]> git.giacomolli.uk - onechat/blob - turn_test.go
Run the turns of an agent in a loop
[onechat] / turn_test.go
1 package main
2
3 import (
4 "context"
5 "errors"
6 "net/http"
7 "strings"
8 "sync"
9 "testing"
10 "time"
11
12 "github.com/stretchr/testify/require"
13 )
14
15 func TestTurnRunsToolsUntilTheFinalAnswer(t *testing.T) {
16 server := newFakeResponses(t)
17 server.reply = func(w http.ResponseWriter, call fakeCall) {
18 if call.number == 0 {
19 writeEvents(w, completedEvent([]any{
20 reasoningItem("rs_1", "opaque thought"),
21 functionCallItem("fc_1", "call_1", "shell", `{"command":"echo hi"}`),
22 }, 100, 0, 5))
23
24 return
25 }
26
27 writeEvents(w, completedEvent([]any{messageItem("It said hi.")}, 120, 100, 5))
28 }
29
30 test := newTestLoop(t, server)
31 _, err := test.loop.Submit("Run echo hi.")
32 require.NoError(t, err)
33 test.events.WaitFor(t, "idle", 1)
34 require.Equal(t, 2, server.Calls())
35
36 first := server.Call(0).body
37 second := server.Call(1).body
38 instructions, err := test.agent.Prompt()
39 require.NoError(t, err)
40 require.Equal(t, instructions, first["instructions"])
41 require.Equal(t, test.agent.config.Model, first["model"])
42
43 // The second request begins with the whole first one.
44 firstInput := first["input"].([]any)
45 secondInput := second["input"].([]any)
46 require.Equal(t, firstInput, secondInput[:len(firstInput)])
47 require.Equal(t, []any{
48 map[string]any{"type": "reasoning", "encrypted_content": "opaque thought", "summary": []any{}},
49 map[string]any{"type": "function_call", "call_id": "call_1", "name": "shell", "arguments": `{"command":"echo hi"}`},
50 map[string]any{"type": "function_call_output", "call_id": "call_1", "output": "hi\n"},
51 }, secondInput[len(firstInput):])
52
53 require.Equal(t, []string{
54 "user: Run echo hi.",
55 `tool: shell {"command":"echo hi"}`,
56 "echo: hi\n",
57 "reply: It said hi.",
58 }, test.Log(t))
59 require.Equal(t, []string{"user", "busy", "tool", "echo", "reply", "idle"}, test.events.Kinds())
60 }
61
62 func TestFirstRequestHoldsTheViewTheStateAndTheMessage(t *testing.T) {
63 server := newFakeResponses(t)
64 test := newTestLoop(t, server)
65 _, err := test.memory.Append("user", "an earlier message")
66 require.NoError(t, err)
67
68 _, err = test.loop.Submit("The new message.")
69 require.NoError(t, err)
70 test.events.WaitFor(t, "idle", 1)
71
72 require.Equal(t, []string{
73 "<chat>\n0+1|user: an earlier message\n</chat>\n",
74 "Now: Thursday 2026-01-01 12:00 +00:00\n",
75 "The new message.",
76 }, inputParts(t, server.Call(0)))
77 }
78
79 func TestMessageDuringATurnJoinsTheNextRequest(t *testing.T) {
80 server := newFakeResponses(t)
81 arrived := make(chan struct{})
82 release := make(chan struct{})
83 server.reply = func(w http.ResponseWriter, call fakeCall) {
84 if call.number == 0 {
85 close(arrived)
86 <-release
87 writeEvents(w, completedEvent([]any{functionCallItem("fc_1", "call_1", "date", `{"id":0}`)}, 100, 0, 5))
88
89 return
90 }
91
92 writeEvents(w, completedEvent([]any{messageItem("done")}, 100, 0, 5))
93 }
94
95 test := newTestLoop(t, server)
96 _, err := test.loop.Submit("first")
97 require.NoError(t, err)
98
99 <-arrived
100 _, err = test.loop.Submit("second, during the turn")
101 require.NoError(t, err)
102 close(release)
103
104 test.events.WaitFor(t, "idle", 1)
105 input := server.Call(1).body["input"].([]any)
