Prechádzať zdrojové kódy

test: finish HTTP mock processor coverage

Kit Langton 2 týždňov pred
rodič
commit
8d2385ad49

+ 299 - 187
packages/opencode/test/lib/llm-server.ts

@@ -1,39 +1,12 @@
-import { NodeHttpServer } from "@effect/platform-node"
+import { NodeHttpServer, NodeHttpServerRequest } from "@effect/platform-node"
 import * as Http from "node:http"
 import { Deferred, Effect, Layer, ServiceMap, Stream } from "effect"
 import * as HttpServer from "effect/unstable/http/HttpServer"
 import { HttpRouter, HttpServerRequest, HttpServerResponse } from "effect/unstable/http"
 
-type Usage = { input: number; output: number }
+export type Usage = { input: number; output: number }
 
-type Step =
-  | {
-      type: "text"
-      text: string
-      usage?: Usage
-    }
-  | {
-      type: "tool"
-      tool: string
-      input: unknown
-    }
-  | {
-      type: "fail"
-      message: string
-    }
-  | {
-      type: "httpError"
-      status: number
-      body: unknown
-    }
-  | {
-      type: "hang"
-    }
-  | {
-      type: "hold"
-      text: string
-      wait: PromiseLike<unknown>
-    }
+type Line = Record<string, unknown>
 
 type Hit = {
   url: URL
@@ -45,163 +18,293 @@ type Wait = {
   ready: Deferred.Deferred<void>
 }
 
-function sse(lines: unknown[]) {
-  return HttpServerResponse.stream(
-    Stream.fromIterable([
-      [...lines.map((line) => `data: ${JSON.stringify(line)}`), "data: [DONE]"].join("\n\n") + "\n\n",
-    ]).pipe(Stream.encodeText),
-    { contentType: "text/event-stream" },
-  )
+type Sse = {
+  type: "sse"
+  head: unknown[]
+  tail: unknown[]
+  wait?: PromiseLike<unknown>
+  hang?: boolean
+  error?: unknown
+  reset?: boolean
+}
+
+type HttpError = {
+  type: "http-error"
+  status: number
+  body: unknown
+}
+
+export type Item = Sse | HttpError
+
+const done = Symbol("done")
+
+function line(input: unknown) {
+  if (input === done) return "data: [DONE]\n\n"
+  return `data: ${JSON.stringify(input)}\n\n`
+}
+
+function tokens(input?: Usage) {
+  if (!input) return
+  return {
+    prompt_tokens: input.input,
+    completion_tokens: input.output,
+    total_tokens: input.input + input.output,
+  }
 }
 
-function text(step: Extract<Step, { type: "text" }>) {
-  const finish: Record<string, unknown> = {
+function chunk(input: { delta?: Record<string, unknown>; finish?: string; usage?: Usage }) {
+  return {
     id: "chatcmpl-test",
     object: "chat.completion.chunk",
-    choices: [{ delta: {}, finish_reason: "stop" }],
-  }
-  if (step.usage) {
-    finish.usage = {
-      prompt_tokens: step.usage.input,
-      completion_tokens: step.usage.output,
-      total_tokens: step.usage.input + step.usage.output,
-    }
-  }
-  return sse([
-    {
-      id: "chatcmpl-test",
-      object: "chat.completion.chunk",
-      choices: [{ delta: { role: "assistant" } }],
-    },
-    {
-      id: "chatcmpl-test",
-      object: "chat.completion.chunk",
-      choices: [{ delta: { content: step.text } }],
-    },
-    finish,
-  ])
+    choices: [
+      {
+        delta: input.delta ?? {},
+        ...(input.finish ? { finish_reason: input.finish } : {}),
+      },
+    ],
+    ...(input.usage ? { usage: tokens(input.usage) } : {}),
+  } satisfies Line
 }
 
-function httpError(step: Extract<Step, { type: "httpError" }>) {
-  return HttpServerResponse.text(JSON.stringify(step.body), {
-    status: step.status,
-    contentType: "application/json",
-  })
+function role() {
+  return chunk({ delta: { role: "assistant" } })
 }
 
