]> git.giacomolli.uk - onechat/blob - turn_test.go
Format short daemon events with slog
[onechat] / turn_test.go
1 package main
2
3 import (
4 "context"
5 "encoding/json"
6 "log"
7 "net/http"
8 "path/filepath"
9 "strings"
10 "sync"
11 "testing"
12 "time"
13
14 "github.com/stretchr/testify/require"
15 )
16
17 func TestTurnRunsToolsUntilTheFinalAnswer(t *testing.T) {
18 server := newModelFake(t)
19 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
20 if call.number == 0 {
21 modelWriteEvents(w, modelCompletedEvent([]any{
22 modelReasoningItem("rs_1", "opaque thought"),
23 modelFunctionCallItem("fc_1", "call_1", "shell", `{"command":"echo hi"}`),
24 }, 100, 0, 5))
25
26 return
27 }
28
29 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("It said hi.")}, 120, 100, 5))
30 }
31
32 test := newTestTurnLoop(t, server)
33 _, err := test.loop.Submit(terminalSource, "Run echo hi.")
34 require.NoError(t, err)
35 test.events.WaitFor(t, "idle", 1)
36 require.Equal(t, 2, server.Calls())
37
38 first := server.Call(0).body
39 second := server.Call(1).body
40 instructions, err := test.agent.Prompt()
41 require.NoError(t, err)
42 require.Equal(t, instructions, first["instructions"])
43 require.Equal(t, test.agent.config.Model, first["model"])
44
45 // The second request begins with the whole first one.
46 firstInput := first["input"].([]any)
47 secondInput := second["input"].([]any)
48 require.Equal(t, firstInput, secondInput[:len(firstInput)])
49 require.Equal(t, []any{
50 map[string]any{"type": "reasoning", "encrypted_content": "opaque thought", "summary": []any{}},
51 map[string]any{"type": "function_call", "call_id": "call_1", "name": "shell", "arguments": `{"command":"echo hi"}`},
52 map[string]any{"type": "function_call_output", "call_id": "call_1", "output": "hi\n"},
53 }, secondInput[len(firstInput):])
54
55 require.Equal(t, []string{
56 "user: Run echo hi.",
57 `tool: shell {"command":"echo hi"}`,
58 "echo: hi\n",
59 "reply: It said hi.",
60 }, test.Log(t))
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(), "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 // Log returns the logged messages as "kind: text".
268 func (l *TurnTestLoop) Log(t *testing.T) []string {
269 t.Helper()
270
271 var lines []string
272
273 for id := range l.memory.Len() {
274 message, err := l.memory.Message(id)
275 require.NoError(t, err)
276
277 lines = append(lines, message.Kind+": "+message.Text)
278 }
279
280 return lines
281 }
282
283 // TurnEventRecorder keeps the events of a loop.
284 type TurnEventRecorder struct {
285 mutex sync.Mutex
286 events []Event
287 }
288
289 // WaitFor waits until count events of kind came.
290 func (r *TurnEventRecorder) WaitFor(t *testing.T, kind string, count int) {
291 t.Helper()
292
293 require.Eventually(t, func() bool {
294 return r.Count(kind) >= count
295 }, 10*time.Second, time.Millisecond)
296 }
297
298 func (r *TurnEventRecorder) Kinds() []string {
299 r.mutex.Lock()
300 defer r.mutex.Unlock()
301
302 kinds := make([]string, len(r.events))
303 for n, event := range r.events {
304 kinds[n] = event.Kind
305 }
306
307 return kinds
308 }
309
310 func (r *TurnEventRecorder) Count(kind string) int {
311 count := 0
312 for _, each := range r.Kinds() {
313 if each == kind {
314 count++
315 }
316 }
317
318 return count
319 }
320
321 // Last returns the text of the last event of kind.
322 func (r *TurnEventRecorder) Last(kind string) string {
323 return r.LastEvent(kind).Text
324 }
325
326 // LastEvent returns the last event of kind.
327 func (r *TurnEventRecorder) LastEvent(kind string) Event {
328 r.mutex.Lock()
329 defer r.mutex.Unlock()
330
331 for n := len(r.events) - 1; n >= 0; n-- {
332 if r.events[n].Kind == kind {
333 return r.events[n]
334 }
335 }
336
337 return Event{}
338 }
339
340 func (r *TurnEventRecorder) Add(event Event) {
341 r.mutex.Lock()
342 defer r.mutex.Unlock()
343
344 r.events = append(r.events, event)
345 }
346
347 func newTestTurnLoop(t *testing.T, server *ModelFake) *TurnTestLoop {
348 t.Helper()
349
350 return newTestTurnLoopWith(t, server, newFakeSummarizer())
351 }
352
353 // turnInputTexts returns the text of each user message of the input of call.
354 func turnInputTexts(t *testing.T, call ModelFakeCall) []string {
355 t.Helper()
356
357 var body struct {
358 Input []struct {
359 Role string `json:"role"`
360 Content []struct {
361 Text string `json:"text"`
362 } `json:"content"`
363 } `json:"input"`
364 }
365
366 require.NoError(t, json.Unmarshal(call.raw, &body))
367
368 var texts []string
369
370 for _, item := range body.Input {
371 if item.Role != "user" {
372 continue
373 }
374
375 var text strings.Builder
376
377 for _, part := range item.Content {
378 text.WriteString(part.Text)
379 }
380
381 texts = append(texts, text.String())
382 }
383
384 return texts
385 }
386
387 // newTestTurnLoopWith returns a running loop on a new agent whose builds use
388 // summarizer and whose commands run in a temporary directory.
389 func newTestTurnLoopWith(t *testing.T, server *ModelFake, summarizer Summarizer) *TurnTestLoop {
390 t.Helper()
391
392 agent, err := openAgent(t.TempDir())
393 require.NoError(t, err)
394
395 agent.config.Workdir = t.TempDir()
396 clock := newTestClock()
397 memory, err := openMemory(t.Context(), t.TempDir(), clock.Now, summarizer, discardLog())
398 require.NoError(t, err)
399
400 test := &TurnTestLoop{
401 agent: agent,
402 memory: memory,
403 events: &TurnEventRecorder{},
404 logs: &SyncBuffer{},
405 usagePath: filepath.Join(t.TempDir(), "usage.jsonl"),
406 }
407
408 client := newModelClient(server.URL, ModelAPIKey(modelTestKey), agent.CacheKey(), test.usagePath)
409 whatsapp, _, _ := whatsappStartTest(t, t.TempDir(), clock)
410 test.loop, err = newTurnLoop(agent, memory, newTestSubagents(t), whatsapp, newTestSchedules(t, clock.Now), client, clock.Now, log.New(test.logs, "", 0), test.events.Add)
411 require.NoError(t, err)
412
413 ctx, cancel := context.WithCancel(t.Context())
414 done := make(chan error)
415 go func() {
416 done <- test.loop.Run(ctx)
417 }()
418
419 t.Cleanup(func() {
420 cancel()
421 require.ErrorIs(t, <-done, context.Canceled)
422 require.NoError(t, memory.Close())
423 })
424
425 return test
426 }