]> git.giacomolli.uk - onechat/blob - turn_test.go
Share the Markdown marks and block parser
[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 require.Equal(t, []string{"user", "busy", "tool", "echo", "reply", "idle"}, test.events.Kinds())
61
62 var turns []string
63
64 for _, record := range modelReadUsage(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 := newModelFake(t)
73 test := newTestTurnLoop(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 }, turnInputTexts(t, server.Call(0)))
86 }
87
88 func TestMessageDuringATurnJoinsTheNextRequest(t *testing.T) {
89 server := newModelFake(t)
90 arrived := make(chan struct{})
91 release := make(chan struct{})
92 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
93 if call.number == 0 {
94 close(arrived)
95 <-release
96 modelWriteEvents(w, modelCompletedEvent([]any{modelFunctionCallItem("fc_1", "call_1", "date", `{"id":0}`)}, 100, 0, 5))
97
98 return
99 }
100
101 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("done")}, 100, 0, 5))
102 }
103
104 test := newTestTurnLoop(t, server)
105 _, err := test.loop.Submit(terminalSource, "first")
106 require.NoError(t, err)
107
108 <-arrived
109
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
116 input := server.Call(1).body["input"].([]any)
117 require.Equal(t, map[string]any{
118 "role": "user",
119 "content": []any{map[string]any{"type": "input_text", "text": "second, during the turn"}},
120 }, input[len(input)-1])
121
122 time.Sleep(50 * time.Millisecond)
123 require.Equal(t, 2, server.Calls())
124 require.Equal(t, 1, test.events.Count("busy"))
125 }
126
127 func TestMessageDuringTheFinalRequestStartsANewTurn(t *testing.T) {
128 server := newModelFake(t)
129 arrived := make(chan struct{})
130 release := make(chan struct{})
131 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
132 if call.number == 0 {
133 close(arrived)
134 <-release
135 }
136
137 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("answer")}, 100, 0, 5))
138 }
139
140 test := newTestTurnLoop(t, server)
141 _, err := test.loop.Submit(terminalSource, "first")
142 require.NoError(t, err)
143
144 <-arrived
145
146 _, err = test.loop.Submit(terminalSource, "second")
147 require.NoError(t, err)
148 close(release)
149
150 test.events.WaitFor(t, "idle", 2)
151 require.Equal(t, 2, server.Calls())
152
153 texts := turnInputTexts(t, server.Call(1))
154 require.Equal(t, "second", texts[len(texts)-1])
155 require.Contains(t, strings.Join(texts, ""), "|reply: answer\n")
156 }
157
158 func TestTurnWaitsForTheSummaries(t *testing.T) {
159 server := newModelFake(t)
160 summarizer := newFakeSummarizer()
161 summarizer.gate = make(chan struct{})
162 test := newTestTurnLoopWith(t, server, summarizer)
163
164 _, err := test.memory.Append("echo", strings.Repeat("long output ", 100))
165 require.NoError(t, err)
166
167 _, err = test.loop.Submit(terminalSource, "go on")
168 require.NoError(t, err)
169
170 time.Sleep(50 * time.Millisecond)
171 require.Equal(t, 0, server.Calls())
172
173 close(summarizer.gate)
174 test.events.WaitFor(t, "idle", 1)
175 require.Equal(t, 1, server.Calls())
176 require.NotContains(t, turnInputTexts(t, server.Call(0))[0], viewPending)
177 }
178
179 func TestInterruptEndsTheTurn(t *testing.T) {
180 server := newModelFake(t)
181 arrived := make(chan struct{})
182 release := make(chan struct{})
183 server.reply = func(w http.ResponseWriter, call ModelFakeCall) {
184 if call.number == 0 {
185 close(arrived)
186 <-release
187 }
188
189 modelWriteEvents(w, modelCompletedEvent([]any{modelMessageItem("answer")}, 100, 0, 5))
190 }
191
192 test := newTestTurnLoop(t, server)
193 _, err := test.loop.Submit(terminalSource, "a long task")
194 require.NoError(t, err)
195
196 <-arrived
197 require.True(t, test.loop.Interrupt())
198 test.events.WaitFor(t, "idle", 1)
199 close(release)
200
201 require.Equal(t, "interrupted", test.events.Last("error"))
202 require.Equal(t, []string{"user: a long task"}, test.Log(t))
203 require.False(t, test.loop.Interrupt())
204
205 _, err = test.loop.Submit(terminalSource, "again")
206 require.NoError(t, err)
207 test.events.WaitFor(t, "idle", 2)
208 require.Equal(t, "reply: answer", test.Log(t)[2])
209 require.Empty(t, test.logs.String())
210 }
211
212 func TestFailedCallEndsTheTurnWithoutARetry(t *testing.T) {
213 server := newModelFake(t)
214 server.reply = func(w http.ResponseWriter, _ ModelFakeCall) {
215 http.Error(w, "busy", http.StatusServiceUnavailable)
216 }
217
218 test := newTestTurnLoop(t, server)
219 _, err := test.loop.Submit(terminalSource, "hello")
220 require.NoError(t, err)
221 test.events.WaitFor(t, "idle", 1)
222
223 time.Sleep(50 * time.Millisecond)
224 require.Equal(t, 1, server.Calls())
225 require.Contains(t, test.events.Last("error"), "HTTP 503")
226 require.Equal(t, 1, strings.Count(test.logs.String(), "\n"))
227 require.Contains(t, test.logs.String(), "<3>turn failed: ")
228 require.Contains(t, test.logs.String(), "HTTP 503")
229 require.NotContains(t, test.logs.String(), "hello")
230 }
231
232 func TestViewMessagesOnlyGrowBetweenTurns(t *testing.T) {
233 server := newModelFake(t)
234 test := newTestTurnLoop(t, server)
235 for n, text := range []string{"one", "two", "three"} {
236 _, err := test.loop.Submit(terminalSource, text)
237 require.NoError(t, err)
238 test.events.WaitFor(t, "idle", n+1)
239 }
240
241 first := server.Call(1).body["input"].([]any)
242 second := server.Call(2).body["input"].([]any)
243
244 require.Len(t, first, 4)
245 require.Len(t, second, 5)
246 require.Equal(t, first[:2], second[:2])
247 require.Equal(t, []string{
248 "<chat>\n",
249 "0+1|user: one\n1+1|reply: a line\n",
250 "2+1|user: two\n3+1|reply: a line\n",
251 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
252 "three",
253 }, turnInputTexts(t, server.Call(2)))
254 }
255
256 // TurnTestLoop is a loop that runs in the background of a test, with its parts.
