test: add llm.test.ts (#11375)

packages/opencode/test/session/llm.test.ts

@@ -1,6 +1,15 @@
-import { describe, expect, test } from "bun:test"
-import { LLM } from "../../src/session/llm"
+import { afterAll, beforeAll, beforeEach, describe, expect, test } from "bun:test"
+import path from "path"
 import type { ModelMessage } from "ai"
+import { LLM } from "../../src/session/llm"
+import { Global } from "../../src/global"
+import { Instance } from "../../src/project/instance"
+import { Provider } from "../../src/provider/provider"
+import { ProviderTransform } from "../../src/provider/transform"
+import { ModelsDev } from "../../src/provider/models"
+import { tmpdir } from "../fixture/fixture"
+import type { Agent } from "../../src/agent/agent"
+import type { MessageV2 } from "../../src/session/message-v2"
 
 describe("session.llm.hasToolCalls", () => {
   test("returns false for empty messages array", () => {
@@ -88,3 +97,609 @@ describe("session.llm.hasToolCalls", () => {
     expect(LLM.hasToolCalls(messages)).toBe(true)
   })
 })
+
+type Capture = {
+  url: URL
+  headers: Headers
+  body: Record<string, unknown>
+}
+
+const state = {
+  server: null as ReturnType<typeof Bun.serve> | null,
+  queue: [] as Array<{ path: string; response: Response; resolve: (value: Capture) => void }>,
+}
+
+function deferred<T>() {
+  const result = {} as { promise: Promise<T>; resolve: (value: T) => void }
+  result.promise = new Promise((resolve) => {
+    result.resolve = resolve
+  })
+  return result
+}
+
+function waitRequest(pathname: string, response: Response) {
+  const pending = deferred<Capture>()
+  state.queue.push({ path: pathname, response, resolve: pending.resolve })
+  return pending.promise
+}
+
+beforeAll(() => {
+  state.server = Bun.serve({
+    port: 0,
+    async fetch(req) {
+      const next = state.queue.shift()
+      if (!next) {
+        return new Response("unexpected request", { status: 500 })
+      }
+
+      const url = new URL(req.url)
+      const body = (await req.json()) as Record<string, unknown>
+      next.resolve({ url, headers: req.headers, body })
+
+      if (!url.pathname.endsWith(next.path)) {
+        return new Response("not found", { status: 404 })
+      }
+
+      return next.response
+    },
+  })
+})
+
+beforeEach(() => {
+  state.queue.length = 0
+})
+
+afterAll(() => {
+  state.server?.stop()
+})
+
+function createChatStream(text: string) {
+  const payload =
+    [
+      `data: ${JSON.stringify({
+        id: "chatcmpl-1",
+        object: "chat.completion.chunk",
+        choices: [{ delta: { role: "assistant" } }],
+      })}`,
+      `data: ${JSON.stringify({
+        id: "chatcmpl-1",
+        object: "chat.completion.chunk",
+        choices: [{ delta: { content: text } }],
+      })}`,
+      `data: ${JSON.stringify({
+        id: "chatcmpl-1",
+        object: "chat.completion.chunk",
+        choices: [{ delta: {}, finish_reason: "stop" }],
+      })}`,
+      "data: [DONE]",
+    ].join("\n\n") + "\n\n"
+
+  const encoder = new TextEncoder()
+  return new ReadableStream<Uint8Array>({
+    start(controller) {
+      controller.enqueue(encoder.encode(payload))
+      controller.close()
+    },
+  })
+}
+
+async function loadFixture(providerID: string, modelID: string) {
+  const fixturePath = path.join(import.meta.dir, "../tool/fixtures/models-api.json")
+  const data = (await Bun.file(fixturePath).json()) as Record<string, ModelsDev.Provider>
+  const provider = data[providerID]
+  if (!provider) {
+    throw new Error(`Missing provider in fixture: ${providerID}`)
+  }
+  const model = provider.models[modelID]
+  if (!model) {
+    throw new Error(`Missing model in fixture: ${modelID}`)
+  }
+  return { provider, model }
+}
+
+async function writeModels(models: Record<string, ModelsDev.Provider>) {
+  const modelsPath = path.join(Global.Path.cache, "models.json")
+  await Bun.write(modelsPath, JSON.stringify(models))
+  ModelsDev.Data.reset()
+}
+
+function createEventStream(chunks: unknown[], includeDone = false) {
+  const lines = chunks.map((chunk) => `data: ${typeof chunk === "string" ? chunk : JSON.stringify(chunk)}`)
+  if (includeDone) {
+    lines.push("data: [DONE]")
+  }
+  const payload = lines.join("\n\n") + "\n\n"
+  const encoder = new TextEncoder()
+  return new ReadableStream<Uint8Array>({
+    start(controller) {
+      controller.enqueue(encoder.encode(payload))
+      controller.close()
+    },
+  })
+}
+
+function createEventResponse(chunks: unknown[], includeDone = false) {
+  return new Response(createEventStream(chunks, includeDone), {
+    status: 200,
+    headers: { "Content-Type": "text/event-stream" },
+  })
+}
+
+describe("session.llm.stream", () => {
