plugin-apply.spec.ts 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305
  1. import { createServer } from 'node:http'
  2. import type { IncomingMessage, Server, ServerResponse } from 'node:http'
  3. import { mkdtemp, rm } from 'node:fs/promises'
  4. import { join } from 'node:path'
  5. import { tmpdir } from 'node:os'
  6. import { PassThrough, Writable } from 'node:stream'
  7. import { afterEach, describe, expect, it, vi } from 'vitest'
  8. import { Context } from 'cordis'
  9. import * as agentCore from '@deepseek-ai/dsh-agent-spine-demo'
  10. import SessionPersistenceJsonl from '@deepseek-ai/dsh-session-persistence-jsonl'
  11. import * as jsonrpc from '../src/index.ts'
  12. /**
  13. * Mount the real namespace plugin with in-memory stdio and exit seams. Covers
  14. * the full transport/server path, response-before-exit shutdown exactly once,
  15. * and bare-fiber disposal without process exit.
  16. */
  17. /** One ordered frame, write completion, or exit observation. */
  18. type WireEvent =
  19. | { kind: 'frame'; frame: Record<string, unknown> }
  20. | { kind: 'write-complete'; ids: (string | number)[] }
  21. | { kind: 'root-disposed' }
  22. | { kind: 'exit'; code: number }
  23. interface ApplyHarness {
  24. ctx: Context
  25. /** The plugin fiber used by the bare-dispose case. */
  26. fiber: Awaited<ReturnType<Context['plugin']>>
  27. /** Frames, write completions, and exits in observation order. */
  28. events: WireEvent[]
  29. outputErrors: Error[]
  30. send(frame: Record<string, unknown>): void
  31. sendRaw(text: string): void
  32. frames(): Record<string, unknown>[]
  33. exits(): number[]
  34. waitForFrame(predicate: (frame: Record<string, unknown>) => boolean, description: string): Promise<Record<string, unknown>>
  35. dispose(): Promise<void>
  36. }
  37. /** Poll asynchronous output for up to five seconds. */
  38. async function waitFor<T>(get: () => T | undefined, description: string): Promise<T> {
  39. const deadline = Date.now() + 5000
  40. for (;;) {
  41. const value = get()
  42. if (value !== undefined) return value
  43. if (Date.now() > deadline) throw new Error(`timed out waiting for ${description}`)
  44. await new Promise(resolve => setTimeout(resolve, 5))
  45. }
  46. }
  47. /** Drain asynchronous work before a negative assertion. */
  48. async function settle(): Promise<void> {
  49. await new Promise(resolve => setTimeout(resolve, 25))
  50. }
  51. /** Mount the real plugin on a minimal harness with in-memory stdio and exit. */
  52. async function mountPlugin(
  53. storageDir: string,
  54. options: { writeDelayMs?: number; failFlush?: boolean } = {},
  55. ): Promise<ApplyHarness> {
  56. const ctx = new Context()
  57. await ctx.plugin(agentCore, { workspaceContext: false })
  58. await ctx.plugin(SessionPersistenceJsonl, { root: storageDir })
  59. await new Promise(resolve => setTimeout(resolve, 50))
  60. const input = new PassThrough()
  61. const events: WireEvent[] = []
  62. const outputErrors: Error[] = []
  63. let pendingOutput = ''
  64. // Record frame admission separately from write completion so delayed output
  65. // tests the flush barrier.
  66. const output = new Writable({
  67. write(chunk: Buffer, _encoding, callback) {
  68. const ids: (string | number)[] = []
  69. pendingOutput += chunk.toString('utf8')
  70. for (;;) {
  71. const newline = pendingOutput.indexOf('\n')
  72. if (newline < 0) break
  73. const line = pendingOutput.slice(0, newline).trim()
  74. pendingOutput = pendingOutput.slice(newline + 1)
  75. if (line) {
  76. const frame = JSON.parse(line) as Record<string, unknown>
  77. events.push({ kind: 'frame', frame })
  78. if (typeof frame.id === 'string' || typeof frame.id === 'number') ids.push(frame.id)
  79. }
  80. }
  81. const complete = (): void => {
  82. if (options.failFlush === true && chunk.length === 0) {
  83. callback(new Error('flush callback failed'))
  84. return
  85. }
  86. events.push({ kind: 'write-complete', ids })
  87. callback()
  88. }
  89. if ((options.writeDelayMs ?? 0) > 0) setTimeout(complete, options.writeDelayMs)
  90. else complete()
  91. },
  92. })
  93. output.on('error', (error: Error) => { outputErrors.push(error) })
  94. const exit = (code: number): void => { events.push({ kind: 'exit', code }) }
  95. ctx.effect(() => () => { events.push({ kind: 'root-disposed' }) }, 'jsonrpc test root-disposal witness')
  96. const fiber = await ctx.plugin(jsonrpc, { input, output, exit })
