]> git.giacomolli.uk - onechat/blob - turn_test.go
Lay out turn requests for the prompt cache
[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("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
64 func TestFirstRequestHoldsTheViewTheStateAndTheMessage(t *testing.T) {
65 server := newFakeResponses(t)
66 test := newTestLoop(t, server)
67 _, err := test.memory.Append("user", "an earlier message")
68 require.NoError(t, err)
69
70 _, err = test.loop.Submit("The new message.")
71 require.NoError(t, err)
72 test.events.WaitFor(t, "idle", 1)
73
74 require.Equal(t, []string{
75 "<chat>\n0+1|user: an earlier message\n",
76 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
77 "The new message.",
78 }, inputTexts(t, server.Call(0)))
79 }
80
81 func TestMessageDuringATurnJoinsTheNextRequest(t *testing.T) {
82 server := newFakeResponses(t)
83 arrived := make(chan struct{})
84 release := make(chan struct{})
85 server.reply = func(w http.ResponseWriter, call fakeCall) {
86 if call.number == 0 {
87 close(arrived)
88 <-release
89 writeEvents(w, completedEvent([]any{functionCallItem("fc_1", "call_1", "date", `{"id":0}`)}, 100, 0, 5))
90
91 return
92 }
93
94 writeEvents(w, completedEvent([]any{messageItem("done")}, 100, 0, 5))
95 }
96
97 test := newTestLoop(t, server)
98 _, err := test.loop.Submit("first")
99 require.NoError(t, err)
100
101 <-arrived
102 _, err = test.loop.Submit("second, during the turn")
103 require.NoError(t, err)
104 close(release)
105
106 test.events.WaitFor(t, "idle", 1)
107 input := server.Call(1).body["input"].([]any)
108 require.Equal(t, map[string]any{
109 "role": "user",
110 "content": []any{map[string]any{"type": "input_text", "text": "second, during the turn"}},
111 }, input[len(input)-1])
112
113 time.Sleep(50 * time.Millisecond)
114 require.Equal(t, 2, server.Calls())
115 require.Equal(t, 1, test.events.Count("busy"))
116 }
117
118 func TestMessageDuringTheFinalRequestStartsANewTurn(t *testing.T) {
119 server := newFakeResponses(t)
120 arrived := make(chan struct{})
121 release := make(chan struct{})
122 server.reply = func(w http.ResponseWriter, call fakeCall) {
123 if call.number == 0 {
124 close(arrived)
125 <-release
126 }
127
128 writeEvents(w, completedEvent([]any{messageItem("answer")}, 100, 0, 5))
129 }
130
131 test := newTestLoop(t, server)
132 _, err := test.loop.Submit("first")
133 require.NoError(t, err)
134
135 <-arrived
136 _, err = test.loop.Submit("second")
137 require.NoError(t, err)
138 close(release)
139
140 test.events.WaitFor(t, "idle", 2)
141 require.Equal(t, 2, server.Calls())
142
143 texts := inputTexts(t, server.Call(1))
144 require.Equal(t, "second", texts[len(texts)-1])
145 require.Contains(t, strings.Join(texts, ""), "|reply: answer\n")
146 }
147
148 func TestTurnWaitsForTheSummaries(t *testing.T) {
149 server := newFakeResponses(t)
150 summarizer := newFakeSummarizer()
151 summarizer.gate = make(chan struct{})
152 test := newTestLoopWith(t, server, summarizer, false)
153
154 _, err := test.memory.Append("echo", strings.Repeat("long output ", 100))
155 require.NoError(t, err)
156
157 _, err = test.loop.Submit("go on")
158 require.NoError(t, err)
159
160 time.Sleep(50 * time.Millisecond)
161 require.Equal(t, 0, server.Calls())
162
163 close(summarizer.gate)
164 test.events.WaitFor(t, "idle", 1)
165 require.Equal(t, 1, server.Calls())
166 require.NotContains(t, inputTexts(t, server.Call(0))[0], pending)
167 }
168
169 func TestInterruptEndsTheTurn(t *testing.T) {
170 server := newFakeResponses(t)
171 arrived := make(chan struct{})
172 release := make(chan struct{})
173 server.reply = func(w http.ResponseWriter, call fakeCall) {
174 if call.number == 0 {
175 close(arrived)
176 <-release
177 }
178
179 writeEvents(w, completedEvent([]any{messageItem("answer")}, 100, 0, 5))
180 }
181
182 test := newTestLoop(t, server)
183 _, err := test.loop.Submit("a long task")
184 require.NoError(t, err)
185
