opencode/packages/llm/test/provider/openai-chat.test.ts

import { describe, expect } from "bun:test"
import { Effect, Schema, Stream } from "effect"
import { HttpClientRequest } from "effect/unstable/http"
import { LLM, LLMError } from "../../src"
import * as Azure from "../../src/providers/azure"
import * as OpenAI from "../../src/providers/openai"
import * as OpenAIChat from "../../src/protocols/openai-chat"
import { LLMClient } from "../../src/route"
import { it } from "../lib/effect"
import { dynamicResponse, fixedResponse, truncatedStream } from "../lib/http"
import { deltaChunk, usageChunk } from "../lib/openai-chunks"
import { sseEvents } from "../lib/sse"

const TargetJson = Schema.fromJsonString(Schema.Unknown)
const encodeJson = Schema.encodeSync(TargetJson)
const decodeJson = Schema.decodeUnknownSync(TargetJson)

const model = OpenAIChat.model({
  id: "gpt-4o-mini",
  baseURL: "https://api.openai.test/v1/",
  headers: { authorization: "Bearer test" },
})

const request = LLM.request({
  id: "req_1",
  model,
  system: "You are concise.",
  prompt: "Say hello.",
  generation: { maxTokens: 20, temperature: 0 },
})

describe("OpenAI Chat route", () => {
  it.effect("prepares OpenAI Chat payload", () =>
    Effect.gen(function* () {
      // Pass the OpenAIChat payload type so `prepared.body` is statically
      // typed to the route's native shape — the assertions below read field
      // names without `unknown` casts.
      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(request)
      const _typed: { readonly model: string; readonly stream: true } = prepared.body

      expect(prepared.body).toEqual({
        model: "gpt-4o-mini",
        messages: [
          { role: "system", content: "You are concise." },
          { role: "user", content: "Say hello." },
        ],
        stream: true,
        stream_options: { include_usage: true },
        max_tokens: 20,
        temperature: 0,
      })
    }),
  )

  it.effect("maps OpenAI provider options to Chat options", () =>
    Effect.gen(function* () {
      const prepared = yield* LLMClient.prepare<OpenAIChat.OpenAIChatBody>(
        LLM.request({
          model: OpenAI.chat("gpt-4o-mini", { baseURL: "https://api.openai.test/v1/" }),
          prompt: "think",
          providerOptions: { openai: { reasoningEffort: "low" } },
        }),
      )

      expect(prepared.body.store).toBe(false)
      expect(prepared.body.reasoning_effort).toBe("low")
    }),
  )

  it.effect("adds native query params to the Chat Completions URL", () =>
    LLMClient.generate(
      LLM.updateRequest(request, { model: OpenAIChat.model({ ...model, queryParams: { "api-version": "v1" } }) }),
    ).pipe(
      Effect.provide(
        dynamicResponse((input) =>
          Effect.gen(function* () {
            const web = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
            expect(web.url).toBe("https://api.openai.test/v1/chat/completions?api-version=v1")
            return input.respond(sseEvents(deltaChunk({}, "stop")), {
              headers: { "content-type": "text/event-stream" },
            })
          }),
        ),
      ),
    ),
  )

  it.effect("uses Azure api-key header for static OpenAI Chat keys", () =>
    LLMClient.generate(
      LLM.updateRequest(request, {
        model: Azure.chat("gpt-4o-mini", {
          baseURL: "https://opencode-test.openai.azure.com/openai/v1/",
          apiKey: "azure-key",
          headers: { authorization: "Bearer stale" },
        }),
      }),
    ).pipe(
      Effect.provide(
        dynamicResponse((input) =>
          Effect.gen(function* () {
            const web = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
            expect(web.headers.get("api-key")).toBe("azure-key")
            expect(web.headers.get("authorization")).toBeNull()
            return input.respond(sseEvents(deltaChunk({}, "stop")), {
              headers: { "content-type": "text/event-stream" },
            })
          }),
        ),
      ),
    ),
  )

