Files
reverse-engineering-anything/tests/boundary/mcp/commandObservationViews.test.ts
T
morluto 858ff73880 perf(tests): give MCP test clients the shared lazy schema validation (#10960)
* perf(tests): give MCP test clients the shared lazy schema validation

The MCP SDK client compiles every listed tool's output schema on its
first tool call. With the SDK's default Ajv options, REA's catalog costs
gigabytes of allocation per client (inspect_analysis_view alone: 1.7 s and
~500 MiB retained), so native MCP files peaked at 1-2 GiB per worker.

tests/fixtures/mcpSchemaValidation already owns lazy, reference-preserving
validation. Route the native MCP, process-boundary and shared harness
clients through it, share compiled validators across clients in a worker
(bounded LRU keyed by schema text), and add a counted architecture guard so
default-validation test clients cannot spread; the remaining copies are the
baseline.

* fix(guards): read call arguments past parentheses inside string literals

* perf(tests): give MCP boundary test clients the shared schema validation (#11022)

* perf(tests): give MCP boundary test clients the shared schema validation

* perf(tests): give the remaining MCP test clients the shared schema validation (#11025)

* fix(guards): read an MCP client's validator option from parsed code

The default-validation counter searched each new Client(...) call's text for jsonSchemaValidator, so an inline comment, a string literal or a client name naming the option hid a default client. Parse the source with @babel/parser and count NewExpressions of Client whose object-literal arguments have no jsonSchemaValidator property; a source that does not parse counts every textual call. The guard reads raw test sources, and the generated-child-script allowlist entry, which no longer counts as a call, is removed.

* fix(guards): read an MCP client's options only from a whole object-literal argument

The default-validation counter skipped files without the text
'new Client(', so a comment between the tokens hid a client, and it took
any depth-two key as an option, so 'enabled ? { jsonSchemaValidator } : {}'
passed while its false branch uses the SDK default. The skip now only
requires the name Client, and the validator counts only when it is a key
of an object literal that is a whole argument; anything else counts as the
default.
2026-11-15 21:53:53 -05:00