+  test("sends temperature, tokens, and reasoning options for openai-compatible models", async () => {
+    const server = state.server
+    if (!server) {
+      throw new Error("Server not initialized")
+    }
+
+    const providerID = "alibaba"
+    const modelID = "qwen-plus"
+    const fixture = await loadFixture(providerID, modelID)
+    const provider = fixture.provider
+    const model = fixture.model
+
+    const request = waitRequest(
+      "/chat/completions",
+      new Response(createChatStream("Hello"), {
+        status: 200,
+        headers: { "Content-Type": "text/event-stream" },
+      }),
+    )
+
+    await writeModels({ [providerID]: provider })
+
+    await using tmp = await tmpdir({
+      init: async (dir) => {
+        await Bun.write(
+          path.join(dir, "opencode.json"),
+          JSON.stringify({
+            $schema: "https://opencode.ai/config.json",
+            enabled_providers: [providerID],
+            provider: {
+              [providerID]: {
+                options: {
+                  apiKey: "test-key",
+                  baseURL: `${server.url.origin}/v1`,
+                },
+              },
+            },
+          }),
+        )
+      },
+    })
+
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const resolved = await Provider.getModel(providerID, model.id)
+        const sessionID = "session-test-1"
+        const agent = {
+          name: "test",
+          mode: "primary",
+          options: {},
+          permission: [{ permission: "*", pattern: "*", action: "allow" }],
+          temperature: 0.4,
+          topP: 0.8,
+        } satisfies Agent.Info
+
+        const user = {
+          id: "user-1",
+          sessionID,
+          role: "user",
+          time: { created: Date.now() },
+          agent: agent.name,
+          model: { providerID, modelID: resolved.id },
+          variant: "high",
+        } satisfies MessageV2.User
+
+        const stream = await LLM.stream({
+          user,
+          sessionID,
+          model: resolved,
+          agent,
+          system: ["You are a helpful assistant."],
+          abort: new AbortController().signal,
+          messages: [{ role: "user", content: "Hello" }],
+          tools: {},
+        })
+
+        for await (const _ of stream.fullStream) {
+        }
+
+        const capture = await request
+        const body = capture.body
+        const headers = capture.headers
+        const url = capture.url
+
+        expect(url.pathname.startsWith("/v1/")).toBe(true)
+        expect(url.pathname.endsWith("/chat/completions")).toBe(true)
+        expect(headers.get("Authorization")).toBe("Bearer test-key")
+        expect(headers.get("User-Agent") ?? "").toMatch(/^opencode\//)
+
+        expect(body.model).toBe(resolved.api.id)
+        expect(body.temperature).toBe(0.4)
+        expect(body.top_p).toBe(0.8)
+        expect(body.stream).toBe(true)
+
+        const maxTokens = (body.max_tokens as number | undefined) ?? (body.max_output_tokens as number | undefined)
+        const expectedMaxTokens = ProviderTransform.maxOutputTokens(
+          resolved.api.npm,
+          ProviderTransform.options({ model: resolved, sessionID }),
+          resolved.limit.output,
+          LLM.OUTPUT_TOKEN_MAX,
+        )
+        expect(maxTokens).toBe(expectedMaxTokens)
+
+        const reasoning = (body.reasoningEffort as string | undefined) ?? (body.reasoning_effort as string | undefined)
+        expect(reasoning).toBe("high")
+      },
+    })
+  })
+
+  test("sends responses API payload for OpenAI models", async () => {
+    const server = state.server
+    if (!server) {
+      throw new Error("Server not initialized")
+    }
+
+    const source = await loadFixture("github-copilot", "gpt-5.1")
+    const model = source.model
+
+    const responseChunks = [
+      {
+        type: "response.created",
+        response: {
+          id: "resp-1",
+          created_at: Math.floor(Date.now() / 1000),
+          model: model.id,
+          service_tier: null,
+        },
+      },
+      {
+        type: "response.output_text.delta",
+        item_id: "item-1",
+        delta: "Hello",
+        logprobs: null,
+      },
+      {
+        type: "response.completed",
+        response: {
+          incomplete_details: null,
+          usage: {
+            input_tokens: 1,
+            input_tokens_details: null,
+            output_tokens: 1,
+            output_tokens_details: null,
+          },
+          service_tier: null,
+        },
+      },
+    ]
+    const request = waitRequest("/responses", createEventResponse(responseChunks, true))
+
+    await writeModels({})
+
+    await using tmp = await tmpdir({
+      init: async (dir) => {
+        await Bun.write(