-function tool(step: Extract<Step, { type: "tool" }>, seq: number) {
-  const id = `call_${seq}`
-  const args = JSON.stringify(step.input)
-  return sse([
-    {
-      id: "chatcmpl-test",
-      object: "chat.completion.chunk",
-      choices: [{ delta: { role: "assistant" } }],
-    },
-    {
-      id: "chatcmpl-test",
-      object: "chat.completion.chunk",
-      choices: [
+function textLine(value: string) {
+  return chunk({ delta: { content: value } })
+}
+
+function reasonLine(value: string) {
+  return chunk({ delta: { reasoning_content: value } })
+}
+
+function finishLine(reason: string, usage?: Usage) {
+  return chunk({ finish: reason, usage })
+}
+
+function toolStartLine(id: string, name: string) {
+  return chunk({
+    delta: {
+      tool_calls: [
         {
-          delta: {
-            tool_calls: [
-              {
-                index: 0,
-                id,
-                type: "function",
-                function: {
-                  name: step.tool,
-                  arguments: "",
-                },
-              },
-            ],
+          index: 0,
+          id,
+          type: "function",
+          function: {
+            name,
+            arguments: "",
           },
         },
       ],
     },
-    {
-      id: "chatcmpl-test",
-      object: "chat.completion.chunk",
-      choices: [
+  })
+}
+
+function toolArgsLine(value: string) {
+  return chunk({
+    delta: {
+      tool_calls: [
         {
-          delta: {
-            tool_calls: [
-              {
-                index: 0,
-                function: {
-                  arguments: args,
-                },
-              },
-            ],
+          index: 0,
+          function: {
+            arguments: value,
           },
         },
       ],
     },
-    {
-      id: "chatcmpl-test",
-      object: "chat.completion.chunk",
-      choices: [{ delta: {}, finish_reason: "tool_calls" }],
-    },
-  ])
+  })
 }
 
-function fail(step: Extract<Step, { type: "fail" }>) {
-  return HttpServerResponse.stream(
-    Stream.fromIterable([
-      'data: {"id":"chatcmpl-test","object":"chat.completion.chunk","choices":[{"delta":{"role":"assistant"}}]}\n\n',
-    ]).pipe(Stream.encodeText, Stream.concat(Stream.fail(new Error(step.message)))),
-    { contentType: "text/event-stream" },
-  )
+function bytes(input: Iterable<unknown>) {
+  return Stream.fromIterable([...input].map(line)).pipe(Stream.encodeText)
 }
 
-function hang() {
-  return HttpServerResponse.stream(
-    Stream.fromIterable([
-      'data: {"id":"chatcmpl-test","object":"chat.completion.chunk","choices":[{"delta":{"role":"assistant"}}]}\n\n',
-    ]).pipe(Stream.encodeText, Stream.concat(Stream.never)),
-    { contentType: "text/event-stream" },
-  )
+function send(item: Sse) {
+  const head = bytes(item.head)
+  const tail = bytes([...item.tail, ...(item.hang || item.error ? [] : [done])])
+  const empty = Stream.fromIterable<Uint8Array>([])
+  const wait = item.wait
+  const body: Stream.Stream<Uint8Array, unknown> = wait
+    ? Stream.concat(head, Stream.fromEffect(Effect.promise(() => wait)).pipe(Stream.flatMap(() => tail)))
+    : Stream.concat(head, tail)
+  let end: Stream.Stream<Uint8Array, unknown> = empty
+  if (item.error) end = Stream.concat(empty, Stream.fail(item.error))
+  else if (item.hang) end = Stream.concat(empty, Stream.never)
+
+  return HttpServerResponse.stream(Stream.concat(body, end), { contentType: "text/event-stream" })
 }
 