186 <-arrived
187 require.True(t, test.loop.Interrupt())
188 test.events.WaitFor(t, "idle", 1)
189 close(release)
190
191 require.Equal(t, "interrupted", test.events.Last("error"))
192 require.Equal(t, []string{"user: a long task"}, test.Log(t))
193 require.False(t, test.loop.Interrupt())
194
195 _, err = test.loop.Submit("again")
196 require.NoError(t, err)
197 test.events.WaitFor(t, "idle", 2)
198 require.Equal(t, "reply: answer", test.Log(t)[2])
199 }
200
201 func TestFailedCallEndsTheTurnWithoutARetry(t *testing.T) {
202 server := newFakeResponses(t)
203 server.reply = func(w http.ResponseWriter, call fakeCall) {
204 http.Error(w, "busy", http.StatusServiceUnavailable)
205 }
206
207 test := newTestLoop(t, server)
208 _, err := test.loop.Submit("hello")
209 require.NoError(t, err)
210 test.events.WaitFor(t, "idle", 1)
211
212 time.Sleep(50 * time.Millisecond)
213 require.Equal(t, 1, server.Calls())
214 require.Contains(t, test.events.Last("error"), "HTTP 503")
215 }
216
217 func TestViewMessagesOnlyGrowBetweenTurns(t *testing.T) {
218 server := newFakeResponses(t)
219 test := newTestLoop(t, server)
220 for n, text := range []string{"one", "two", "three"} {
221 _, err := test.loop.Submit(text)
222 require.NoError(t, err)
223 test.events.WaitFor(t, "idle", n+1)
224 }
225
226 first := server.Call(1).body["input"].([]any)
227 second := server.Call(2).body["input"].([]any)
228 require.Len(t, first, 4)
229 require.Len(t, second, 5)
230 require.Equal(t, first[:2], second[:2])
231 require.Equal(t, []string{
232 "<chat>\n",
233 "0+1|user: one\n1+1|reply: a line\n",
234 "2+1|user: two\n3+1|reply: a line\n",
235 "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00",
236 "three",
237 }, inputTexts(t, server.Call(2)))
238 }
239
240 func TestWarmUpEndsWhereTheNextTurnReadsTheCache(t *testing.T) {
241 server := newFakeResponses(t)
242 server.reply = func(w http.ResponseWriter, call fakeCall) {
243 if call.number == 1 {
244 writeEvents(w, map[string]any{"type": "response.output_item.added", "item": reasoningItem("rs_1", "")})
245 return
246 }
247
248 writeEvents(w, completedEvent([]any{messageItem("answer")}, 100, 0, 5))
249 }
250
251 test := newTestLoopWith(t, server, newFakeSummarizer(), true)
252 _, err := test.loop.Submit("one")
253 require.NoError(t, err)
254 require.Eventually(t, func() bool {
255 return server.Calls() == 2
256 }, 10*time.Second, time.Millisecond)
257
258 _, err = test.loop.Submit("two")
259 require.NoError(t, err)
260 test.events.WaitFor(t, "idle", 2)
261 require.Eventually(t, func() bool {
262 return server.Calls() == 4
263 }, 10*time.Second, time.Millisecond)
264
265 warmUp := server.Call(1).body
266 require.Equal(t, server.Call(0).body["instructions"], warmUp["instructions"])
267 require.Equal(t, server.Call(0).body["tools"], warmUp["tools"])
268 require.Equal(t, []string{"<chat>\n", "0+1|user: one\n1+1|reply: answer\n"}, inputTexts(t, server.Call(1)))
269
270 // The next turn begins with the whole warm-up, so its message end after
271 // the view is the end of the warm-up entry.
272 warmInput := warmUp["input"].([]any)
273 turnInput := server.Call(2).body["input"].([]any)
274 require.Equal(t, warmInput, turnInput[:len(warmInput)])
275 require.Equal(t, "</chat>\n\nNow: Thursday 2026-01-01 12:00 +00:00", inputTexts(t, server.Call(2))[len(warmInput)])
276
277 turns := []string{}
278 for _, record := range readUsage(t, test.usagePath) {
279 turns = append(turns, record.Turn)
280 }
281
282 require.Equal(t, []string{"first", "warmup", "first", "warmup"}, turns)
283 require.Equal(t, []string{"user: one", "reply: answer", "user: two", "reply: answer"}, test.Log(t))
284 }
285
286 // testLoop is a loop that runs in the background of a test, with its parts.
287 type testLoop struct {
288 agent *Agent
289 memory *Memory
290 loop *loop
291 events *eventRecorder
292 usagePath string
293 }
294
295 // Log returns the logged messages as "kind: text".