333 lines
10 KiB
TypeScript

import { unknownTarget } from "../../../src/domain/targetIdentity.js";
import { Client, InMemoryTransport } from "@modelcontextprotocol/client";
import pino from "pino";
import { expect, it, onTestFinished, vi } from "vitest";
import { EvidenceMcpServer } from "../../../src/server/EvidenceMcpServer.js";
import {
registerServiceTools,
type GdbToolOptions,
} from "../../../src/server/serviceToolBindings.js";
import { createToolResultDelivery } from "../../../src/server/toolResult.js";
import {
createEvidence,
parseEvidence,
type Evidence,
} from "../../../src/domain/evidence.js";
import { ok } from "../../../src/domain/result.js";
import { inspectAnalysisView } from "../../../src/application/analysisView/AnalysisViewService.js";
import { TestReverseEngineeringService } from "../../support/reverseEngineeringService.js";
import { ownedOutput } from "../../support/reverseEngineeringCompositionFixtures.js";
import { advertisedMcpToolSchemas } from "../../fixtures/advertisedMcpTool.js";
import { createMcpSchemaValidation } from "../../fixtures/mcpSchemaValidation.js";
const sessionId = "00000000-0000-4000-8000-000000000001";
const capture = {
command: "show architecture",
mi: "7^done",
records: [
'~"architecture\\n"',
'=thread-created,id="1"',
"unrecognized 🧪 record",
"7^done",
],
console: '~"architecture\\n"',
target: "",
log: "",
process_exit_observed: false,
process_exit_code: null,
process_exit_signal: null,
output_truncated: true,
output_invalid_utf8: true as const,
completion_status: "unknown" as const,
};
const gdbEvidence = (
parameters = { session_id: sessionId, command: capture.command },
) =>
createEvidence(
unknownTarget("not-observed"),
{ id: "gdb", name: "GDB", version: null },
{
operation: "gdb_console",
parameters,
result: capture,
rawResult: capture,
limitations: ["Capture is partial."],
},
);
const connect = async (retain = true) => {
const ledger = new Map<string, Evidence>();
const record = (evidence: Evidence) => {
ledger.set(evidence.evidence_id, evidence);
return ok("added" as const);
};
const lookup = (id: string) => ledger.get(id);
const delivery = createToolResultDelivery(2 * 1024 * 1024);
const server = new EvidenceMcpServer(
{ name: "command-views", version: "1" },
{},
retain ? record : undefined,
delivery,
);
const execute = vi.fn(async () =>
ok({ result: capture, evidence: gdbEvidence() }),
);
const manager: GdbToolOptions["manager"] = {
start: async () =>
ok({ session_id: sessionId, mi_version: "mi3" as const }),
execute,
status: () => undefined,
close: async () => ok({ target_state: "none_observed" as const }),
retryCleanupFailures: async () => ok(null),
};
const run = vi.fn(async () =>
ownedOutput("BFD advertised targets\n", "diagnostic\n"),
);
const service = new TestReverseEngineeringService({ runOwned: run });
const logger = pino({ enabled: false });
registerServiceTools(
server,
{
analysisView: {},
gdb: { manager, session: undefined },
reverseEngineering: { service, session: undefined },
},
{
logger,
evidenceById: retain ? lookup : undefined,
recordEvidence: retain ? record : undefined,
},
);
const client = new Client(
{ name: "command-view-client", version: "1" },
{ jsonSchemaValidator: createMcpSchemaValidation() },
);
onTestFinished(async () => {
await client.close();
await server.close();
await service.close();
});
const [ct, st] = InMemoryTransport.createLinkedPair();
await server.connect(st);
await client.connect(ct);
return { client, ledger, execute, run };
};
it("delivers ordered GDB records once and queries their retained parent after close through advertised SDK schemas", async () => {
const { client, ledger } = await connect();
const schemas = await advertisedMcpToolSchemas(client, "gdb_console");
const viewSchemas = await advertisedMcpToolSchemas(
client,
"inspect_analysis_view",
);
const input = { session_id: sessionId, command: capture.command };
const full = await client.callTool(
{ name: "gdb_console", arguments: input },
{ toolDefinition: schemas.toolDefinition },
);
const summary = await client.callTool(
{ name: "gdb_console", arguments: { ...input, detail: "summary" } },
{ toolDefinition: schemas.toolDefinition },
);
for (const reply of [full, summary]) {
expect(reply.isError).not.toBe(true);
}
expect(parseEvidence(full.structuredContent)).toEqual(gdbEvidence());
const selected = parseEvidence(summary.structuredContent);
expect(selected.raw_result).toBeNull();
expect(selected.subject.identity.kind).toBe("unknown");
expect(selected.normalized_result).toMatchObject({
parent_evidence_id: gdbEvidence().evidence_id,
artifact: null,
source_provider: gdbEvidence().provider,
source_parameters: gdbEvidence().parameters,
summary: {