-function hold(step: Extract<Step, { type: "hold" }>) {
-  return HttpServerResponse.stream(
-    Stream.fromIterable([
-      'data: {"id":"chatcmpl-test","object":"chat.completion.chunk","choices":[{"delta":{"role":"assistant"}}]}\n\n',
-    ]).pipe(
-      Stream.encodeText,
-      Stream.concat(
-        Stream.fromEffect(Effect.promise(() => step.wait)).pipe(
-          Stream.flatMap(() =>
-            Stream.fromIterable([
-              `data: ${JSON.stringify({
-                id: "chatcmpl-test",
-                object: "chat.completion.chunk",
-                choices: [{ delta: { content: step.text } }],
-              })}\n\n`,
-              `data: ${JSON.stringify({
-                id: "chatcmpl-test",
-                object: "chat.completion.chunk",
-                choices: [{ delta: {}, finish_reason: "stop" }],
-              })}\n\n`,
-              "data: [DONE]\n\n",
-            ]).pipe(Stream.encodeText),
-          ),
-        ),
-      ),
-    ),
-    { contentType: "text/event-stream" },
-  )
+const reset = Effect.fn("TestLLMServer.reset")(function* (item: Sse) {
+  const req = yield* HttpServerRequest.HttpServerRequest
+  const res = NodeHttpServerRequest.toServerResponse(req)
+  yield* Effect.sync(() => {
+    res.writeHead(200, { "content-type": "text/event-stream" })
+    for (const part of item.head) res.write(line(part))
+    for (const part of item.tail) res.write(line(part))
+    res.destroy(new Error("connection reset"))
+  })
+  yield* Effect.never
+})
+
+function fail(item: HttpError) {
+  return HttpServerResponse.text(JSON.stringify(item.body), {
+    status: item.status,
+    contentType: "application/json",
+  })
+}
+
+export class Reply {
+  #head: unknown[] = [role()]
+  #tail: unknown[] = []
+  #usage: Usage | undefined
+  #finish: string | undefined
+  #wait: PromiseLike<unknown> | undefined
+  #hang = false
+  #error: unknown
+  #reset = false
+  #seq = 0
+
+  #id() {
+    this.#seq += 1
+    return `call_${this.#seq}`
+  }
+
+  text(value: string) {
+    this.#tail = [...this.#tail, textLine(value)]
+    return this
+  }
+
+  reason(value: string) {
+    this.#tail = [...this.#tail, reasonLine(value)]
+    return this
+  }
+
+  usage(value: Usage) {
+    this.#usage = value
+    return this
+  }
+
+  wait(value: PromiseLike<unknown>) {
+    this.#wait = value
+    return this
+  }
+
+  stop() {
+    this.#finish = "stop"
+    this.#hang = false
+    this.#error = undefined
+    this.#reset = false
+    return this
+  }
+
+  toolCalls() {
+    this.#finish = "tool_calls"
+    this.#hang = false
+    this.#error = undefined
+    this.#reset = false
+    return this
+  }
+
+  tool(name: string, input: unknown) {
+    const id = this.#id()
+    const args = JSON.stringify(input)
+    this.#tail = [...this.#tail, toolStartLine(id, name), toolArgsLine(args)]
+    return this.toolCalls()
+  }
+
+  pendingTool(name: string, input: unknown) {
+    const id = this.#id()
+    const args = JSON.stringify(input)
+    const size = Math.max(1, Math.floor(args.length / 2))
+    this.#tail = [...this.#tail, toolStartLine(id, name), toolArgsLine(args.slice(0, size))]
+    return this
+  }
+
+  hang() {
+    this.#finish = undefined
+    this.#hang = true
+    this.#error = undefined
+    this.#reset = false
+    return this
+  }
+
+  streamError(error: unknown = "boom") {
+    this.#finish = undefined
+    this.#hang = false
+    this.#error = error
+    this.#reset = false
+    return this
+  }
+
+  reset() {
+    this.#finish = undefined
+    this.#hang = false
+    this.#error = undefined
+    this.#reset = true
+    return this
+  }
+
+  item(): Item {
+    return {
+      type: "sse",
+      head: this.#head,
+      tail: this.#finish ? [...this.#tail, finishLine(this.#finish, this.#usage)] : this.#tail,
+      wait: this.#wait,
+      hang: this.#hang,
+      error: this.#error,
+      reset: this.#reset,
+    }
+  }
+}
+
+export function reply() {
+  return new Reply()
+}
+
+export function httpError(status: number, body: unknown): Item {
+  return {
+    type: "http-error",
+    status,
+    body,
+  }
+}
+
+export function raw(input: {
+  chunks?: unknown[]
+  head?: unknown[]
+  tail?: unknown[]
+  wait?: PromiseLike<unknown>
+  hang?: boolean
+  error?: unknown
+  reset?: boolean
+}): Item {
+  return {
+    type: "sse",
+    head: input.head ?? input.chunks ?? [],
+    tail: input.tail ?? [],
+    wait: input.wait,
+    hang: input.hang,
+    error: input.error,
+    reset: input.reset,
+  }
+}
+
+function item(input: Item | Reply) {
+  return input instanceof Reply ? input.item() : input
 }
 