106 require.Equal(t, map[string]any{
107 "role": "user",
108 "content": []any{map[string]any{"type": "input_text", "text": "second, during the turn"}},
109 }, input[len(input)-1])
110
111 time.Sleep(50 * time.Millisecond)
112 require.Equal(t, 2, server.Calls())
113 require.Equal(t, 1, test.events.Count("busy"))
114 }
115
116 func TestMessageDuringTheFinalRequestStartsANewTurn(t *testing.T) {
117 server := newFakeResponses(t)
118 arrived := make(chan struct{})
119 release := make(chan struct{})
120 server.reply = func(w http.ResponseWriter, call fakeCall) {
121 if call.number == 0 {
122 close(arrived)
123 <-release
124 }
125
126 writeEvents(w, completedEvent([]any{messageItem("answer")}, 100, 0, 5))
127 }
128
129 test := newTestLoop(t, server)
130 _, err := test.loop.Submit("first")
131 require.NoError(t, err)
132
133 <-arrived
134 _, err = test.loop.Submit("second")
135 require.NoError(t, err)
136 close(release)
137
138 test.events.WaitFor(t, "idle", 2)
139 require.Equal(t, 2, server.Calls())
140
141 parts := inputParts(t, server.Call(1))
142 require.Equal(t, "second", parts[len(parts)-1])
143 require.Contains(t, parts[0], "|reply: answer\n")
144 }
145
146 func TestTurnWaitsForTheSummaries(t *testing.T) {
147 server := newFakeResponses(t)
148 summarizer := newFakeSummarizer()
149 summarizer.gate = make(chan struct{})
150 test := newTestLoopWith(t, server, summarizer)
151
152 _, err := test.memory.Append("echo", strings.Repeat("long output ", 100))
153 require.NoError(t, err)
154
155 _, err = test.loop.Submit("go on")
156 require.NoError(t, err)
157
158 time.Sleep(50 * time.Millisecond)
159 require.Equal(t, 0, server.Calls())
160
161 close(summarizer.gate)
162 test.events.WaitFor(t, "idle", 1)
163 require.Equal(t, 1, server.Calls())
164 require.NotContains(t, inputParts(t, server.Call(0))[0], pending)
165 }
166
167 func TestInterruptEndsTheTurn(t *testing.T) {
168 server := newFakeResponses(t)
169 arrived := make(chan struct{})
170 release := make(chan struct{})
171 server.reply = func(w http.ResponseWriter, call fakeCall) {
172 if call.number == 0 {
173 close(arrived)
174 <-release
175 }
176
177 writeEvents(w, completedEvent([]any{messageItem("answer")}, 100, 0, 5))
178 }
179
180 test := newTestLoop(t, server)
181 _, err := test.loop.Submit("a long task")
182 require.NoError(t, err)
183
184 <-arrived
185 require.True(t, test.loop.Interrupt())
186 test.events.WaitFor(t, "idle", 1)
187 close(release)
188
189 require.Equal(t, "interrupted", test.events.Last("error"))
190 require.Equal(t, []string{"user: a long task"}, test.Log(t))
191 require.False(t, test.loop.Interrupt())
192
193 _, err = test.loop.Submit("again")
194 require.NoError(t, err)
195 test.events.WaitFor(t, "idle", 2)
196 require.Equal(t, "reply: answer", test.Log(t)[2])
197 }
198
199 func TestFailedCallEndsTheTurnWithoutARetry(t *testing.T) {
200 server := newFakeResponses(t)
201 server.reply = func(w http.ResponseWriter, call fakeCall) {
202 http.Error(w, "busy", http.StatusServiceUnavailable)
203 }
204
205 test := newTestLoop(t, server)
206 _, err := test.loop.Submit("hello")
207 require.NoError(t, err)
208 test.events.WaitFor(t, "idle", 1)
209
210 time.Sleep(50 * time.Millisecond)
211 require.Equal(t, 1, server.Calls())
212 require.Contains(t, test.events.Last("error"), "HTTP 503")
213 }
214
215 // testLoop is a loop that runs in the background of a test, with its parts.