records: capture.records,
mi: capture.mi,
completion_status: "unknown",
output_truncated: true,
output_invalid_utf8: true,
process_exit_code: null,
available_counts: {
records: 4,
console_utf16_units: capture.console.length,
target_utf16_units: 0,
log_utf16_units: 0,
},
},
});
expect(selected.normalized_result).not.toHaveProperty("summary.console");
expect(selected.normalized_result).not.toHaveProperty("summary.target");
expect(selected.normalized_result).not.toHaveProperty("summary.log");
await client.callTool({
name: "close_gdb_session",
arguments: { session_id: sessionId },
});
const queried = await client.callTool(
{
name: "inspect_analysis_view",
arguments: {
source: {
kind: "retained-evidence",
evidence_id: gdbEvidence().evidence_id,
},
view: { kind: "summary" },
},
},
{ toolDefinition: viewSchemas.toolDefinition },
);
expect(parseEvidence(queried.structuredContent)).toEqual(selected);
expect(ledger.get(gdbEvidence().evidence_id)).toEqual(gdbEvidence());
});
it("keeps pathless objdump observations subjectless and preserves the default complete capture", async () => {
const { client, ledger } = await connect();
const schemas = await advertisedMcpToolSchemas(
client,
"inspect_with_objdump",
);
expect(
schemas.ajv.validate(schemas.inputSchema, {
operation: "architectures",
detail: "summary",
}),
).toBe(true);
const full = await client.callTool(
{ name: "inspect_with_objdump", arguments: { operation: "architectures" } },
{ toolDefinition: schemas.toolDefinition },
);
const selected = await client.callTool(
{
name: "inspect_with_objdump",
arguments: { operation: "architectures", detail: "summary" },
},
{ toolDefinition: schemas.toolDefinition },
);
const parent = parseEvidence(full.structuredContent);
const summary = parseEvidence(selected.structuredContent);
expect(summary.normalized_result).toMatchObject({
parent_evidence_id: parent.evidence_id,
artifact: null,
source_parameters: { operation: "architectures" },
summary: parent.normalized_result,
});
expect(summary.raw_result).toBeNull();
expect(summary.subject.identity.kind).toBe("unknown");
expect(ledger.get(parent.evidence_id)).toEqual(parent);
});
it("rejects summary without retention before executing either native command", async () => {
const { client, execute, run } = await connect(false);
for (const input of [
{
name: "gdb_console",
arguments: {
session_id: sessionId,
command: capture.command,
detail: "summary",
},
},
{
name: "inspect_with_objdump",
arguments: { operation: "architectures", detail: "summary" },
},
])
expect((await client.callTool(input)).isError).toBe(true);
expect(execute).not.toHaveBeenCalled();
expect(run).not.toHaveBeenCalled();
});
it("rejects malformed historical GDB facts and conflicting source commands", async () => {
const source = gdbEvidence({
session_id: sessionId,
command: "a different command",
});
expect(
(
await inspectAnalysisView({
source: { kind: "inline", evidence: source },
view: { kind: "summary" },
})
).ok,
).toBe(false);
const malformed = createEvidence(
unknownTarget("not-observed"),
source.provider,
{
operation: "gdb_console",
parameters: {},
result: { ...capture, records: [123] },
},
);
expect(
(
await inspectAnalysisView({
source: { kind: "inline", evidence: malformed },
view: { kind: "summary" },
})
).ok,
).toBe(false);
});
it.each([true, false, null])(
"preserves opaque objdump capture and unknown/change facts (%s)",
async (changed) => {
const report = {
stdout: "opaque output 🧪\n",
stderr: "diagnostic",
exit_code: "ETIMEDOUT",
signal: "SIGTERM",
output_truncated: true,
completion_status: "unknown",
artifact_sha256_after:
changed === null ? null : (changed ? "b" : "a").repeat(64),
artifact_changed: changed,
};
const parent = createEvidence(
{
path: "/artifacts/input",
sha256: "a".repeat(64),
format: "elf",
architecture: "x86_64",
},
{ id: "gnu.objdump", name: "GNU objdump", version: null },
{
operation: "inspect_with_objdump",
parameters: { path: "/artifacts/input", operation: "symbols" },
result: report,
rawResult: report,
},
);
const selected = await inspectAnalysisView({
source: { kind: "inline", evidence: parent },
view: { kind: "summary" },
});
if (!selected.ok) throw selected.error;
expect(selected.value.normalized_result).toMatchObject({ summary: report });
if (changed !== null) {
const conflicting = createEvidence(
{
path: "/artifacts/input",
sha256: "a".repeat(64),
format: "elf",
architecture: "x86_64",
},
parent.provider,
{
operation: parent.operation,
parameters: parent.parameters,
result: { ...report, artifact_changed: !changed },
},
);
expect(
(
await inspectAnalysisView({
source: { kind: "inline", evidence: conflicting },
view: { kind: "summary" },
})
).ok,
).toBe(false);
}
},
);