]> git.giacomolli.uk - onechat/blob - turn_test.go
Separate multiline Go statements
[onechat] / turn_test.go
1 package main
2
3 import (
4 "context"
5 "encoding/json"
6 "net/http"
7 "path/filepath"
8 "strings"
9 "sync"
10 "testing"
11 "time"
12
13 "github.com/stretchr/testify/require"
14 )
15
16 func TestTurnRunsToolsUntilTheFinalAnswer(t *testing.T) {
17 server := newModelFake(t)
18 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
19 if call.number == 0 {
20 modelWriteEvents(w, modelCompletedEvent([]any{
21 modelReasoningItem("rs_1", "opaque thought"),
22 modelFunctionCallItem("fc_1", "call_1", "shell", `{"command":"echo hi"}`),
23 }, 100, 0, 5))
24
25 return
26 }
27
28 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("It said hi.")}, 120, 100, 5))
29 }
30
31 test := newTestTurnLoop(t, server)
32 _, err := test.loop.Submit(terminalSource, "Run echo hi.")
33 require.NoError(t, err)
34 test.events.WaitFor(t, "idle", 1)
35 require.Equal(t, 2, server.Calls())
36
37 first := server.Call(0).body
38 second := server.Call(1).body
39 instructions, err := test.agent.Prompt()
40 require.NoError(t, err)
41 require.Equal(t, instructions, first["instructions"])
42 require.Equal(t, test.agent.config.Model, first["model"])
43
44 // The second request begins with the whole first one.
45 firstInput := first["input"].([]any)
46 secondInput := second["input"].([]any)
47 require.Equal(t, firstInput, secondInput[:len(firstInput)])
48 require.Equal(t, []any{
49 map[string]any{"type": "reasoning", "encrypted_content": "opaque thought", "summary": []any{}},
50 map[string]any{"type": "function_call", "call_id": "call_1", "name": "shell", "arguments": `{"command":"echo hi"}`},
51 map[string]any{"type": "function_call_output", "call_id": "call_1", "output": "hi\n"},
52 }, secondInput[len(firstInput):])
53
54 require.Equal(t, []string{
55 "user: Run echo hi.",
56 `tool: shell {"command":"echo hi"}`,
57 "echo: hi\n",
58 "reply: It said hi.",
59 }, test.Log(t))
60
61 require.Equal(t, []string{"user", "busy", "tool", "echo", "reply", "idle"}, test.events.Kinds())
62
63 var turns []string
64
65 for _, record := range modelReadUsage(t, test.usagePath) {
66 turns = append(turns, record.Turn)
67 }
68
69 require.Equal(t, []string{"first", "next"}, turns)
70 }
71
72 func TestFirstRequestHoldsTheViewTheStateAndTheMessage(t *testing.T) {
73 server := newModelFake(t)
74 test := newTestTurnLoop(t, server)
75 _, err := test.memory.Append("user", "an earlier message")
76 require.NoError(t, err)
77
78 _, err = test.loop.Submit(terminalSource, "The new message.")
79 require.NoError(t, err)
80 test.events.WaitFor(t, "idle", 1)
81
82 require.Equal(t, []string{
83 "<chat>\n0+1|user: an earlier message\n",
84 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
85 "The new message.",
86 }, turnInputTexts(t, server.Call(0)))
87 }
88
89 func TestMessageDuringATurnJoinsTheNextRequest(t *testing.T) {
90 server := newModelFake(t)
91 arrived := make(chan struct{})
92 release := make(chan struct{})
93 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
94 if call.number == 0 {
95 close(arrived)
96 <-release
97 modelWriteEvents(w, modelCompletedEvent([]any{modelFunctionCallItem("fc_1", "call_1", "date", `{"id":0}`)}, 100, 0, 5))
98
99 return
100 }
101
102 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("done")}, 100, 0, 5))
103 }
104
105 test := newTestTurnLoop(t, server)
106 _, err := test.loop.Submit(terminalSource, "first")
107 require.NoError(t, err)
108
109 <-arrived
110
111 _, err = test.loop.Submit(terminalSource, "second, during the turn")
112 require.NoError(t, err)
113 close(release)
114
115 test.events.WaitFor(t, "idle", 1)
116
117 input := server.Call(1).body["input"].([]any)
118 require.Equal(t, map[string]any{
119 "role": "user",
120 "content": []any{map[string]any{"type": "input_text", "text": "second, during the turn"}},
121 }, input[len(input)-1])
122
123 time.Sleep(50 * time.Millisecond)
124 require.Equal(t, 2, server.Calls())
125 require.Equal(t, 1, test.events.Count("busy"))
126 }
127
128 func TestMessageDuringTheFinalRequestStartsANewTurn(t *testing.T) {
129 server := newModelFake(t)
130 arrived := make(chan struct{})
131 release := make(chan struct{})
132 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
