Files
openclaw/test/scripts/ci-changed-node-test-plan.test.ts
T
Peter Steinberger 34b8a6515d test(gateway): boot a minimal test gateway in the gateway lane (#121134)
* test(gateway): boot a minimal test gateway in the gateway lane

A gateway startup stall (#120926, awaited chat-metadata refresh in minimal
mode) shipped with green CI and first surfaced by hanging every checks-ui-e2e
suite that boots a minimal test gateway: no gateway-lane test booted one with
bundled plugins enabled (gateway.test.ts disables them, which masked the
stall), and the changed-scope classifier only selects the ui-e2e lane for
ui-touching diffs.

Add a minimal-gateway boot smoke in src/gateway that mirrors the ui-e2e boot
environment (bundled plugins enabled, minimal skips, strict time budget).
Selection is automatic: the smoke imports the gateway server, so the changed
node test plan picks it up through the import graph for any diff that can
affect startup, and full-suite plans run it in the
agentic-control-plane-startup-core shard. Reintroducing the #120926 hunk makes
the smoke fail (watchdog kill) in the gateway lane; healthy code boots in ~10s.

A planner regression test keeps the smoke classified as a gateway-server test
file and import-graph-reachable from gateway startup sources, so a rename or
graph-invisible import shape cannot silently drop the coverage again.

* test(ui): resolve labs rows by registry id, not positional index

Unblocks landing onto red main: 8fdf7570a1 (#120727) added the Cloud Worker
Desktop labs entry with a hardcoded row index that collides with Message audit
metadata, so enabling it toggled the neighboring row and the full-suite lane
failed deterministically (labs-page.test.ts, exposed only on PRs that run the
compact full suite). Derive each table case's row index from LAB_FEATURES by
feature id — the idiom the rest of the suite already uses — so a new labs
entry can no longer silently retarget an existing case.
2026-08-09 10:32:35 -07:00