 namespace TestLLMServer {
   export interface Service {
     readonly url: string
+    readonly push: (...input: (Item | Reply)[]) => Effect.Effect<void>
     readonly text: (value: string, opts?: { usage?: Usage }) => Effect.Effect<void>
-    readonly tool: (tool: string, input: unknown) => Effect.Effect<void>
-    readonly fail: (message?: string) => Effect.Effect<void>
+    readonly tool: (name: string, input: unknown) => Effect.Effect<void>
+    readonly toolHang: (name: string, input: unknown) => Effect.Effect<void>
+    readonly reason: (value: string, opts?: { text?: string; usage?: Usage }) => Effect.Effect<void>
+    readonly fail: (message?: unknown) => Effect.Effect<void>
     readonly error: (status: number, body: unknown) => Effect.Effect<void>
     readonly hang: Effect.Effect<void>
-    readonly hold: (text: string, wait: PromiseLike<unknown>) => Effect.Effect<void>
+    readonly hold: (value: string, wait: PromiseLike<unknown>) => Effect.Effect<void>
     readonly hits: Effect.Effect<Hit[]>
     readonly calls: Effect.Effect<number>
     readonly wait: (count: number) => Effect.Effect<void>
@@ -218,12 +321,11 @@ export class TestLLMServer extends ServiceMap.Service<TestLLMServer, TestLLMServ
       const router = yield* HttpRouter.HttpRouter
 
       let hits: Hit[] = []
-      let list: Step[] = []
-      let seq = 0
+      let list: Item[] = []
       let waits: Wait[] = []
 
-      const push = (step: Step) => {
-        list = [...list, step]
+      const queue = (...input: (Item | Reply)[]) => {
+        list = [...list, ...input.map(item)]
       }
 
       const notify = Effect.fnUntraced(function* () {
@@ -234,11 +336,10 @@ export class TestLLMServer extends ServiceMap.Service<TestLLMServer, TestLLMServ
       })
 
       const pull = () => {
-        const step = list[0]
-        if (!step) return { step: undefined, seq }
-        seq += 1
+        const first = list[0]
+        if (!first) return
         list = list.slice(1)
-        return { step, seq }
+        return first
       }
 
       yield* router.add(
@@ -247,22 +348,22 @@ export class TestLLMServer extends ServiceMap.Service<TestLLMServer, TestLLMServ
         Effect.gen(function* () {
           const req = yield* HttpServerRequest.HttpServerRequest
           const next = pull()
-          if (!next.step) return HttpServerResponse.text("unexpected request", { status: 500 })
-          const json = yield* req.json.pipe(Effect.orElseSucceed(() => ({})))
+          if (!next) return HttpServerResponse.text("unexpected request", { status: 500 })
+          const body = yield* req.json.pipe(Effect.orElseSucceed(() => ({})))
           hits = [
             ...hits,
             {
               url: new URL(req.originalUrl, "http://localhost"),
-              body: json && typeof json === "object" ? (json as Record<string, unknown>) : {},
+              body: body && typeof body === "object" ? (body as Record<string, unknown>) : {},
             },
           ]
           yield* notify()
-          if (next.step.type === "text") return text(next.step)
-          if (next.step.type === "tool") return tool(next.step, next.seq)
-          if (next.step.type === "fail") return fail(next.step)
-          if (next.step.type === "httpError") return httpError(next.step)
-          if (next.step.type === "hang") return hang()
-          return hold(next.step)
+          if (next.type === "sse" && next.reset) {
+            yield* reset(next)
+            return HttpServerResponse.empty()
+          }
+          if (next.type === "sse") return send(next)
+          return fail(next)
         }),
       )
 