  it.effect("applies serializable HTTP overlays after payload lowering", () =>
    LLMClient.generate(
      LLM.updateRequest(request, {
        model: OpenAIChat.model({ ...model, apiKey: "fresh-key", headers: { authorization: "Bearer stale" } }),
        http: {
          body: { metadata: { source: "test" } },
          headers: { authorization: "Bearer request", "x-custom": "yes" },
          query: { debug: "1" },
        },
      }),
    ).pipe(
      Effect.provide(
        dynamicResponse((input) =>
          Effect.gen(function* () {
            const web = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
            expect(web.url).toBe("https://api.openai.test/v1/chat/completions?debug=1")
            expect(web.headers.get("authorization")).toBe("Bearer fresh-key")
            expect(web.headers.get("x-custom")).toBe("yes")
            expect(decodeJson(input.text)).toMatchObject({
              stream: true,
              stream_options: { include_usage: true },
              metadata: { source: "test" },
            })
            return input.respond(sseEvents(deltaChunk({}, "stop")), {
              headers: { "content-type": "text/event-stream" },
            })
          }),
        ),
      ),
    ),
  )

  it.effect("prepares assistant tool-call and tool-result messages", () =>
    Effect.gen(function* () {
      const prepared = yield* LLMClient.prepare(
        LLM.request({
          id: "req_tool_result",
          model,
          messages: [
            LLM.user("What is the weather?"),
            LLM.assistant([LLM.toolCall({ id: "call_1", name: "lookup", input: { query: "weather" } })]),
            LLM.toolMessage({ id: "call_1", name: "lookup", result: { forecast: "sunny" } }),
          ],
        }),
      )

      expect(prepared.body).toEqual({
        model: "gpt-4o-mini",
        messages: [
          { role: "user", content: "What is the weather?" },
          {
            role: "assistant",
            content: null,
            tool_calls: [
              {
                id: "call_1",
                type: "function",
                function: { name: "lookup", arguments: encodeJson({ query: "weather" }) },
              },
            ],
          },
          { role: "tool", tool_call_id: "call_1", content: encodeJson({ forecast: "sunny" }) },
        ],
        stream: true,
        stream_options: { include_usage: true },
      })
    }),
  )

  it.effect("rejects unsupported user media content", () =>
    Effect.gen(function* () {
      const error = yield* LLMClient.prepare(
        LLM.request({
          id: "req_media",
          model,
          messages: [LLM.user({ type: "media", mediaType: "image/png", data: "AAECAw==" })],
        }),
      ).pipe(Effect.flip)

      expect(error.message).toContain("OpenAI Chat user messages only support text content for now")
    }),
  )

  it.effect("rejects unsupported assistant reasoning content", () =>
    Effect.gen(function* () {
      const error = yield* LLMClient.prepare(
        LLM.request({
          id: "req_reasoning",
          model,
          messages: [LLM.assistant({ type: "reasoning", text: "hidden" })],
        }),
      ).pipe(Effect.flip)

      expect(error.message).toContain("OpenAI Chat assistant messages only support text and tool-call content for now")
    }),
  )

  it.effect("parses text and usage stream fixtures", () =>
    Effect.gen(function* () {
      const body = sseEvents(
        deltaChunk({ role: "assistant", content: "Hello" }),
        deltaChunk({ content: "!" }),
        deltaChunk({}, "stop"),
        usageChunk({
          prompt_tokens: 5,
          completion_tokens: 2,
          total_tokens: 7,
          prompt_tokens_details: { cached_tokens: 1 },
          completion_tokens_details: { reasoning_tokens: 0 },
        }),
      )
      const response = yield* LLMClient.generate(request).pipe(Effect.provide(fixedResponse(body)))

      expect(response.text).toBe("Hello!")
      expect(response.events).toEqual([
        { type: "text-delta", text: "Hello" },
        { type: "text-delta", text: "!" },
        {
          type: "request-finish",
          reason: "stop",
          usage: {
            inputTokens: 5,
            outputTokens: 2,
            reasoningTokens: 0,
            cacheReadInputTokens: 1,
            totalTokens: 7,
            native: {
              prompt_tokens: 5,
              completion_tokens: 2,
              total_tokens: 7,
              prompt_tokens_details: { cached_tokens: 1 },
              completion_tokens_details: { reasoning_tokens: 0 },
            },
          },
        },
      ])
    }),
  )