  97. const frames = (): Record<string, unknown>[] =>
  98. events.flatMap(event => event.kind === 'frame' ? [event.frame] : [])
  99. return {
  100. ctx,
  101. fiber,
  102. events,
  103. outputErrors,
  104. send: (frame) => { input.write(`${JSON.stringify(frame)}\n`) },
  105. sendRaw: (text) => { input.write(text) },
  106. frames,
  107. exits: () => events.flatMap(event => event.kind === 'exit' ? [event.code] : []),
  108. waitForFrame: (predicate, description) => waitFor(() => frames().find(predicate), description),
  109. dispose: async () => { await ctx.fiber.dispose() },
  110. }
  111. }
  112. const servers: Server[] = []
  113. afterEach(async () => {
  114. await Promise.all(servers.splice(0).map(server => new Promise(resolve => server.close(resolve))))
  115. vi.unstubAllEnvs()
  116. })
  117. /** Keyless SSE endpoint for completing a prompt turn. */
  118. async function mockCompletionServer(): Promise<{ url: string; requests: unknown[] }> {
  119. const requests: unknown[] = []
  120. const server = createServer((request: IncomingMessage, response: ServerResponse) => {
  121. let body = ''
  122. request.on('data', (chunk: Buffer) => { body += chunk.toString('utf8') })
  123. request.on('end', () => {
  124. requests.push(JSON.parse(body))
  125. response.writeHead(200, { 'content-type': 'text/event-stream' })
  126. response.write('data: {"choices":[{"delta":{"role":"assistant","content":null,"reasoning_content":""}}]}\n\n')
  127. response.write('data: {"choices":[{"delta":{"content":"done"}}]}\n\n')
  128. response.write('data: {"choices":[{"delta":{"content":""},"finish_reason":"stop"}],"usage":{"prompt_tokens":3,"completion_tokens":1}}\n\n')
  129. response.write('data: [DONE]\n\n')
  130. response.end()
  131. })
  132. })
  133. servers.push(server)
  134. await new Promise<void>(resolve => server.listen(0, '127.0.0.1', resolve))
  135. const address = server.address()
  136. if (address === null || typeof address === 'string') throw new Error('no port')
  137. return { url: `http://127.0.0.1:${address.port}`, requests }
  138. }
  139. describe('dsh-jsonrpc plugin apply', () => {
  140. it('serves initialize over the injected stdio pair', async () => {
  141. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-init-'))
  142. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  143. const harness = await mountPlugin(storageDir)
  144. try {
  145. harness.send({ jsonrpc: '2.0', id: 'init-1', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'apply-model' } })
  146. const response = await harness.waitForFrame(frame => frame.id === 'init-1', 'initialize response')
  147. expect(response).toEqual({
  148. jsonrpc: '2.0',
  149. id: 'init-1',
  150. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime', version: '0.0.1' } },
  151. })
  152. expect(harness.exits()).toEqual([])
  153. } finally {
  154. await harness.dispose()
  155. await rm(storageDir, { recursive: true, force: true })
  156. }
  157. })
  158. it('drives a session/prompt turn end-to-end and forwards session notifications as output frames', async () => {
  159. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-prompt-'))
  160. const llmServer = await mockCompletionServer()
  161. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  162. vi.stubEnv('DEEPSEEK_BASE_URL', llmServer.url)
  163. const harness = await mountPlugin(storageDir)
  164. try {
  165. harness.send({ jsonrpc: '2.0', id: 1, method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'dsagent-model' } })
  166. await harness.waitForFrame(frame => frame.id === 1, 'initialize response')
  167. harness.send({
  168. jsonrpc: '2.0',
  169. id: 2,
  170. method: 'session/prompt',
  171. params: { sessionId: 'main', contentBlocks: [{ type: 'text', text: 'fix it' }] },
  172. })
  173. const response = await harness.waitForFrame(frame => frame.id === 2, 'prompt response')
  174. expect((response.result as { messageId?: unknown }).messageId).toBeTypeOf('string')
  175. await harness.waitForFrame(
  176. frame => frame.method === 'session.status'
  177. && (frame.params as { status?: string } | undefined)?.status === 'idle',
  178. 'idle session status',
  179. )
  180. expect(llmServer.requests).toHaveLength(1)
  181. const body = llmServer.requests[0] as { model: string; messages: { role: string }[] }
  182. expect(body.model).toBe('dsagent-model')
  183. expect(body.messages.at(-1)?.role).toBe('user')
  184. // Notifications use the same transport and arrive as id-less frames.
