buenosair/lvmh
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

web: fix args-object React crash; e2e plugin-contract scenario

Browse Source
This commit is contained in:
Raphael Westphal
2026-08-18 16:32:21 +02:00
parent 7f0672f5d8
commit 1cc5d9a978
5 changed files with 625 additions and 536 deletions
Download Patch File Download Diff File Expand all files Collapse all files
e2e/driver.mjs
+24 -7
View File
@@ -81,7 +81,9 @@ process.on("SIGTERM", () => {
async function main() {
console.log("=== lvmh e2e integration ===");
const built = ensureWebDist();
console.log(`web dist: ${built ? "built now" : "reused"} ${path.relative(REPO_ROOT, WEB_DIST)}`);
console.log(
`web dist: ${built ? "built now" : "reused"} ${path.relative(REPO_ROOT, WEB_DIST)}`,
);
fs.mkdirSync(path.join(REPO_ROOT, ".pi", "scratch"), { recursive: true });
tmpDir = fs.mkdtempSync(path.join(REPO_ROOT, ".pi", "scratch", "e2e-"));
@@ -99,10 +101,13 @@ async function main() {
LVMH_CONTAINER_LVMH_URL: `ws://${addr}/agent/ws`,
LVMH_WORKER_DOCKERFILE: path.join(tmpDir, "absent-worker.Dockerfile"),
};
const boot = () => startDaemon({ addr, dbPath, webdist: WEB_DIST, logPath, env: daemonEnv });
const boot = () =>
startDaemon({ addr, dbPath, webdist: WEB_DIST, logPath, env: daemonEnv });
daemon = await boot();
console.log(`daemon: ${daemon.baseUrl} (go run . in ${path.relative(REPO_ROOT, DAEMON_DIR)}, db ${dbPath})`);
console.log(
`daemon: ${daemon.baseUrl} (go run . in ${path.relative(REPO_ROOT, DAEMON_DIR)}, db ${dbPath})`,
);
console.log(`fake gitlab: ${fakeGitLab.url}`);
const ctx = {
@@ -125,18 +130,30 @@ async function main() {
await scenario(ctx);
} catch (err) {
r.check(`${name}: scenario crashed`, false, String(err?.stack ?? err));
const log = fs.existsSync(logPath) ? fs.readFileSync(logPath, "utf8").split("\n").slice(-15).join("\n") : "";
if (log) console.log(` daemon.log tail:\n${log.split("\n").map((l) => " " + l).join("\n")}`);
const log = fs.existsSync(logPath)
? fs.readFileSync(logPath, "utf8").split("\n").slice(-15).join("\n")
: "";
if (log)
console.log(
` daemon.log tail:\n${log
.split("\n")
.map((l) => " " + l)
.join("\n")}`,
);
}
}
const { ok, fail, xfail, xpass, total } = r.summary();
console.log("\n──────────────────────────────");
console.log(`passed ${ok} / ${total} failed ${fail} xfail ${xfail} (known bugs) xpass ${xpass}`);
console.log(
`passed ${ok} / ${total} failed ${fail} xfail ${xfail} (known bugs) xpass ${xpass}`,
);
if (fail > 0) {
console.log("failed checks:");
for (const e of r.entries.filter((e) => e.status === "fail")) {
console.log(` [${e.scenario}] ${e.name}${e.detail ? " — " + e.detail : ""}`);
console.log(
` [${e.scenario}] ${e.name}${e.detail ? " — " + e.detail : ""}`,
);
}
}
console.log(`RESULT: ${fail === 0 ? "GREEN" : "RED"}`);
e2e/scenarios/plugin-contract.mjs
+50 -12
View File
@@ -9,30 +9,45 @@
// exact drift that crashed React #31 in production. This scenario captures a
// real session via the real plugin when possible and hard-fails on any frame
// shape the web UI cannot consume safely.
import { WSSock, agentHello, sid, sessionSnapshot, rest, sleep } from "../lib.mjs";
import {
WSSock,
agentHello,
sid,
sessionSnapshot,
rest,
sleep,
} from "../lib.mjs";
const SESSION_ID = sid("e2e-contract");
// Shape predicates — the web's actual consumption rules (web/src/derive.ts).
function frameRenderViolations(frame) {
const errs = [];
if (frame.type === "tool_execution_start" || frame.type === "tool_execution_update") {
if (
frame.type === "tool_execution_start" ||
frame.type === "tool_execution_update"
) {
// args: unknown is fine (deriveChat normalizes) — but circular/unserializable
// payloads would break the daemon's persistence; verify JSON round-trip.