216 type testLoop struct {
217 agent *Agent
218 memory *Memory
219 loop *loop
220 events *eventRecorder
221 }
222
223 // Log returns the logged messages as "kind: text".
224 func (l *testLoop) Log(t *testing.T) []string {
225 t.Helper()
226
227 var lines []string
228 for id := range l.memory.Len() {
229 message, err := l.memory.Message(id)
230 require.NoError(t, err)
231
232 lines = append(lines, message.Kind+": "+message.Text)
233 }
234
235 return lines
236 }
237
238 // eventRecorder keeps the events of a loop.
239 type eventRecorder struct {
240 mutex sync.Mutex
241 events []Event
242 }
243
244 func (r *eventRecorder) Add(event Event) {
245 r.mutex.Lock()
246 defer r.mutex.Unlock()
247
248 r.events = append(r.events, event)
249 }
250
251 func (r *eventRecorder) Kinds() []string {
252 r.mutex.Lock()
253 defer r.mutex.Unlock()
254
255 kinds := make([]string, len(r.events))
256 for n, event := range r.events {
257 kinds[n] = event.Kind
258 }
259
260 return kinds
261 }
262
263 func (r *eventRecorder) Count(kind string) int {
264 count := 0
265 for _, each := range r.Kinds() {
266 if each == kind {
267 count++
268 }
269 }
270
271 return count
272 }
273
274 // Last returns the text of the last event of kind.
275 func (r *eventRecorder) Last(kind string) string {
276 r.mutex.Lock()
277 defer r.mutex.Unlock()
278
279 for n := len(r.events) - 1; n >= 0; n-- {
280 if r.events[n].Kind == kind {
281 return r.events[n].Text
282 }
283 }
284
285 return ""
286 }
287
288 // WaitFor waits until count events of kind came.
289 func (r *eventRecorder) WaitFor(t *testing.T, kind string, count int) {
290 t.Helper()
291
292 require.Eventually(t, func() bool {
293 return r.Count(kind) >= count
294 }, 10*time.Second, time.Millisecond)
295 }
296
297 func newTestLoop(t *testing.T, server *fakeResponses) *testLoop {
298 t.Helper()
299
300 return newTestLoopWith(t, server, newFakeSummarizer())
301 }
302
303 // newTestLoopWith returns a running loop on a new agent whose builds use
304 // summarizer and whose commands run in a temporary directory.
305 func newTestLoopWith(t *testing.T, server *fakeResponses, summarizer Summarizer) *testLoop {
306 t.Helper()
307
308 agent, err := openAgent(t.TempDir())
309 require.NoError(t, err)
310
311 agent.config.Workdir = t.TempDir()
312 clock := newTestClock()
313 memory, err := openMemory(t.Context(), t.TempDir(), clock.Now, summarizer)
314 require.NoError(t, err)
315
316 test := &testLoop{
317 agent: agent,
318 memory: memory,
319 events: &eventRecorder{},
320 }
321
322 test.loop = newLoop(agent, memory, newTestClient(t, server), clock.Now, test.events.Add)
323 ctx, cancel := context.WithCancel(t.Context())
324 done := make(chan error)
325 go func() {
326 done <- test.loop.Run(ctx)
327 }()
328
329 t.Cleanup(func() {
330 cancel()
331 require.True(t, errors.Is(<-done, context.Canceled))
332 require.NoError(t, memory.Close())
333 })
334
335 return test
336 }
337
338 func reasoningItem(id, encrypted string) map[string]any {
339 return map[string]any{
340 "type": "reasoning",
341 "id": id,
342 "encrypted_content": encrypted,
343 "summary": []any{},
344 }
345 }
346
347 func functionCallItem(id, callID, name, arguments string) map[string]any {
348 return map[string]any{
349 "type": "function_call",
350 "id": id,
351 "call_id": callID,
352 "name": name,
353 "arguments": arguments,
354 }
355 }