  185. const notifications = harness.frames().filter(frame => frame.id === undefined)
  186. expect(notifications.some(frame => frame.method === 'session.event')).toBe(true)
  187. expect(notifications.findLast(frame => frame.method === 'session.status')).toMatchObject({
  188. jsonrpc: '2.0',
  189. params: { sessionId: 'main', status: 'idle' },
  190. })
  191. } finally {
  192. await harness.dispose()
  193. await rm(storageDir, { recursive: true, force: true })
  194. }
  195. })
  196. it('answers shutdown before exiting 0 exactly once, even against a racing second shutdown', async () => {
  197. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-shutdown-'))
  198. const harness = await mountPlugin(storageDir, { writeDelayMs: 10 })
  199. try {
  200. // One chunk makes the two deferred exit callbacks race.
  201. const first = { jsonrpc: '2.0', id: 'sd-1', method: 'shutdown' }
  202. const second = { jsonrpc: '2.0', id: 'sd-2', method: 'shutdown' }
  203. harness.sendRaw(`${JSON.stringify(first)}\n${JSON.stringify(second)}\n`)
  204. await waitFor(() => harness.exits().length > 0 ? true : undefined, 'exit recorder call')
  205. expect(harness.exits()).toEqual([0])
  206. // Both response writes and the flush barrier complete before exit.
  207. const exitIndex = harness.events.findIndex(event => event.kind === 'exit')
  208. const firstResponse = harness.events.findIndex(event => event.kind === 'frame' && event.frame.id === 'sd-1')
  209. const secondResponse = harness.events.findIndex(event => event.kind === 'frame' && event.frame.id === 'sd-2')
  210. const firstComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.includes('sd-1'))
  211. const secondComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.includes('sd-2'))
  212. const flushComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.length === 0)
  213. const rootDisposed = harness.events.findIndex(event => event.kind === 'root-disposed')
  214. expect(firstResponse).toBeGreaterThanOrEqual(0)
  215. expect(secondResponse).toBeGreaterThanOrEqual(0)
  216. expect(firstComplete).toBeGreaterThan(firstResponse)
  217. expect(secondComplete).toBeGreaterThan(secondResponse)
  218. expect(flushComplete).toBeGreaterThan(firstComplete)
  219. expect(flushComplete).toBeGreaterThan(secondComplete)
  220. expect(rootDisposed).toBeGreaterThan(flushComplete)
  221. expect(exitIndex).toBeGreaterThan(rootDisposed)
  222. await settle()
  223. expect(harness.exits()).toEqual([0])
  224. expect(harness.events.filter(event => event.kind === 'root-disposed')).toHaveLength(1)
  225. const before = harness.frames().length
  226. harness.send({ jsonrpc: '2.0', id: 'after-exit', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  227. await settle()
  228. expect(harness.frames().length).toBe(before)
  229. } finally {
  230. await harness.dispose()
  231. await rm(storageDir, { recursive: true, force: true })
  232. }
  233. })
  234. it('still disposes and exits once when the flush callback fails', async () => {
  235. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-flush-failure-'))
  236. const harness = await mountPlugin(storageDir, { failFlush: true })
  237. try {
  238. harness.send({ jsonrpc: '2.0', id: 'sd-fail', method: 'shutdown' })
  239. await waitFor(() => harness.exits().length > 0 ? true : undefined, 'exit after flush failure')
  240. await settle()
  241. expect(harness.exits()).toEqual([0])
  242. expect(harness.events.filter(event => event.kind === 'root-disposed')).toHaveLength(1)
  243. expect(harness.outputErrors.map(error => error.message)).toEqual(['flush callback failed'])
  244. const before = harness.frames().length
  245. harness.send({ jsonrpc: '2.0', id: 'after-flush-failure', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  246. await settle()
  247. expect(harness.frames().length).toBe(before)
  248. } finally {
  249. await harness.dispose()
  250. await rm(storageDir, { recursive: true, force: true })
  251. }
  252. })
  253. it('stops serving on a bare fiber dispose (HMR-style unload) without calling exit', async () => {
  254. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-dispose-'))
  255. const harness = await mountPlugin(storageDir)
  256. try {
  257. // Prove the handler-rejection path is live before disposal.
  258. harness.send({ jsonrpc: '2.0', id: 'probe-1', method: 'nope/unknown' })
  259. const error = await harness.waitForFrame(frame => frame.id === 'probe-1', 'error response for unknown method')
  260. expect(error.error).toMatchObject({
  261. code: -32603,
  262. message: 'unknown DeepSeek Harness SDK runtime method: nope/unknown',
  263. })
  264. await harness.fiber.dispose()
  265. expect(harness.events.some(event => event.kind === 'root-disposed')).toBe(false)
  266. const before = harness.frames().length
  267. harness.send({ jsonrpc: '2.0', id: 'probe-2', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  268. await settle()
  269. expect(harness.frames().length).toBe(before)
  270. expect(harness.exits()).toEqual([])
  271. } finally {
  272. await harness.dispose()
  273. await rm(storageDir, { recursive: true, force: true })
  274. }
  275. })
  276. })