if (frame.args !== undefined && frame.args !== null) {
try {
JSON.parse(JSON.stringify(frame.args));
} catch {
errs.push(`${frame.type}.args not JSON-safe: ${String(frame.args).slice(0, 60)}`);
errs.push(
`${frame.type}.args not JSON-safe: ${String(frame.args).slice(0, 60)}`,
);
}
}
}
if (frame.type === "message_end") {
const m = frame.message;
if (typeof m?.text !== "string") errs.push("message_end.message.text not string");
if (typeof m?.text !== "string")
errs.push("message_end.message.text not string");
for (const c of m?.toolCalls ?? []) {
if (typeof c?.argsJson !== "string") errs.push(`toolCall ${c?.id} argsJson not string`);
if (typeof c?.name !== "string") errs.push(`toolCall ${c?.id} name not string`);
if (typeof c?.argsJson !== "string")
errs.push(`toolCall ${c?.id} argsJson not string`);
if (typeof c?.name !== "string")
errs.push(`toolCall ${c?.id} name not string`);
}
}
return errs;
@@ -41,7 +56,9 @@ function frameRenderViolations(frame) {
export async function run(ctx) {
const { r, base, token, agentUrl } = ctx;
const session = sessionSnapshot(SESSION_ID);
const agent = new WSSock(agentUrl, { headers: { Authorization: `Bearer ${token}` } });
const agent = new WSSock(agentUrl, {
headers: { Authorization: `Bearer ${token}` },
});
if (!(await agent.opened())) {
r.check("contract agent connects", false, "agent WS failed to open");
return;
@@ -95,26 +112,46 @@ export async function run(ctx) {
id: "msg-1",
text: "Reading src/main.ts",
thinking: null,
toolCalls: [{ id: "call-read-1", name: "read", argsJson: '{"path":"src/main.ts","offset":1}' }],
toolCalls: [
{
id: "call-read-1",
name: "read",
argsJson: '{"path":"src/main.ts","offset":1}',
},
],
toolCallId: null,
},
},
{ type: "agent_settled", seq: 8 },
];
for (const f of frames) {
agent.send({ v: 1, sessionId: session.id, ts: Date.now(), ...f, seq: f.seq });
agent.send({
v: 1,
sessionId: session.id,
ts: Date.now(),
...f,
seq: f.seq,
});
}
// Wait for persistence to catch up, then read back everything the web would.
const persisted = await (async () => {
for (let i = 0; i < 40; i++) {
const res = await rest(base, token, `/api/sessions/${session.id}/events?after=0&limit=100`);
const res = await rest(
base,
token,
`/api/sessions/${session.id}/events?after=0&limit=100`,
);
if (res.json?.length >= 6) return res.json;
await sleep(25);
}
return [];
})();
r.check("contract: all 6 persisted frames round-tripped", persisted.length === 6, `got ${persisted.length}`);
r.check(
"contract: all 6 persisted frames round-tripped",
persisted.length === 6,
`got ${persisted.length}`,
);
r.check(
"contract: message_update not persisted",
persisted.every((f) => f.type !== "message_update"),
@@ -130,7 +167,8 @@ export async function run(ctx) {
const objectArgs = persisted.find((f) => f.type === "tool_execution_start");
r.check(
"contract: object tool args survive persistence verbatim",
typeof objectArgs?.args === "object" && objectArgs.args?.path === "src/main.ts",
typeof objectArgs?.args === "object" &&
objectArgs.args?.path === "src/main.ts",
`args=${JSON.stringify(objectArgs?.args)}`,
);
web/src/derive.test.ts
+3 -1
View File
@@ -54,7 +54,9 @@ function toolEnd(id: string, isError?: boolean, preview?: string): EventFrame {
// render as-is (React #31 crash) — deriveChat must normalize to text.
describe("object tool args from the real plugin", () => {
it("argsText normalizes every shape the plugin emits", () => {
expect(argsText({ path: "src/main.ts" })).toBe('{\n "path": "src/main.ts"\n}');
expect(argsText({ path: "src/main.ts" })).toBe(
'{\n "path": "src/main.ts"\n}',
);
expect(argsText('"plain"')).toBe('"plain"');
expect(argsText(42)).toBe("42");
expect(argsText(null)).toBe("");
web/src/derive.ts
+41 -13
View File
@@ -2,7 +2,10 @@ import type { EventFrame } from "./protocol";
// ---------- event list merge ----------
export function mergeEvents(existing: EventFrame[], incoming: EventFrame[]): EventFrame[] {
export function mergeEvents(
existing: EventFrame[],
incoming: EventFrame[],
): EventFrame[] {
const bySeq = new Map<number, EventFrame>();
for (const e of existing) bySeq.set(e.seq, e);
for (const e of incoming) bySeq.set(e.seq, e);
@@ -13,7 +16,8 @@ export function mergeEvents(existing: EventFrame[], incoming: EventFrame[]): Eve
* must not advance the refetch cursor. */
export function lastPersistedSeq(events: EventFrame[]): number {
let max = 0;
for (const e of events) if (e.type !== "message_update" && e.seq > max) max = e.seq;
for (const e of events)
if (e.type !== "message_update" && e.seq > max) max = e.seq;
return max;
}
@@ -79,7 +83,8 @@ export function deriveChat(events: EventFrame[]): ChatDerivation {
break;
}
case "tool_execution_end": {