133 if call.number == 0 {
134 close(arrived)
135 <-release
136 }
137
138 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("answer")}, 100, 0, 5))
139 }
140
141 test := newTestTurnLoop(t, server)
142 _, err := test.loop.Submit(terminalSource, "first")
143 require.NoError(t, err)
144
145 <-arrived
146
147 _, err = test.loop.Submit(terminalSource, "second")
148 require.NoError(t, err)
149 close(release)
150
151 test.events.WaitFor(t, "idle", 2)
152 require.Equal(t, 2, server.Calls())
153
154 texts := turnInputTexts(t, server.Call(1))
155 require.Equal(t, "second", texts[len(texts)-1])
156 require.Contains(t, strings.Join(texts, ""), "|reply: answer\n")
157 }
158
159 func TestTurnWaitsForTheSummaries(t *testing.T) {
160 server := newModelFake(t)
161 summarizer := newFakeSummarizer()
162 summarizer.gate = make(chan struct{})
163 test := newTestTurnLoopWith(t, server, summarizer)
164
165 _, err := test.memory.Append("echo", strings.Repeat("long output ", 100))
166 require.NoError(t, err)
167
168 _, err = test.loop.Submit(terminalSource, "go on")
169 require.NoError(t, err)
170
171 time.Sleep(50 * time.Millisecond)
172 require.Equal(t, 0, server.Calls())
173
174 close(summarizer.gate)
175 test.events.WaitFor(t, "idle", 1)
176 require.Equal(t, 1, server.Calls())
177 require.NotContains(t, turnInputTexts(t, server.Call(0))[0], viewPending)
178 }
179
180 func TestInterruptEndsTheTurn(t *testing.T) {
181 server := newModelFake(t)
182 arrived := make(chan struct{})
183 release := make(chan struct{})
184 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
185 if call.number == 0 {
186 close(arrived)
187 <-release
188 }
189
190 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("answer")}, 100, 0, 5))
191 }
192
193 test := newTestTurnLoop(t, server)
194 _, err := test.loop.Submit(terminalSource, "a long task")
195 require.NoError(t, err)
196
197 <-arrived
198 require.True(t, test.loop.Interrupt())
199 test.events.WaitFor(t, "idle", 1)
200 close(release)
201
202 require.Equal(t, "interrupted", test.events.Last("error"))
203 require.Equal(t, []string{"user: a long task"}, test.Log(t))
204 require.False(t, test.loop.Interrupt())
205
206 _, err = test.loop.Submit(terminalSource, "again")
207 require.NoError(t, err)
208 test.events.WaitFor(t, "idle", 2)
209 require.Equal(t, "reply: answer", test.Log(t)[2])
210 require.Empty(t, test.logs.String())
211 }
212
213 func TestFailedCallEndsTheTurnWithoutARetry(t *testing.T) {
214 server := newModelFake(t)
215 server.reply = func(w http.ResponseWriter, _ ModelFakeCall) {
216 http.Error(w, "busy", http.StatusServiceUnavailable)
217 }
218
219 test := newTestTurnLoop(t, server)
220 _, err := test.loop.Submit(terminalSource, "hello")
221 require.NoError(t, err)
222 test.events.WaitFor(t, "idle", 1)
223
224 time.Sleep(50 * time.Millisecond)
225 require.Equal(t, 1, server.Calls())
226 require.Contains(t, test.events.Last("error"), "HTTP 503")
227 require.Equal(t, 1, strings.Count(test.logs.String(), "\n"))
228 require.Contains(t, test.logs.String(), "<3>turn failed: ")
229 require.Contains(t, test.logs.String(), "HTTP 503")
230 require.NotContains(t, test.logs.String(), "hello")
231 }
232
233 func TestViewMessagesOnlyGrowBetweenTurns(t *testing.T) {
234 server := newModelFake(t)
235 test := newTestTurnLoop(t, server)
236 for n, text := range []string{"one", "two", "three"} {
237 _, err := test.loop.Submit(terminalSource, text)
238 require.NoError(t, err)
239 test.events.WaitFor(t, "idle", n+1)
240 }
241
242 first := server.Call(1).body["input"].([]any)
243 second := server.Call(2).body["input"].([]any)
244
245 require.Len(t, first, 4)
246 require.Len(t, second, 5)
247 require.Equal(t, first[:2], second[:2])
248 require.Equal(t, []string{
249 "<chat>\n",
250 "0+1|user: one\n1+1|reply: a line\n",
251 "2+1|user: two\n3+1|reply: a line\n",
252 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
253 "three",
254 }, turnInputTexts(t, server.Call(2)))
255 }
256
257 // TurnTestLoop is a loop that runs in the background of a test, with its parts.
258 type TurnTestLoop struct {
259 agent *Agent
260 memory *Memory
261 loop *TurnLoop
262 events *TurnEventRecorder
263 logs *SyncBuffer
264 usagePath string
265 }
266
267 // TurnEventRecorder keeps the events of a loop.