296 func (l *testLoop) Log(t *testing.T) []string {
297 t.Helper()
298
299 var lines []string
300 for id := range l.memory.Len() {
301 message, err := l.memory.Message(id)
302 require.NoError(t, err)
303
304 lines = append(lines, message.Kind+": "+message.Text)
305 }
306
307 return lines
308 }
309
310 // eventRecorder keeps the events of a loop.
311 type eventRecorder struct {
312 mutex sync.Mutex
313 events []Event
314 }
315
316 func (r *eventRecorder) Add(event Event) {
317 r.mutex.Lock()
318 defer r.mutex.Unlock()
319
320 r.events = append(r.events, event)
321 }
322
323 func (r *eventRecorder) Kinds() []string {
324 r.mutex.Lock()
325 defer r.mutex.Unlock()
326
327 kinds := make([]string, len(r.events))
328 for n, event := range r.events {
329 kinds[n] = event.Kind
330 }
331
332 return kinds
333 }
334
335 func (r *eventRecorder) Count(kind string) int {
336 count := 0
337 for _, each := range r.Kinds() {
338 if each == kind {
339 count++
340 }
341 }
342
343 return count
344 }
345
346 // Last returns the text of the last event of kind.
347 func (r *eventRecorder) Last(kind string) string {
348 r.mutex.Lock()
349 defer r.mutex.Unlock()
350
351 for n := len(r.events) - 1; n >= 0; n-- {
352 if r.events[n].Kind == kind {
353 return r.events[n].Text
354 }
355 }
356
357 return ""
358 }
359
360 // WaitFor waits until count events of kind came.
361 func (r *eventRecorder) WaitFor(t *testing.T, kind string, count int) {
362 t.Helper()
363
364 require.Eventually(t, func() bool {
365 return r.Count(kind) >= count
366 }, 10*time.Second, time.Millisecond)
367 }
368
369 func newTestLoop(t *testing.T, server *fakeResponses) *testLoop {
370 t.Helper()
371
372 return newTestLoopWith(t, server, newFakeSummarizer(), false)
373 }
374
375 // newTestLoopWith returns a running loop on a new agent whose builds use
376 // summarizer and whose commands run in a temporary directory. warmUp turns the
377 // warm-up on.
378 func newTestLoopWith(t *testing.T, server *fakeResponses, summarizer Summarizer, warmUp bool) *testLoop {
379 t.Helper()
380
381 agent, err := openAgent(t.TempDir())
382 require.NoError(t, err)
383
384 agent.config.Workdir = t.TempDir()
385 clock := newTestClock()
386 memory, err := openMemory(t.Context(), t.TempDir(), clock.Now, summarizer)
387 require.NoError(t, err)
388
389 test := &testLoop{
390 agent: agent,
391 memory: memory,
392 events: &eventRecorder{},
393 usagePath: filepath.Join(t.TempDir(), "usage.jsonl"),
394 }
395
396 client := newClient(server.URL, apiKey(testKey), agent.CacheKey(), test.usagePath)
397 test.loop, err = newLoop(agent, memory, client, clock.Now, test.events.Add)
398 require.NoError(t, err)
399
400 test.loop.warmUp = warmUp
401 ctx, cancel := context.WithCancel(t.Context())
402 done := make(chan error)
403 go func() {
404 done <- test.loop.Run(ctx)
405 }()
406
407 t.Cleanup(func() {
408 cancel()
409 require.True(t, errors.Is(<-done, context.Canceled))
410 require.NoError(t, memory.Close())
411 })
412
413 return test
414 }
415
416 // inputTexts returns the text of each user message of the input of call.
417 func inputTexts(t *testing.T, call fakeCall) []string {
418 t.Helper()
419
420 var body struct {
421 Input []struct {
422 Role string `json:"role"`
423 Content []struct {
424 Text string `json:"text"`
425 } `json:"content"`
426 } `json:"input"`
427 }
428
429 require.NoError(t, json.Unmarshal(call.raw, &body))
430
431 var texts []string
432 for _, item := range body.Input {
433 if item.Role != "user" {
434 continue
435 }
436
437 var text strings.Builder
438 for _, part := range item.Content {
439 text.WriteString(part.Text)
440 }
441
442 texts = append(texts, text.String())
443 }
444
445 return texts
446 }
447
448 func reasoningItem(id, encrypted string) map[string]any {
449 return map[string]any{
450 "type": "reasoning",
451 "id": id,
452 "encrypted_content": encrypted,
453 "summary": []any{},
454 }
455 }
456
457 func functionCallItem(id, callID, name, arguments string) map[string]any {
458 return map[string]any{
459 "type": "function_call",
460 "id": id,
461 "call_id": callID,
462 "name": name,
463 "arguments": arguments,
464 }
465 }