  it.effect("assembles streamed tool call input", () =>
    Effect.gen(function* () {
      const body = sseEvents(
        deltaChunk({
          role: "assistant",
          tool_calls: [{ index: 0, id: "call_1", function: { name: "lookup", arguments: '{"query"' } }],
        }),
        deltaChunk({ tool_calls: [{ index: 0, function: { arguments: ':"weather"}' } }] }),
        deltaChunk({}, "tool_calls"),
      )
      const response = yield* LLMClient.generate(
        LLM.updateRequest(request, {
          tools: [{ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } }],
        }),
      ).pipe(Effect.provide(fixedResponse(body)))

      expect(response.events).toEqual([
        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
        { type: "tool-call", id: "call_1", name: "lookup", input: { query: "weather" } },
        { type: "request-finish", reason: "tool-calls", usage: undefined },
      ])
    }),
  )

  it.effect("does not finalize streamed tool calls without a finish reason", () =>
    Effect.gen(function* () {
      const body = sseEvents(
        deltaChunk({
          role: "assistant",
          tool_calls: [{ index: 0, id: "call_1", function: { name: "lookup", arguments: '{"query"' } }],
        }),
        deltaChunk({ tool_calls: [{ index: 0, function: { arguments: ':"weather"}' } }] }),
      )
      const response = yield* LLMClient.generate(
        LLM.updateRequest(request, {
          tools: [{ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } }],
        }),
      ).pipe(Effect.provide(fixedResponse(body)))

      expect(response.events).toEqual([
        { type: "tool-input-delta", id: "call_1", name: "lookup", text: '{"query"' },
        { type: "tool-input-delta", id: "call_1", name: "lookup", text: ':"weather"}' },
      ])
      expect(response.toolCalls).toEqual([])
    }),
  )

  it.effect("fails on malformed stream events", () =>
    Effect.gen(function* () {
      const body = sseEvents(deltaChunk({ content: 123 }))
      const error = yield* LLMClient.generate(request).pipe(Effect.provide(fixedResponse(body)), Effect.flip)

      expect(error.message).toContain("Invalid openai/openai-chat stream event")
    }),
  )

  it.effect("surfaces transport errors that occur mid-stream", () =>
    Effect.gen(function* () {
      const layer = truncatedStream([
        `data: ${JSON.stringify(deltaChunk({ role: "assistant", content: "Hello" }))}\n\n`,
      ])
      const error = yield* LLMClient.generate(request).pipe(Effect.provide(layer), Effect.flip)

      expect(error.message).toContain("Failed to read openai/openai-chat stream")
    }),
  )

  it.effect("fails HTTP provider errors before stream parsing", () =>
    Effect.gen(function* () {
      const error = yield* LLMClient.generate(request).pipe(
        Effect.provide(
          fixedResponse('{"error":{"message":"Bad request","type":"invalid_request_error"}}', {
            status: 400,
            headers: { "content-type": "application/json" },
          }),
        ),
        Effect.flip,
      )

      expect(error).toBeInstanceOf(LLMError)
      expect(error.reason).toMatchObject({ _tag: "InvalidRequest" })
      expect(error.message).toContain("HTTP 400")
    }),
  )

  it.effect("short-circuits the upstream stream when the consumer takes a prefix", () =>
    Effect.gen(function* () {
      // The body has more chunks than we'll consume. If `Stream.take(1)` did
      // not interrupt the upstream HTTP body the test would hang waiting for
      // the rest of the stream to drain.
      const body = sseEvents(
        deltaChunk({ role: "assistant", content: "Hello" }),
        deltaChunk({ content: " world" }),
        deltaChunk({}, "stop"),
      )

      const events = Array.from(
        yield* LLMClient.stream(request).pipe(Stream.take(1), Stream.runCollect, Effect.provide(fixedResponse(body))),
      )
      expect(events.map((event) => event.type)).toEqual(["text-delta"])
    }),
  )
})