268 type TurnEventRecorder struct {
269 mutex sync.Mutex
270 events []Event
271 }
272
273 func newTestTurnLoop(t *testing.T, server *ModelFake) *TurnTestLoop {
274 t.Helper()
275
276 return newTestTurnLoopWith(t, server, newFakeSummarizer())
277 }
278
279 // newTestTurnLoopWith returns a running loop on a new agent whose builds use
280 // summarizer and whose commands run in a temporary directory.
281 func newTestTurnLoopWith(t *testing.T, server *ModelFake, summarizer Summarizer) *TurnTestLoop {
282 t.Helper()
283
284 agent, err := openAgent(filepath.Join(t.TempDir(), "agent"))
285 require.NoError(t, err)
286
287 agent.config.Workdir = t.TempDir()
288 clock := newTestClock()
289 memory, err := openMemory(t.Context(), t.TempDir(), clock.Now, summarizer, discardLog())
290 require.NoError(t, err)
291
292 test := &TurnTestLoop{
293 agent: agent,
294 memory: memory,
295 events: &TurnEventRecorder{},
296 logs: &SyncBuffer{},
297 usagePath: filepath.Join(t.TempDir(), "usage.jsonl"),
298 }
299
300 client := newModelClient(server.URL, ModelAPIKey(modelTestKey), agent.CacheKey(), test.usagePath)
301 whatsapp, _, _ := whatsappStartTest(t, t.TempDir(), clock)
302 test.loop, err = newTurnLoop(agent, memory, newTestSubagents(t), whatsapp, newTestSchedules(t, clock.Now), client, clock.Now, daemonTestLogger(test.logs), test.events.Add)
303 require.NoError(t, err)
304
305 ctx, cancel := context.WithCancel(t.Context())
306 done := make(chan error)
307 go func() {
308 done <- test.loop.Run(ctx)
309 }()
310
311 t.Cleanup(func() {
312 cancel()
313 require.ErrorIs(t, <-done, context.Canceled)
314 require.NoError(t, memory.Close())
315 })
316
317 return test
318 }
319
320 // Log returns the logged messages as "kind: text".
321 func (l *TurnTestLoop) Log(t *testing.T) []string {
322 t.Helper()
323
324 var lines []string
325
326 for id := range l.memory.Len() {
327 message, err := l.memory.Message(id)
328 require.NoError(t, err)
329
330 lines = append(lines, message.Kind+": "+message.Text)
331 }
332
333 return lines
334 }
335
336 // WaitFor waits until count events of kind came.
337 func (r *TurnEventRecorder) WaitFor(t *testing.T, kind string, count int) {
338 t.Helper()
339
340 require.Eventually(t, func() bool {
341 return r.Count(kind) >= count
342 }, 10*time.Second, time.Millisecond)
343 }
344
345 func (r *TurnEventRecorder) Kinds() []string {
346 r.mutex.Lock()
347 defer r.mutex.Unlock()
348
349 kinds := make([]string, len(r.events))
350 for n, event := range r.events {
351 kinds[n] = event.Kind
352 }
353
354 return kinds
355 }
356
357 func (r *TurnEventRecorder) Count(kind string) int {
358 count := 0
359 for _, each := range r.Kinds() {
360 if each == kind {
361 count++
362 }
363 }
364
365 return count
366 }
367
368 // Last returns the text of the last event of kind.
369 func (r *TurnEventRecorder) Last(kind string) string {
370 return r.LastEvent(kind).Text
371 }
372
373 // LastEvent returns the last event of kind.
374 func (r *TurnEventRecorder) LastEvent(kind string) Event {
375 r.mutex.Lock()
376 defer r.mutex.Unlock()
377
378 for n := len(r.events) - 1; n >= 0; n-- {
379 if r.events[n].Kind == kind {
380 return r.events[n]
381 }
382 }
383
384 return Event{}
385 }
386
387 func (r *TurnEventRecorder) Add(event Event) {
388 r.mutex.Lock()
389 defer r.mutex.Unlock()
390
391 r.events = append(r.events, event)
392 }
393
394 // turnInputTexts returns the text of each user message of the input of call.
395 func turnInputTexts(t *testing.T, call ModelFakeCall) []string {
396 t.Helper()
397
398 var body struct {
399 Input []struct {
400 Role string `json:"role"`
401 Content []struct {
402 Text string `json:"text"`
403 } `json:"content"`
404 } `json:"input"`
405 }
406
407 require.NoError(t, json.Unmarshal(call.raw, &body))
408
409 var texts []string
410
411 for _, item := range body.Input {
412 if item.Role != "user" {
413 continue
414 }
415
416 var text strings.Builder
417
418 for _, part := range item.Content {
419 text.WriteString(part.Text)
420 }
421
422 texts = append(texts, text.String())
423 }
424
425 return texts
426 }
427
428 func modelReasoningItem(id, encrypted string) map[string]any {
429 return map[string]any{
430 "type": "reasoning",
431 "id": id,
432 "encrypted_content": encrypted,
433 "summary": []any{},
434 }
435 }