+          path.join(dir, "opencode.json"),
+          JSON.stringify({
+            $schema: "https://opencode.ai/config.json",
+            enabled_providers: ["openai"],
+            provider: {
+              openai: {
+                name: "OpenAI",
+                env: ["OPENAI_API_KEY"],
+                npm: "@ai-sdk/openai",
+                api: "https://api.openai.com/v1",
+                models: {
+                  [model.id]: model,
+                },
+                options: {
+                  apiKey: "test-openai-key",
+                  baseURL: `${server.url.origin}/v1`,
+                },
+              },
+            },
+          }),
+        )
+      },
+    })
+
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const resolved = await Provider.getModel("openai", model.id)
+        const sessionID = "session-test-2"
+        const agent = {
+          name: "test",
+          mode: "primary",
+          options: {},
+          permission: [{ permission: "*", pattern: "*", action: "allow" }],
+          temperature: 0.2,
+        } satisfies Agent.Info
+
+        const user = {
+          id: "user-2",
+          sessionID,
+          role: "user",
+          time: { created: Date.now() },
+          agent: agent.name,
+          model: { providerID: "openai", modelID: resolved.id },
+          variant: "high",
+        } satisfies MessageV2.User
+
+        const stream = await LLM.stream({
+          user,
+          sessionID,
+          model: resolved,
+          agent,
+          system: ["You are a helpful assistant."],
+          abort: new AbortController().signal,
+          messages: [{ role: "user", content: "Hello" }],
+          tools: {},
+        })
+
+        for await (const _ of stream.fullStream) {
+        }
+
+        const capture = await request
+        const body = capture.body
+
+        expect(capture.url.pathname.endsWith("/responses")).toBe(true)
+        expect(body.model).toBe(resolved.api.id)
+        expect(body.stream).toBe(true)
+        expect((body.reasoning as { effort?: string } | undefined)?.effort).toBe("high")
+
+        const maxTokens = body.max_output_tokens as number | undefined
+        const expectedMaxTokens = ProviderTransform.maxOutputTokens(
+          resolved.api.npm,
+          ProviderTransform.options({ model: resolved, sessionID }),
+          resolved.limit.output,
+          LLM.OUTPUT_TOKEN_MAX,
+        )
+        expect(maxTokens).toBe(expectedMaxTokens)
+      },
+    })
+  })
+
+  test("sends messages API payload for Anthropic models", async () => {
+    const server = state.server
+    if (!server) {
+      throw new Error("Server not initialized")
+    }
+
+    const providerID = "anthropic"
+    const modelID = "claude-3-5-sonnet-20241022"
+    const fixture = await loadFixture(providerID, modelID)
+    const provider = fixture.provider
+    const model = fixture.model
+
+    const chunks = [
+      {
+        type: "message_start",
+        message: {
+          id: "msg-1",
+          model: model.id,
+          usage: {
+            input_tokens: 3,
+            cache_creation_input_tokens: null,
+            cache_read_input_tokens: null,
+          },
+        },
+      },
+      {
+        type: "content_block_start",
+        index: 0,
+        content_block: { type: "text", text: "" },
+      },
+      {
+        type: "content_block_delta",
+        index: 0,
+        delta: { type: "text_delta", text: "Hello" },
+      },
+      { type: "content_block_stop", index: 0 },
+      {
+        type: "message_delta",
+        delta: { stop_reason: "end_turn", stop_sequence: null, container: null },
+        usage: {
+          input_tokens: 3,
+          output_tokens: 2,
+          cache_creation_input_tokens: null,
+          cache_read_input_tokens: null,
+        },
+      },
+      { type: "message_stop" },
+    ]
+    const request = waitRequest("/messages", createEventResponse(chunks))
+
+    await writeModels({ [providerID]: provider })
+
+    await using tmp = await tmpdir({
+      init: async (dir) => {
+        await Bun.write(
+          path.join(dir, "opencode.json"),
+          JSON.stringify({
+            $schema: "https://opencode.ai/config.json",
+            enabled_providers: [providerID],
+            provider: {
+              [providerID]: {
+                options: {
+                  apiKey: "test-anthropic-key",
+                  baseURL: `${server.url.origin}/v1`,
+                },
+              },
+            },
+          }),
+        )
+      },
+    })
+
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const resolved = await Provider.getModel(providerID, model.id)
+        const sessionID = "session-test-3"
+        const agent = {
