corrected the stale
   CLI-TOOLS.md agent count (8->9).

5. setup-qwen container guard — since #10057 the container guard exits 2
   on ephemeral runtimes; the two tests exercising the merge/write path
   were environment-sensitive (red on container devboxes). Fix: pass
   allowContainerWrite so the tests are hermetic everywhere; the guard
   keeps its own dedicated coverage.

6. i18n health verdict namespace (real bug, fixed in messages) — #11224
   added the verdict/diagnostics strings to the `sidebar` namespace but
   health/page.tsx reads them via useTranslations("health"), so the page
   rendered raw keys in every locale and the "direct translation calls
   have English messages" gate went red. Fix: keys moved sidebar->health
   in en.json + vi.json (the only locales that had them), plus
   health.healthSubtitle added. The sidebar never referenced them
   (verified: no usage), and sidebar.healthSubtitle (its real sidebar
   key) is untouched.

7. i18n pt-BR drift (22 keys) — the 08-23 wave (#11224/#11228/#11215/
   #11204/#11195) added English keys never translated: common.batch*,
   endpoint.*, cliCommon.concept.acp.warning, resilienceConnections.*,
   plus the moved health.* keys and the prime-agent cliTools
   description. Fix: pt-BR translations added — 0 missing keys vs en;
   vi strict parity re-verified (0 missing, 0 extra).

Validation: every touched test file RED->GREEN individually
(node --import tsx/esm --test), typecheck:core clean, docs-counts-sync
soft-pass, cli-i18n gate PASS, 8/8 unit shards re-run on the branch.