@@ -273,23 +374,37 @@ export class TestLLMServer extends ServiceMap.Service<TestLLMServer, TestLLMServ
           server.address._tag === "TcpAddress"
             ? `http://127.0.0.1:${server.address.port}/v1`
             : `unix://${server.address.path}/v1`,
+        push: Effect.fn("TestLLMServer.push")(function* (...input: (Item | Reply)[]) {
+          queue(...input)
+        }),
         text: Effect.fn("TestLLMServer.text")(function* (value: string, opts?: { usage?: Usage }) {
-          push({ type: "text", text: value, usage: opts?.usage })
+          const out = reply().text(value)
+          if (opts?.usage) out.usage(opts.usage)
+          queue(out.stop().item())
         }),
-        tool: Effect.fn("TestLLMServer.tool")(function* (tool: string, input: unknown) {
-          push({ type: "tool", tool, input })
+        tool: Effect.fn("TestLLMServer.tool")(function* (name: string, input: unknown) {
+          queue(reply().tool(name, input).item())
         }),
-        fail: Effect.fn("TestLLMServer.fail")(function* (message = "boom") {
-          push({ type: "fail", message })
+        toolHang: Effect.fn("TestLLMServer.toolHang")(function* (name: string, input: unknown) {
+          queue(reply().pendingTool(name, input).hang().item())
+        }),
+        reason: Effect.fn("TestLLMServer.reason")(function* (value: string, opts?: { text?: string; usage?: Usage }) {
+          const out = reply().reason(value)
+          if (opts?.text) out.text(opts.text)
+          if (opts?.usage) out.usage(opts.usage)
+          queue(out.stop().item())
+        }),
+        fail: Effect.fn("TestLLMServer.fail")(function* (message: unknown = "boom") {
+          queue(reply().streamError(message).item())
         }),
-        hang: Effect.gen(function* () {
-          push({ type: "hang" })
-        }).pipe(Effect.withSpan("TestLLMServer.hang")),
         error: Effect.fn("TestLLMServer.error")(function* (status: number, body: unknown) {
-          push({ type: "httpError", status, body })
+          queue(httpError(status, body))
         }),
-        hold: Effect.fn("TestLLMServer.hold")(function* (text: string, wait: PromiseLike<unknown>) {
-          push({ type: "hold", text, wait })
+        hang: Effect.gen(function* () {
+          queue(reply().hang().item())
+        }).pipe(Effect.withSpan("TestLLMServer.hang")),
+        hold: Effect.fn("TestLLMServer.hold")(function* (value: string, wait: PromiseLike<unknown>) {
+          queue(reply().wait(wait).text(value).stop().item())
         }),
         hits: Effect.sync(() => [...hits]),
         calls: Effect.sync(() => hits.length),
@@ -303,8 +418,5 @@ export class TestLLMServer extends ServiceMap.Service<TestLLMServer, TestLLMServ
         pending: Effect.sync(() => list.length),
       })
     }),
-  ).pipe(
-    Layer.provide(HttpRouter.layer), //
-    Layer.provide(NodeHttpServer.layer(() => Http.createServer(), { port: 0 })),
-  )
+  ).pipe(Layer.provide(HttpRouter.layer), Layer.provide(NodeHttpServer.layer(() => Http.createServer(), { port: 0 })))
 }

+ 65 - 201
packages/opencode/test/session/processor-effect.test.ts

@@ -1,7 +1,6 @@
 import { NodeFileSystem } from "@effect/platform-node"
 import { expect } from "bun:test"
-import { APICallError, jsonSchema, tool } from "ai"
-import { Cause, Effect, Exit, Fiber, Layer, ServiceMap, Stream } from "effect"
+import { Cause, Effect, Exit, Fiber, Layer } from "effect"
 import path from "path"
 import type { Agent } from "../../src/agent/agent"
 import { Agent as AgentSvc } from "../../src/agent/agent"
@@ -20,19 +19,12 @@ import { SessionStatus } from "../../src/session/status"
 import { Snapshot } from "../../src/snapshot"
 import { Log } from "../../src/util/log"
 import * as CrossSpawnSpawner from "../../src/effect/cross-spawn-spawner"
-import { provideTmpdirInstance, provideTmpdirServer } from "../fixture/fixture"
+import { provideTmpdirServer } from "../fixture/fixture"
 import { testEffect } from "../lib/effect"
-import { TestLLMServer } from "../lib/llm-server"
+import { reply, TestLLMServer } from "../lib/llm-server"
 