267 lines
12 KiB
TypeScript

import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
import { tmpdir } from "node:os";
import path from "node:path";
import { describe, expect, it } from "vitest";
import {
createChangedNodeTestShards,
hasBuildArtifactAffectingChange,
hasPromptSnapshotAffectingChange,
hasQaSmokeAffectingChange,
} from "../../scripts/lib/ci-changed-node-test-plan.mts";
import { hasImportGraphImpactOnTargets } from "../../scripts/test-projects.test-support.mts";
import { listGitTrackedFiles } from "../../src/test-utils/repo-files.js";
import { isGatewayServerTestFile } from "../vitest/vitest.gateway-server-paths.mjs";
describe("CI changed Node test plan", () => {
it("routes a focused source change into one targeted job", () => {
expect(createChangedNodeTestShards(["src/agents/live-model-filter.ts"])).toEqual([
{
checkName: "checks-node-changed",
configs: [],
requiresDist: false,
runner: "blacksmith-8vcpu-ubuntu-2404",
shardName: "changed",
targets: [
"src/agents/live-model-filter.test.ts",
"src/agents/live-model-dynamic-candidates.test.ts",
"src/agents/model-compat.test.ts",
],
},
]);
});
it("keeps boundary coverage on test-only diffs without the build-artifacts lane", () => {
// Test-only diffs skip build-artifacts (which hosts the full boundary
// gate), so the plan carries its own nondist boundary shard instead.
expect(createChangedNodeTestShards(["test/extension-import-boundaries.test.ts"])).toEqual([
{
checkName: "checks-node-changed",
configs: [],
requiresDist: false,
runner: "blacksmith-8vcpu-ubuntu-2404",
shardName: "changed",
targets: ["test/extension-import-boundaries.test.ts"],
},
{
checkName: "checks-node-changed-boundary",
configs: ["test/vitest/vitest.boundary.config.ts"],
requiresDist: false,
runner: "blacksmith-8vcpu-ubuntu-2404",
shardName: "changed-boundary",
},
]);
});
it("classifies build-artifact and QA smoke impact by changed surface", () => {
expect(hasBuildArtifactAffectingChange(["src/agents/foo.test.ts", "test/helpers/x.ts"])).toBe(
false,
);
expect(hasBuildArtifactAffectingChange(["src/agents/foo.ts"])).toBe(true);
// Build-input classification: only sources and the build pipeline can
// change dist bytes; repo scripts, workflows, and qa scenarios cannot.
expect(hasBuildArtifactAffectingChange(["scripts/build-all.mts"])).toBe(true);
expect(hasBuildArtifactAffectingChange(["tsconfig.json"])).toBe(true);
expect(hasBuildArtifactAffectingChange(["scripts/run-vitest.mjs"])).toBe(false);
expect(hasBuildArtifactAffectingChange([".github/workflows/ci.yml"])).toBe(false);
expect(hasBuildArtifactAffectingChange(["qa/scenarios/index.yaml"])).toBe(false);
expect(hasBuildArtifactAffectingChange(["ui/src/app.ts"])).toBe(false);
expect(hasQaSmokeAffectingChange(["extensions/qa-lab/src/ci-smoke-plan.ts"])).toBe(true);
expect(hasQaSmokeAffectingChange(["ui/src/app.ts"])).toBe(true);
// Inside the packaged CLI's import graph -> the smoke scenarios can see it.
expect(hasQaSmokeAffectingChange(["src/infra/retry.ts"])).toBe(true);
// Smoke drives matrix + telegram; other channel plugins are invisible to it.
expect(hasQaSmokeAffectingChange(["extensions/telegram/src/index.ts"])).toBe(true);
expect(hasQaSmokeAffectingChange(["extensions/discord/src/index.ts"])).toBe(false);
expect(hasQaSmokeAffectingChange(["scripts/run-vitest.mjs"])).toBe(false);
expect(hasQaSmokeAffectingChange(["test/scripts/ci-node-test-plan.test.ts"])).toBe(false);
// The QA lane's own orchestration must not be able to skip the lane.
expect(hasQaSmokeAffectingChange([".github/workflows/ci.yml"])).toBe(true);
expect(hasQaSmokeAffectingChange([".github/actions/setup-node-env/action.yml"])).toBe(true);
expect(hasQaSmokeAffectingChange(["scripts/lib/ci-changed-node-test-plan.mts"])).toBe(true);
expect(hasQaSmokeAffectingChange([".github/workflows/labeler.yml"])).toBe(false);
// Deleted source files cannot be graphed; fail safe to running QA smoke.
expect(hasQaSmokeAffectingChange(["src/infra/definitely-deleted-module.ts"])).toBe(true);
});
it("classifies prompt-snapshot impact by surface and generator import graph", () => {
// Inside the generator's import graph -> regenerated output can change.
expect(hasPromptSnapshotAffectingChange(["src/auto-reply/reply/prompt-prelude.ts"])).toBe(true);
// The codex extension loads through a dynamic bundled-plugin module id the
// graph walk cannot see; it stays on the always-run surface.
expect(hasPromptSnapshotAffectingChange(["extensions/codex/src/index.ts"])).toBe(true);
expect(
hasPromptSnapshotAffectingChange([
"test/fixtures/agents/prompt-snapshots/codex-runtime-happy-path/README.md",
]),
).toBe(true);
expect(hasPromptSnapshotAffectingChange(["scripts/generate-prompt-snapshots.ts"])).toBe(true);
// Workspace packages feed the generator through package-specifier imports
// the relative graph walk cannot see.
expect(hasPromptSnapshotAffectingChange(["packages/llm-core/src/index.ts"])).toBe(true);
// The gate's own orchestration must not be able to skip the gated lane.
expect(hasPromptSnapshotAffectingChange([".github/workflows/ci.yml"])).toBe(true);
expect(hasPromptSnapshotAffectingChange(["scripts/lib/ci-changed-node-test-plan.mts"])).toBe(
true,
);
// Outside the surface and the generator graph -> the lane may skip.
expect(hasPromptSnapshotAffectingChange(["ui/src/app.ts"])).toBe(false);
expect(hasPromptSnapshotAffectingChange(["extensions/discord/src/index.ts"])).toBe(false);
expect(hasPromptSnapshotAffectingChange(["docs/index.md"])).toBe(false);
expect(hasPromptSnapshotAffectingChange(["test/scripts/ci-node-test-plan.test.ts"])).toBe(
false,
);
// Deleted source files cannot be graphed; fail safe to running the check.