257 type TurnTestLoop struct {
258 agent *Agent
259 memory *Memory
260 loop *TurnLoop
261 events *TurnEventRecorder
262 logs *SyncBuffer
263 usagePath string
264 }
265
266 // Log returns the logged messages as "kind: text".
267 func (l *TurnTestLoop) Log(t *testing.T) []string {
268 t.Helper()
269
270 var lines []string
271
272 for id := range l.memory.Len() {
273 message, err := l.memory.Message(id)
274 require.NoError(t, err)
275
276 lines = append(lines, message.Kind+": "+message.Text)
277 }
278
279 return lines
280 }
281
282 // TurnEventRecorder keeps the events of a loop.
283 type TurnEventRecorder struct {
284 mutex sync.Mutex
285 events []Event
286 }
287
288 // WaitFor waits until count events of kind came.
289 func (r *TurnEventRecorder) 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 (r *TurnEventRecorder) Kinds() []string {
298 r.mutex.Lock()
299 defer r.mutex.Unlock()
300
301 kinds := make([]string, len(r.events))
302 for n, event := range r.events {
303 kinds[n] = event.Kind
304 }
305
306 return kinds
307 }
308
309 func (r *TurnEventRecorder) Count(kind string) int {
310 count := 0
311 for _, each := range r.Kinds() {
312 if each == kind {
313 count++
314 }
315 }
316
317 return count
318 }
319
320 // Last returns the text of the last event of kind.
321 func (r *TurnEventRecorder) Last(kind string) string {
322 return r.LastEvent(kind).Text
323 }
324
325 // LastEvent returns the last event of kind.
326 func (r *TurnEventRecorder) LastEvent(kind string) Event {
327 r.mutex.Lock()
328 defer r.mutex.Unlock()
329
330 for n := len(r.events) - 1; n >= 0; n-- {
331 if r.events[n].Kind == kind {
332 return r.events[n]
333 }
334 }
335
336 return Event{}
337 }
338
339 func (r *TurnEventRecorder) Add(event Event) {
340 r.mutex.Lock()
341 defer r.mutex.Unlock()
342
343 r.events = append(r.events, event)
344 }
345
346 func newTestTurnLoop(t *testing.T, server *ModelFake) *TurnTestLoop {
347 t.Helper()
348
349 return newTestTurnLoopWith(t, server, newFakeSummarizer())
350 }
351
352 // turnInputTexts returns the text of each user message of the input of call.
353 func turnInputTexts(t *testing.T, call ModelFakeCall) []string {
354 t.Helper()
355
356 var body struct {
357 Input []struct {
358 Role string `json:"role"`
359 Content []struct {
360 Text string `json:"text"`
361 } `json:"content"`
362 } `json:"input"`
363 }
364
365 require.NoError(t, json.Unmarshal(call.raw, &body))
366
367 var texts []string
368
369 for _, item := range body.Input {
370 if item.Role != "user" {
371 continue
372 }
373
374 var text strings.Builder
375
376 for _, part := range item.Content {
377 text.WriteString(part.Text)
378 }
379
380 texts = append(texts, text.String())
381 }
382
383 return texts
384 }
385
386 // newTestTurnLoopWith returns a running loop on a new agent whose builds use
387 // summarizer and whose commands run in a temporary directory.
388 func newTestTurnLoopWith(t *testing.T, server *ModelFake, summarizer Summarizer) *TurnTestLoop {
389 t.Helper()
390
391 agent, err := openAgent(t.TempDir())
392 require.NoError(t, err)
393
394 agent.config.Workdir = t.TempDir()
395 clock := newTestClock()
396 memory, err := openMemory(t.Context(), t.TempDir(), clock.Now, summarizer, discardLog())
397 require.NoError(t, err)
398
399 test := &TurnTestLoop{
400 agent: agent,
401 memory: memory,
402 events: &TurnEventRecorder{},
403 logs: &SyncBuffer{},
404 usagePath: filepath.Join(t.TempDir(), "usage.jsonl"),
405 }
406
407 client := newModelClient(server.URL, ModelAPIKey(modelTestKey), agent.CacheKey(), test.usagePath)
408 whatsapp, _, _ := whatsappStartTest(t, t.TempDir(), clock)
409 test.loop, err = newTurnLoop(agent, memory, newTestSubagents(t), whatsapp, newTestSchedules(t, clock.Now), client, clock.Now, daemonTestLogger(test.logs), test.events.Add)
410 require.NoError(t, err)
411
412 ctx, cancel := context.WithCancel(t.Context())
413 done := make(chan error)
414 go func() {
415 done <- test.loop.Run(ctx)
416 }()
417
418 t.Cleanup(func() {
419 cancel()
420 require.ErrorIs(t, <-done, context.Canceled)
421 require.NoError(t, memory.Close())
422 })
423
424 return test
425 }