Refs #9985
2026-08-23 14:45:05 -03:00

431 lines
15 KiB
TypeScript

import test from "node:test";
import assert from "node:assert/strict";
import fs from "node:fs";
import os from "node:os";
import path from "node:path";
const TEST_DATA_DIR = fs.mkdtempSync(
path.join(os.tmpdir(), "omniroute-provider-model-routes-codex-")
);
process.env.DATA_DIR = TEST_DATA_DIR;
const core = await import("../../src/lib/db/core.ts");
const providersDb = await import("../../src/lib/db/providers.ts");
const modelsDb = await import("../../src/lib/db/models.ts");
const providerModelsRoute = await import("../../src/app/api/providers/[id]/models/route.ts");
const codexDiscovery = await import("../../src/app/api/providers/[id]/models/discovery/codex.ts");
type RouteModel = {
id: string;
name?: string;
apiFormat?: string;
supportedEndpoints?: string[];
inputTokenLimit?: number;
outputTokenLimit?: number;
[key: string]: unknown;
};
type RouteBody = {
provider?: string;
models?: RouteModel[];
source?: string;
warning?: string;
intentional?: boolean;
discoveredCandidateCount?: number;
};
type ProviderOverrides = {
authType?: string;
apiKey?: string | null;
accessToken?: string | null;
providerSpecificData?: Record<string, unknown>;
};
const originalFetch = globalThis.fetch;
async function resetStorage() {
globalThis.fetch = originalFetch;
codexDiscovery.clearCodexGithubCatalogCacheForTests();
core.resetDbInstance();
fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
}
async function seedCodexConnection(overrides: ProviderOverrides = {}) {
return providersDb.createProviderConnection({
provider: "codex",
authType: overrides.authType || "oauth",
name: `codex-${Math.random().toString(16).slice(2, 8)}`,
apiKey: overrides.apiKey,
accessToken: overrides.accessToken,
isActive: true,
testStatus: "active",
providerSpecificData: overrides.providerSpecificData || {},
});
}
async function callRoute(connectionId: string, search = "") {
return providerModelsRoute.GET(
new Request(`http://localhost/api/providers/${connectionId}/models${search}`),
{ params: { id: connectionId } }
);
}
test.beforeEach(async () => {
await resetStorage();
});
test.after(async () => {
globalThis.fetch = originalFetch;
codexDiscovery.clearCodexGithubCatalogCacheForTests();
core.resetDbInstance();
fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
});
test("provider models route merges live Codex models with the local catalog then filters denylist", async () => {
const connection = await seedCodexConnection({
accessToken: "codex-access-token",
providerSpecificData: { chatgptAccountId: "account-123" },
});
await modelsDb.replaceSyncedAvailableModelsForConnection("codex", connection.id, [
{ id: "stale-codex-model", name: "Stale Codex Model", source: "imported" },
]);
const seenRequests: Array<Record<string, string | null>> = [];
globalThis.fetch = async (url, init) => {
const requestUrl = String(url);
const headers = new Headers(init?.headers as HeadersInit | undefined);
seenRequests.push({
url: requestUrl,
authorization: headers.get("authorization"),
workspaceId: headers.get("chatgpt-account-id"),
originator: headers.get("originator"),
userAgent: headers.get("user-agent"),
});
if (requestUrl.includes("raw.githubusercontent.com/openai/codex")) {
return Response.json({
models: [
{
slug: "gpt-5.6-sol",
display_name: "GPT 5.6 Sol GitHub",
visibility: "list",
supported_in_api: true,
minimal_client_version: "0.144.0",
context_window: 372000,
input_modalities: ["text", "image"],
supported_reasoning_levels: [{ effort: "low" }, { effort: "high" }],
},
{
slug: "gpt-5.4",
display_name: "Retired GPT 5.4 GitHub",
visibility: "list",
supported_in_api: true,
},
],
});
}
return Response.json({
models: [
{ slug: "codex-auto-review", visibility: "hide", supported_in_api: true },
{
slug: "gpt-5.6-sol",
display_name: "GPT 5.6 Sol Live",
visibility: "list",
supported_in_api: true,
max_input_tokens: 999999,
max_output_tokens: 999999,
},
{
slug: "gpt-5.4",
display_name: "Retired GPT 5.4 Live",
visibility: "list",
supported_in_api: true,
},
{ id: "", name: "missing-id" },
],
});
};
const response = await callRoute(connection.id, "?refresh=true");
const body = (await response.json()) as RouteBody;
const modelIds = new Set(body.models?.map((model) => model.id));
const liveModel = body.models?.find((model) => model.id === "gpt-5.6-sol");
const syncedModels = await modelsDb.getSyncedAvailableModelsForConnection("codex", connection.id);
const syncedIds = new Set(syncedModels.map((model) => model.id));
assert.equal(response.status, 200);
assert.equal(body.provider, "codex");
assert.equal(body.source, "api");
assert.equal(body.discoveredCandidateCount, undefined);
assert.deepEqual(seenRequests, [
{