 Log.init({ print: false })
 
-const DEBUG = process.env.OPENCODE_TEST_DEBUG === "1"
-
-function trace(label: string, value: unknown) {
-  if (!DEBUG) return
-  console.log(label, JSON.stringify(value, null, 2))
-}
-
 const ref = {
   providerID: ProviderID.make("test"),
   modelID: ModelID.make("test-model"),
@@ -168,109 +160,6 @@ const env = Layer.mergeAll(TestLLMServer.layer, SessionProcessor.layer.pipe(Laye
 
 const it = testEffect(env)
 
-// ---------------------------------------------------------------------------
-// TestLLM kept only for the reasoning test
-// TODO: reasoning events not available via OpenAI-compatible SSE
-// ---------------------------------------------------------------------------
-type Script = Stream.Stream<LLM.Event, unknown> | ((input: LLM.StreamInput) => Stream.Stream<LLM.Event, unknown>)
-
-class TestLLM extends ServiceMap.Service<
-  TestLLM,
-  {
-    readonly push: (stream: Script) => Effect.Effect<void>
-    readonly reply: (...items: LLM.Event[]) => Effect.Effect<void>
-    readonly calls: Effect.Effect<number>
-    readonly inputs: Effect.Effect<LLM.StreamInput[]>
-  }
->()("@test/SessionProcessorLLM") {}
-
-function reasoningUsage(input = 1, output = 1, total = input + output) {
-  return {
-    inputTokens: input,
-    outputTokens: output,
-    totalTokens: total,
-    inputTokenDetails: {
-      noCacheTokens: undefined,
-      cacheReadTokens: undefined,
-      cacheWriteTokens: undefined,
-    },
-    outputTokenDetails: {
-      textTokens: undefined,
-      reasoningTokens: undefined,
-    },
-  }
-}
-
-const reasoningLlm = Layer.unwrap(
-  Effect.gen(function* () {
-    const queue: Script[] = []
-    const inputs: LLM.StreamInput[] = []
-    let calls = 0
-
-    const push = Effect.fn("TestLLM.push")((item: Script) => {
-      queue.push(item)
-      return Effect.void
-    })
-
-    const reply = Effect.fn("TestLLM.reply")((...items: LLM.Event[]) => push(Stream.make(...items)))
-    return Layer.mergeAll(
-      Layer.succeed(
-        LLM.Service,
-        LLM.Service.of({
-          stream: (input) => {
-            calls += 1
-            inputs.push(input)
-            const item = queue.shift() ?? Stream.empty
-            return typeof item === "function" ? item(input) : item
-          },
-        }),
-      ),
-      Layer.succeed(
-        TestLLM,
-        TestLLM.of({
-          push,
-          reply,
-          calls: Effect.sync(() => calls),
-          inputs: Effect.sync(() => [...inputs]),
-        }),
-      ),
-    )
-  }),
-)
-
-const reasoningDeps = Layer.mergeAll(
-  Session.defaultLayer,
-  Snapshot.defaultLayer,
-  AgentSvc.defaultLayer,
-  Permission.layer,
-  Plugin.defaultLayer,
-  Config.defaultLayer,
-  status,
-  reasoningLlm,
-).pipe(Layer.provideMerge(infra))
-const reasoningEnv = SessionProcessor.layer.pipe(Layer.provideMerge(reasoningDeps))
-const reasoningIt = testEffect(reasoningEnv)
-
-function reasoningModel(context: number): Provider.Model {
-  return {
-    id: "test-model",
-    providerID: "test",
-    name: "Test",
-    limit: { context, output: 10 },
-    cost: { input: 0, output: 0, cache: { read: 0, write: 0 } },
-    capabilities: {
-      toolcall: true,
-      attachment: false,
-      reasoning: false,
-      temperature: true,
-      input: { text: true, image: false, audio: false, video: false },
-      output: { text: true, image: false, audio: false, video: false },
-    },
-    api: { npm: "@ai-sdk/anthropic" },
-    options: {},
-  } as Provider.Model
-}
-
 const boot = Effect.fn("test.boot")(function* () {
   const processors = yield* SessionProcessor.Service
   const session = yield* Session.Service
@@ -367,12 +256,6 @@ it.live("session.processor effect tests stop after token overflow requests compa
 
         const parts = yield* Effect.promise(() => MessageV2.parts(msg.id))
 