expect(hasPromptSnapshotAffectingChange(["src/infra/definitely-deleted-module.ts"])).toBe(true);
});
it("fails safe to the full plan for broad changes", () => {
expect(createChangedNodeTestShards(["package.json"])).toBeNull();
});
it("keeps minimal-gateway boot coverage reachable from gateway startup changes", () => {
// A gateway startup stall must fail in the gateway lane; the boot smoke is
// selected purely through the import graph, so a rename or an import shape
// the graph walker cannot see would silently drop it from targeted plans
// and the stall would first surface on unrelated ui-e2e PRs again.
const bootSmoke = "src/gateway/server-startup-minimal-boot.test.ts";
expect(isGatewayServerTestFile(bootSmoke)).toBe(true);
expect(
hasImportGraphImpactOnTargets(
["src/gateway/server-startup-bootstrap.ts"],
[bootSmoke],
process.cwd(),
),
).toBe(true);
});
it("fails safe whenever a diff deletes source files", () => {
expect(createChangedNodeTestShards(["src/infra/format-time/deleted-helper.ts"])).toBeNull();
expect(
createChangedNodeTestShards([
"src/infra/format-time/deleted-helper.ts",
"src/agents/live-model-filter.ts",
]),
).toBeNull();
});
it("keeps targeting when a diff only deletes test files alongside live source", () => {
const shards = createChangedNodeTestShards([
"src/agents/deleted-obsolete.test.ts",
"src/agents/live-model-filter.ts",
]);
expect(shards).not.toBeNull();
const targets = shards?.flatMap((shard) => shard.targets ?? []) ?? [];
expect(targets).toContain("src/agents/live-model-filter.test.ts");
});
it("runs only the boundary shard when a diff deletes test files", () => {
const cwd = mkdtempSync(path.join(tmpdir(), "openclaw-ci-deleted-test-"));
try {
expect(createChangedNodeTestShards(["src/gone.test.ts"], { cwd })).toEqual([
{
checkName: "checks-node-changed-boundary",
configs: ["test/vitest/vitest.boundary.config.ts"],
requiresDist: false,
runner: "blacksmith-8vcpu-ubuntu-2404",
shardName: "changed-boundary",
},
]);
} finally {
rmSync(cwd, { force: true, recursive: true });
}
});
it("fails safe when an unresolved path is mixed with a precise source change", () => {
expect(
createChangedNodeTestShards(["src/agents/live-model-filter.ts", "tsconfig.json"]),
).toBeNull();
});
it("fails safe when public SDK changes affect extension imports", () => {
expect(createChangedNodeTestShards(["src/plugin-sdk/core.ts"])).toBeNull();
});
it("fails safe when a core change reaches package consumers through the public SDK", () => {
expect(createChangedNodeTestShards(["src/shared/text/strip-markdown.ts"])).toBeNull();
});
it("fails safe when a core change reaches a public SDK wrapper through an import", () => {
expect(createChangedNodeTestShards(["src/channels/chat-meta-shared.ts"])).toBeNull();
});
it("fails safe when workspace package consumers use package imports", () => {
expect(
createChangedNodeTestShards(["packages/gateway-protocol/src/frame-guards.ts"]),
).toBeNull();
});
it("fails safe when a targeted config needs special shard setup", () => {
expect(createChangedNodeTestShards(["scripts/docs-i18n/main.go"])).toBeNull();
expect(createChangedNodeTestShards(["src/tui/tui-pty-harness.e2e.test.ts"])).toBeNull();
});
it("fails safe when an unresolved source only finds an unrelated directory test", () => {
const cwd = mkdtempSync(path.join(tmpdir(), "openclaw-ci-target-"));
try {
mkdirSync(path.join(cwd, "src"));
writeFileSync(path.join(cwd, "src/value.ts"), "export const value = 1;\n");
writeFileSync(path.join(cwd, "src/unrelated.test.ts"), "export const unrelated = true;\n");
expect(createChangedNodeTestShards(["src/value.ts"], { cwd })).toBeNull();
} finally {
rmSync(cwd, { force: true, recursive: true });
}
});
it("fails safe for aggregate full-suite configs", () => {
expect(
createChangedNodeTestShards(["test/vitest/vitest.full-core-support-boundary.config.ts"]),
).toBeNull();
});
it("fails safe for leaf configs split across full-suite processes", () => {
expect(createChangedNodeTestShards(["test/vitest/vitest.commands.config.ts"])).toBeNull();
});
it("fails safe when source targets expand to a whole config", () => {
expect(
createChangedNodeTestShards(["ui/src/app-routes.ts", "ui/src/app-navigation.ts"]),
).toBeNull();
});
it("chunks many targets into bounded parallel jobs", () => {
// A wide test-file diff exercises the multi-chunk path against the real
// tree; the cron suite has well over one chunk's worth of test files.
const changedTests = listGitTrackedFiles({ pathspecs: "src/cron" })
?.filter((file) => file.endsWith(".test.ts") && !/\.(?:e2e|live)\.test\.ts$/u.test(file))
.slice(0, 15);
expect(changedTests?.length).toBe(15);
const shards = createChangedNodeTestShards(changedTests ?? []);
expect(shards).not.toBeNull();
const targetShards = shards?.filter((shard) => shard.targets) ?? [];
expect(targetShards.length).toBeGreaterThan(1);
expect(
targetShards.every((shard, index) => shard.checkName === `checks-node-changed-${index + 1}`),
).toBe(true);
expect(targetShards.every((shard) => (shard.targets?.length ?? 0) <= 12)).toBe(true);
const targets = targetShards.flatMap((shard) => shard.targets ?? []);
expect(new Set(targets).size).toBe(targets.length);
});
it("serializes changed-test chunks that contain Memory Core integration tests", () => {
const shards = createChangedNodeTestShards([
"extensions/memory-core/src/memory/mmr.ts",
"extensions/memory-core/src/memory/mmr.test.ts",
]);
expect(shards).not.toBeNull();
const targetShards = shards?.filter((shard) => shard.targets) ?? [];
expect(targetShards.length).toBeGreaterThan(0);
expect(
targetShards
.filter((shard) =>
shard.targets?.some((target) => target.startsWith("extensions/memory-core/")),
)
.every((shard) => shard.planConcurrency === 1),
).toBe(true);
});
});