url: "https://chatgpt.com/backend-api/codex/models?client_version=0.149.0",
authorization: "Bearer codex-access-token",
workspaceId: "account-123",
originator: "codex_cli_rs",
userAgent: "codex-cli/0.149.0 (Windows 10.0.26200; x64)",
},
{
url: "https://raw.githubusercontent.com/openai/codex/refs/heads/main/codex-rs/models-manager/models.json",
authorization: null,
workspaceId: null,
originator: null,
userAgent: null,
},
]);
assert.ok(modelIds.has("gpt-5.6-sol"));
assert.ok(modelIds.has("gpt-5.6-sol-ultra"));
assert.ok(modelIds.has("gpt-5.6-sol-max"));
// Live payload wins on overlapping fields; local catalog supplies local-only
// variants. EXCEPTION: capacity limits (inputTokenLimit/outputTokenLimit)
// merge conservatively — the smaller of live vs. pinned wins, never the
// larger, so a stale/inflated live number can never make OmniRoute promise
// more context than the account can actually serve (#7012). Here the pinned
// GPT-5.6 Codex contract (872000/128000, see GPT_5_6_CODEX_CAPABILITIES — raised
// from the old 272K pricing tier to the real usable window by #11179)
// is smaller than the live payload's 999999/999999, so the pinned value wins.
assert.equal(liveModel?.name, "GPT 5.6 Sol Live");
assert.equal(liveModel?.inputTokenLimit, 872000);
assert.equal(liveModel?.outputTokenLimit, 128000);
assert.equal(liveModel?.apiFormat, "responses");
assert.deepEqual(liveModel?.supportedEndpoints, ["responses"]);
assert.equal(liveModel?.supportsThinking, true);
assert.equal(liveModel?.supportsVision, true);
assert.ok(modelIds.has("gpt-5.5-low"));
assert.equal(
[...modelIds].some((id) => String(id).startsWith("gpt-5.4")),
false
);
assert.ok(syncedIds.has("gpt-5.6-sol"));
assert.ok(syncedIds.has("gpt-5.5-low"));
assert.equal(
[...syncedIds].some((id) => String(id).startsWith("gpt-5.4")),
false
);
// Stale cache-only ids are replaced when a fresh discovery response is persisted.
assert.equal(modelIds.has("stale-codex-model"), false);
assert.equal(syncedIds.has("stale-codex-model"), false);
});
test("provider models route: live token limit wins when it is SMALLER than the pinned local catalog value", async () => {
const connection = await seedCodexConnection({
accessToken: "codex-access-token",
providerSpecificData: { chatgptAccountId: "account-123" },
});
globalThis.fetch = async (url) => {
const requestUrl = String(url);
if (requestUrl.includes("raw.githubusercontent.com/openai/codex")) {
return Response.json({ models: [] });
}
// Live reports a SMALLER budget than the pinned GPT-5.6 Codex contract
// (272000/128000, GPT_5_6_CODEX_CAPABILITIES) — e.g. a temporary
// account-level cap. The conservative merge must take the smaller live
// value here, not the larger pinned one (#7012).
return Response.json({
models: [
{
slug: "gpt-5.6-sol",
display_name: "GPT 5.6 Sol Live",
visibility: "list",
supported_in_api: true,
max_input_tokens: 100000,
max_output_tokens: 50000,
},
],
});
};
const response = await callRoute(connection.id, "?refresh=true");
const body = (await response.json()) as RouteBody;
const liveModel = body.models?.find((model) => model.id === "gpt-5.6-sol");
assert.equal(response.status, 200);
assert.equal(liveModel?.inputTokenLimit, 100000);
assert.equal(liveModel?.outputTokenLimit, 50000);
});
test("provider models route uses the GitHub Codex catalog when live discovery fails", async () => {
const connection = await seedCodexConnection({ accessToken: "codex-access-token" });
const seenUrls: string[] = [];
globalThis.fetch = async (url) => {
const requestUrl = String(url);
seenUrls.push(requestUrl);
if (requestUrl.includes("raw.githubusercontent.com/openai/codex")) {
return Response.json({
models: [
{
slug: "gpt-5.6-sol",
display_name: "GPT-5.6-Sol",
visibility: "list",
supported_in_api: true,
minimal_client_version: "0.144.0",
context_window: 372000,
},
{
slug: "gpt-5.4",
display_name: "Retired GPT-5.4",
visibility: "list",
supported_in_api: true,
},
],
});
}
return new Response("upstream unavailable", { status: 503 });
};
const response = await callRoute(connection.id, "?refresh=true");
const body = (await response.json()) as RouteBody;
const modelIds = new Set((body.models || []).map((model) => model.id));
assert.equal(response.status, 200);
assert.equal(body.provider, "codex");
assert.equal(body.source, "api");
assert.equal(body.intentional, undefined);
assert.equal(body.warning, "Codex live catalog unavailable — using GitHub model catalog");
assert.equal(body.discoveredCandidateCount, undefined);
assert.ok(seenUrls.some((url) => url.includes("backend-api/codex/models")));
assert.ok(seenUrls.some((url) => url.includes("raw.githubusercontent.com/openai/codex")));
assert.ok(modelIds.has("gpt-5.6-sol"));