const t = e.toolCallId !== undefined ? tools.get(e.toolCallId) : undefined;
const t =
e.toolCallId !== undefined ? tools.get(e.toolCallId) : undefined;
if (t !== undefined) {
t.running = false;
t.isError = e.isError ?? false;
@@ -94,7 +99,8 @@ export function deriveChat(events: EventFrame[]): ChatDerivation {
busy = false;
break;
case "message_start":
if (e.message?.role === "assistant") stream = { id: e.message.id, text: "" };
if (e.message?.role === "assistant")
stream = { id: e.message.id, text: "" };
break;
case "message_update":
if (stream !== null) stream.text += e.delta ?? "";
@@ -190,7 +196,9 @@ function normalizeStatus(raw: unknown): TodoStatus {
}
}
function extractSnapshot(raw: unknown): { content: string; status: TodoStatus }[] | null {
function extractSnapshot(
raw: unknown,
): { content: string; status: TodoStatus }[] | null {
let arr: unknown = raw;
if (Array.isArray(raw) === false && raw !== null && typeof raw === "object") {
const o = raw as Record<string, unknown>;
@@ -224,7 +232,10 @@ export function deriveTasks(events: EventFrame[]): TaskDerivation {
}
// todo snapshots, seq-ordered: args from execution start, result from execution end + toolResult message
const snapshots: { seq: number; items: { content: string; status: TodoStatus }[] }[] = [];
const snapshots: {
seq: number;
items: { content: string; status: TodoStatus }[];
}[] = [];
const subagents: SubagentRun[] = [];
const runningTools: ToolState[] = [];
@@ -236,11 +247,18 @@ export function deriveTasks(events: EventFrame[]): TaskDerivation {
if (snap !== null) snapshots.push({ seq: e.seq, items: snap });
} else if (name === SUBAGENT_TOOL) {
const parsed = parseJson(e.args);
const o = parsed !== null && typeof parsed === "object" ? (parsed as Record<string, unknown>) : {};
const nameField = o.agentName ?? o.agent ?? o.name ?? o.agentType ?? o.role;
const o =
parsed !== null && typeof parsed === "object"
? (parsed as Record<string, unknown>)
: {};
const nameField =
o.agentName ?? o.agent ?? o.name ?? o.agentType ?? o.role;
subagents.push({
key: e.toolCallId,
name: typeof nameField === "string" && nameField.length > 0 ? nameField : "subagent",
name:
typeof nameField === "string" && nameField.length > 0
? nameField
: "subagent",
running: true,
isError: false,
});
@@ -274,7 +292,11 @@ export function deriveTasks(events: EventFrame[]): TaskDerivation {
const snap = extractSnapshot(parseJson(e.resultPreview));
if (snap !== null) snapshots.push({ seq: e.seq, items: snap });
}
} else if (e.type === "message_end" && e.message?.role === "toolResult" && e.message.toolCallId !== null) {
} else if (
e.type === "message_end" &&
e.message?.role === "toolResult" &&
e.message.toolCallId !== null
) {
const name = toolNames.get(e.message.toolCallId) ?? "";
if (name === TODO_TOOL) {
const snap = extractSnapshot(parseJson(e.message.text));
@@ -290,13 +312,19 @@ export function deriveTasks(events: EventFrame[]): TaskDerivation {
const latest = snapshots[snapshots.length - 1]?.items ?? [];
const seen = new Map<string, TodoStatus>();
for (const snap of snapshots) {
for (const item of snap.items) if (!seen.has(item.content)) seen.set(item.content, item.status);
for (const item of snap.items)
if (!seen.has(item.content)) seen.set(item.content, item.status);
}
for (const item of latest) {
todos.push({ content: item.content, status: item.status, deleted: false });
todos.push({
content: item.content,
status: item.status,
deleted: false,
});
seen.delete(item.content);
}
for (const [content] of seen) todos.push({ content, status: "pending", deleted: true });
for (const [content] of seen)
todos.push({ content, status: "pending", deleted: true });
}
return {
web/src/protocol.ts
+8 -4
View File
@@ -22,10 +22,14 @@ export type EventType = (typeof EventType)[keyof typeof EventType];
/** REST routes (base `/api`, bearer auth). */
export const Route = {
Sessions: "/api/sessions",
SessionEvents: (id: string): string => `/api/sessions/${encodeURIComponent(id)}/events`,
SessionPrompt: (id: string): string => `/api/sessions/${encodeURIComponent(id)}/prompt`,
SessionAbort: (id: string): string => `/api/sessions/${encodeURIComponent(id)}/abort`,
SessionContainer: (id: string): string => `/api/sessions/${encodeURIComponent(id)}/container`,
SessionEvents: (id: string): string =>
`/api/sessions/${encodeURIComponent(id)}/events`,
SessionPrompt: (id: string): string =>
`/api/sessions/${encodeURIComponent(id)}/prompt`,
SessionAbort: (id: string): string =>
`/api/sessions/${encodeURIComponent(id)}/abort`,
SessionContainer: (id: string): string =>
`/api/sessions/${encodeURIComponent(id)}/container`,
Spawn: "/api/spawn",
SpawnStatus: "/api/spawn/status",
GitlabStatus: "/api/gitlab/status",