+          name: "test",
+          mode: "primary",
+          options: {},
+          permission: [{ permission: "*", pattern: "*", action: "allow" }],
+          temperature: 0.4,
+          topP: 0.9,
+        } satisfies Agent.Info
+
+        const user = {
+          id: "user-3",
+          sessionID,
+          role: "user",
+          time: { created: Date.now() },
+          agent: agent.name,
+          model: { providerID, modelID: resolved.id },
+        } satisfies MessageV2.User
+
+        const stream = await LLM.stream({
+          user,
+          sessionID,
+          model: resolved,
+          agent,
+          system: ["You are a helpful assistant."],
+          abort: new AbortController().signal,
+          messages: [{ role: "user", content: "Hello" }],
+          tools: {},
+        })
+
+        for await (const _ of stream.fullStream) {
+        }
+
+        const capture = await request
+        const body = capture.body
+
+        expect(capture.url.pathname.endsWith("/messages")).toBe(true)
+        expect(body.model).toBe(resolved.api.id)
+        expect(body.max_tokens).toBe(
+          ProviderTransform.maxOutputTokens(
+            resolved.api.npm,
+            ProviderTransform.options({ model: resolved, sessionID }),
+            resolved.limit.output,
+            LLM.OUTPUT_TOKEN_MAX,
+          ),
+        )
+        expect(body.temperature).toBe(0.4)
+        expect(body.top_p).toBe(0.9)
+      },
+    })
+  })
+
+  test("sends Google API payload for Gemini models", async () => {
+    const server = state.server
+    if (!server) {
+      throw new Error("Server not initialized")
+    }
+
+    const providerID = "google"
+    const modelID = "gemini-2.5-flash"
+    const fixture = await loadFixture(providerID, modelID)
+    const provider = fixture.provider
+    const model = fixture.model
+    const pathSuffix = `/v1beta/models/${model.id}:streamGenerateContent`
+
+    const chunks = [
+      {
+        candidates: [
+          {
+            content: {
+              parts: [{ text: "Hello" }],
+            },
+            finishReason: "STOP",
+          },
+        ],
+        usageMetadata: {
+          promptTokenCount: 1,
+          candidatesTokenCount: 1,
+          totalTokenCount: 2,
+        },
+      },
+    ]
+    const request = waitRequest(pathSuffix, createEventResponse(chunks))
+
+    await writeModels({ [providerID]: provider })
+
+    await using tmp = await tmpdir({
+      init: async (dir) => {
+        await Bun.write(
+          path.join(dir, "opencode.json"),
+          JSON.stringify({
+            $schema: "https://opencode.ai/config.json",
+            enabled_providers: [providerID],
+            provider: {
+              [providerID]: {
+                options: {
+                  apiKey: "test-google-key",
+                  baseURL: `${server.url.origin}/v1beta`,
+                },
+              },
+            },
+          }),
+        )
+      },
+    })
+
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const resolved = await Provider.getModel(providerID, model.id)
+        const sessionID = "session-test-4"
+        const agent = {
+          name: "test",
+          mode: "primary",
+          options: {},
+          permission: [{ permission: "*", pattern: "*", action: "allow" }],
+          temperature: 0.3,
+          topP: 0.8,
+        } satisfies Agent.Info
+
+        const user = {
+          id: "user-4",
+          sessionID,
+          role: "user",
+          time: { created: Date.now() },
+          agent: agent.name,
+          model: { providerID, modelID: resolved.id },
+        } satisfies MessageV2.User
+
+        const stream = await LLM.stream({
+          user,
+          sessionID,
+          model: resolved,
+          agent,
+          system: ["You are a helpful assistant."],
+          abort: new AbortController().signal,
+          messages: [{ role: "user", content: "Hello" }],
+          tools: {},
+        })
+
+        for await (const _ of stream.fullStream) {
+        }
+
+        const capture = await request
+        const body = capture.body
+        const config = body.generationConfig as
+          | { temperature?: number; topP?: number; maxOutputTokens?: number }
+          | undefined
+
+        expect(capture.url.pathname).toBe(pathSuffix)
+        expect(config?.temperature).toBe(0.3)
+        expect(config?.topP).toBe(0.8)
+        expect(config?.maxOutputTokens).toBe(
+          ProviderTransform.maxOutputTokens(
+            resolved.api.npm,
+            ProviderTransform.options({ model: resolved, sessionID }),
+            resolved.limit.output,
+            LLM.OUTPUT_TOKEN_MAX,
+          ),
+        )
+      },
+    })
+  })
+})