assert.ok(modelIds.has("gpt-5.5-low"));
assert.equal(
[...modelIds].some((id) => String(id).startsWith("gpt-5.4")),
false
);
});
test("provider models route returns cached Codex models when refresh discovery fails", async () => {
const connection = await seedCodexConnection({ accessToken: "codex-access-token" });
await modelsDb.replaceSyncedAvailableModelsForConnection("codex", connection.id, [
{
id: "gpt-5.4",
name: "Retired Cached GPT 5.4",
source: "imported",
apiFormat: "responses",
supportedEndpoints: ["responses"],
},
{
id: "gpt-5.6-sol",
name: "Cached GPT 5.6 Sol",
source: "imported",
apiFormat: "responses",
supportedEndpoints: ["responses"],
},
]);
globalThis.fetch = async () => new Response("upstream unavailable", { status: 503 });
const response = await callRoute(connection.id, "?refresh=true");
const body = (await response.json()) as RouteBody;
assert.equal(response.status, 200);
assert.equal(body.provider, "codex");
assert.equal(body.source, "cache");
assert.equal(body.warning, "Codex live catalog unavailable — using cached catalog");
assert.equal(body.discoveredCandidateCount, undefined);
const modelIds = new Set((body.models || []).map((model) => model.id));
assert.ok(modelIds.has("gpt-5.6-sol"));
assert.ok(modelIds.has("gpt-5.6-sol-ultra"));
assert.equal(
[...modelIds].some((id) => String(id).startsWith("gpt-5.4")),
false
);
const syncedModels = await modelsDb.getSyncedAvailableModelsForConnection("codex", connection.id);
const syncedIds = new Set(syncedModels.map((model) => model.id));
assert.ok(syncedIds.has("gpt-5.6-sol-ultra"));
assert.equal(syncedIds.has("gpt-5.4"), false);
});
test("provider models route auto-includes remote-only Codex models after merge", async () => {
const connection = await seedCodexConnection({ accessToken: "codex-access-token" });
globalThis.fetch = async (url) => {
const requestUrl = String(url);
if (requestUrl.includes("raw.githubusercontent.com/openai/codex")) {
return Response.json({ models: [] });
}
return Response.json({
models: [
{
slug: "future-codex-experimental",
display_name: "Future Codex Experimental",
visibility: "list",
supported_in_api: true,
},
{
slug: "gpt-5.6-sol",
display_name: "GPT 5.6 Sol Live",
visibility: "list",
supported_in_api: true,
},
{
slug: "gpt-5.4",
display_name: "Retired GPT 5.4 Live",
visibility: "list",
supported_in_api: true,
},
],
});
};
const response = await callRoute(connection.id, "?refresh=true");
const body = (await response.json()) as RouteBody;
const modelIds = new Set((body.models || []).map((model) => model.id));
const syncedModels = await modelsDb.getSyncedAvailableModelsForConnection("codex", connection.id);
const syncedIds = new Set(syncedModels.map((model) => model.id));
assert.equal(response.status, 200);
assert.equal(body.source, "api");
assert.ok(modelIds.has("future-codex-experimental"));
assert.ok(modelIds.has("gpt-5.6-sol"));
assert.equal(modelIds.has("gpt-5.4"), false);
assert.ok(syncedIds.has("future-codex-experimental"));
assert.equal(syncedIds.has("gpt-5.4"), false);
});
test("provider models route falls back to local Codex catalog when live and GitHub fail", async () => {
const connection = await seedCodexConnection({ accessToken: "codex-access-token" });
globalThis.fetch = async () => new Response("upstream unavailable", { status: 503 });
const response = await callRoute(connection.id, "?refresh=true");
const body = (await response.json()) as RouteBody;
assert.equal(response.status, 200);
assert.equal(body.provider, "codex");
assert.equal(body.source, "local_catalog");
assert.equal(body.intentional, true);
assert.equal(body.warning, "Codex live and GitHub catalogs unavailable — using local catalog");
assert.ok(body.models?.some((model) => model.id === "gpt-5.6-sol"));
assert.ok(body.models?.some((model) => model.id === "gpt-5.5"));
assert.equal(
body.models?.some((model) => model.id.startsWith("gpt-5.4")),
false
);
});
test("provider models route returns curated GPT-5.6 variants when auto-fetch is disabled", async () => {
const connection = await seedCodexConnection({
apiKey: null,
accessToken: "codex-access",
providerSpecificData: { autoFetchModels: false },
});
const response = await callRoute(connection.id);
const body = (await response.json()) as RouteBody;
const modelIds = new Set((body.models || []).map((model) => model.id));
assert.equal(response.status, 200);
assert.equal(body.provider, "codex");
assert.equal(body.source, "local_catalog");
assert.ok(modelIds.has("gpt-5.6-sol-ultra"));
assert.ok(modelIds.has("gpt-5.6-sol-max"));
assert.ok(modelIds.has("gpt-5.6-terra-ultra"));
assert.ok(modelIds.has("gpt-5.6-luna-max"));
assert.equal(
[...modelIds].some((id) => String(id).startsWith("gpt-5.4")),
false
);
});