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