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