-        trace("overflow", {
-          value,
-          parts: parts.map((part) => part.type),
-          inputs: yield* llm.inputs,
-        })
-
         expect(value).toBe("compact")
         expect(parts.some((part) => part.type === "text" && part.text === "after")).toBe(true)
         expect(parts.some((part) => part.type === "step-finish")).toBe(true)
@@ -381,57 +264,66 @@ it.live("session.processor effect tests stop after token overflow requests compa
   ),
 )
 
-// TODO: reasoning events not available via OpenAI-compatible SSE
-reasoningIt.live("session.processor effect tests reset reasoning state across retries", () =>
-  provideTmpdirInstance(
-    (dir) =>
+it.live("session.processor effect tests capture reasoning from http mock", () =>
+  provideTmpdirServer(
+    ({ dir, llm }) =>
       Effect.gen(function* () {
-        const test = yield* TestLLM
-        const processors = yield* SessionProcessor.Service
-        const session = yield* Session.Service
-
-        yield* test.push(
-          Stream.fromIterable<LLM.Event>([
-            { type: "start" },
-            { type: "reasoning-start", id: "r" },
-            { type: "reasoning-delta", id: "r", text: "one" },
-          ]).pipe(
-            Stream.concat(
-              Stream.fail(
-                new APICallError({
-                  message: "boom",
-                  url: "https://example.com/v1/chat/completions",
-                  requestBodyValues: {},
-                  statusCode: 503,
-                  responseHeaders: { "retry-after-ms": "0" },
-                  responseBody: '{"error":"boom"}',
-                  isRetryable: true,
-                }),
-              ),
-            ),
-          ),
-        )
-
-        yield* test.reply(
-          { type: "start" },
-          { type: "reasoning-start", id: "r" },
-          { type: "reasoning-delta", id: "r", text: "two" },
-          { type: "reasoning-end", id: "r" },
-          {
-            type: "finish-step",
-            finishReason: "stop",
-            rawFinishReason: "stop",
-            response: { id: "res", modelId: "test-model", timestamp: new Date() },
-            providerMetadata: undefined,
-            usage: reasoningUsage(),
-          },
-          { type: "finish", finishReason: "stop", rawFinishReason: "stop", totalUsage: reasoningUsage() },
-        )
+        const { processors, session, provider } = yield* boot()
+
+        yield* llm.push(reply().reason("think").text("done").stop())
 
         const chat = yield* session.create({})
         const parent = yield* user(chat.id, "reason")
         const msg = yield* assistant(chat.id, parent.id, path.resolve(dir))
-        const mdl = reasoningModel(100)
+        const mdl = yield* provider.getModel(ref.providerID, ref.modelID)
+        const handle = yield* processors.create({
+          assistantMessage: msg,
+          sessionID: chat.id,
+          model: mdl,
+        })
+
+        const value = yield* handle.process({
+          user: {
+            id: parent.id,
+            sessionID: chat.id,
+            role: "user",
+            time: parent.time,
+            agent: parent.agent,
+            model: { providerID: ref.providerID, modelID: ref.modelID },
+          } satisfies MessageV2.User,
+          sessionID: chat.id,
+          model: mdl,
+          agent: agent(),
+          system: [],
+          messages: [{ role: "user", content: "reason" }],
+          tools: {},
+        })
+
+        const parts = yield* Effect.promise(() => MessageV2.parts(msg.id))
+        const reasoning = parts.find((part): part is MessageV2.ReasoningPart => part.type === "reasoning")
+        const text = parts.find((part): part is MessageV2.TextPart => part.type === "text")
+
+        expect(value).toBe("continue")
+        expect(yield* llm.calls).toBe(1)
+        expect(reasoning?.text).toBe("think")
+        expect(text?.text).toBe("done")
+      }),
+    { git: true, config: (url) => providerCfg(url) },
+  ),
+)
+
+it.live("session.processor effect tests reset reasoning state across retries", () =>
+  provideTmpdirServer(
+    ({ dir, llm }) =>
+      Effect.gen(function* () {
+        const { processors, session, provider } = yield* boot()
+
+        yield* llm.push(reply().reason("one").reset(), reply().reason("two").stop())
+
+        const chat = yield* session.create({})
+        const parent = yield* user(chat.id, "reason")
+        const msg = yield* assistant(chat.id, parent.id, path.resolve(dir))
+        const mdl = yield* provider.getModel(ref.providerID, ref.modelID)
         const handle = yield* processors.create({
           assistantMessage: msg,
           sessionID: chat.id,
@@ -459,11 +351,11 @@ reasoningIt.live("session.processor effect tests reset reasoning state across re
         const reasoning = parts.filter((part): part is MessageV2.ReasoningPart => part.type === "reasoning")
 
         expect(value).toBe("continue")
-        expect(yield* test.calls).toBe(2)
+        expect(yield* llm.calls).toBe(2)
         expect(reasoning.some((part) => part.text === "two")).toBe(true)
         expect(reasoning.some((part) => part.text === "onetwo")).toBe(false)
       }),
-    { git: true },
+    { git: true, config: (url) => providerCfg(url) },
   ),
 )
 
@@ -502,13 +394,6 @@ it.live("session.processor effect tests do not retry unknown json errors", () =>
           tools: {},
         })
 
-        trace("unknown-error", {
-          value,
-          calls: yield* llm.calls,
-          inputs: yield* llm.inputs,
-          error: handle.message.error,
-        })
-
         expect(value).toBe("stop")
         expect(yield* llm.calls).toBe(1)
         expect(handle.message.error?.name).toBe("APIError")
@@ -664,9 +549,8 @@ it.live("session.processor effect tests mark pending tools as aborted on cleanup
     ({ dir, llm }) =>
       Effect.gen(function* () {
         const { processors, session, provider } = yield* boot()
-        const wait = new Promise<{ output: string }>(() => {})
 
-        yield* llm.tool("bash", { cmd: "pwd" })
+        yield* llm.toolHang("bash", { cmd: "pwd" })
 
         const chat = yield* session.create({})
         const parent = yield* user(chat.id, "tool abort")
@@ -693,26 +577,16 @@ it.live("session.processor effect tests mark pending tools as aborted on cleanup
             agent: agent(),
             system: [],
             messages: [{ role: "user", content: "tool abort" }],
-            tools: {
-              bash: tool({
-                description: "Run shell commands",
-                inputSchema: jsonSchema({
-                  type: "object",
-                  properties: {
-                    cmd: { type: "string" },
-                  },
-                  required: ["cmd"],
-                }),
-                execute: async () => wait,
-              }),
-            },
+            tools: {},
           })
           .pipe(Effect.forkChild)
 
         yield* llm.wait(1)
         yield* Effect.promise(async () => {
           const end = Date.now() + 500
-          while (!handle.partFromToolCall("call_1") && Date.now() < end) {
+          while (Date.now() < end) {
+            const parts = await MessageV2.parts(msg.id)
+            if (parts.some((part) => part.type === "tool")) return
             await Bun.sleep(10)
           }
         })
@@ -725,16 +599,6 @@ it.live("session.processor effect tests mark pending tools as aborted on cleanup
         const parts = yield* Effect.promise(() => MessageV2.parts(msg.id))
         const call = parts.find((part): part is MessageV2.ToolPart => part.type === "tool")
 
-        trace("tool-abort", {
-          calls: yield* llm.calls,
-          inputs: yield* llm.inputs,
-          pending: handle.partFromToolCall("call_1"),
-          parts: parts.map((part) => ({
-            type: part.type,
-            ...(part.type === "tool" ? { state: part.state.status } : {}),
-          })),
-        })
-
         expect(Exit.isFailure(exit)).toBe(true)
         if (Exit.isFailure(exit)) {
           expect(Cause.hasInterruptsOnly(exit.cause)).toBe(true)