]> git.giacomolli.uk - onechat/blob - rebuild.go
Name the test group in English
[onechat] / rebuild.go
1 package main
2
3 import (
4 "context"
5 "encoding/json"
6 "errors"
7 "fmt"
8 "io"
9 "io/fs"
10 "log"
11 "os"
12 "path/filepath"
13 "slices"
14 "strings"
15 "time"
16 )
17
18 // rebuild merges added messages with the log of agent, in date order, and
19 // builds a new memory beside it. The caller holds the token lock. The model
20 // uses the original agent prompt and cache key. A failed or stopped build
21 // removes the new directory and leaves the original directory unchanged.
22 // Success copies the other agent files, adds this run's usage, and swaps the
23 // directories, with the original kept as a timestamped backup. Message IDs are
24 // new positions in the merged log, but kind, text, and date stay exact.
25 func rebuild(ctx context.Context, agent *Agent, added []DatedMessage, client *ModelClient, summarizer Summarizer, out io.Writer) (backup string, err error) {
26 err = rebuildIdle(agent.dir)
27 if err != nil {
28 return "", fmt.Errorf("check rebuild work: %w", err)
29 }
30
31 old, err := rebuildReadLog(ctx, agent.dir)
32 if err != nil {
33 return "", fmt.Errorf("read original log: %w", err)
34 }
35
36 entries, err := rebuildMerge(old, added)
37 if err != nil {
38 return "", fmt.Errorf("merge rebuild messages: %w", err)
39 }
40
41 root, err := os.MkdirTemp(filepath.Dir(agent.dir), ".agent-rebuild-")
42 if err != nil {
43 return "", fmt.Errorf("create rebuild directory: %w", err)
44 }
45
46 defer func() {
47 closeErr := os.RemoveAll(root)
48 if closeErr != nil {
49 err = errors.Join(err, fmt.Errorf("remove rebuild directory: %w", closeErr))
50 }
51 }()
52
53 memory, err := openMemory(ctx, root, time.Now, summarizer, log.New(io.Discard, "", 0))
54 if err != nil {
55 return "", fmt.Errorf("open rebuilt memory: %w", err)
56 }
57
58 client.LimitCalls(buildCallLimit)
59 run := BuildRun{
60 memory: memory,
61 client: client,
62 usagePath: client.UsagePath(),
63 }
64
65 appended, stop := run.Run(ctx, entries)
66 sizes, err := memory.buildModelLines()
67 if stop == "" {
68 err = errors.Join(err, memory.Settle(ctx), memory.rebuildComplete(), memory.Flush(ctx))
69 }
70
71 err = errors.Join(err, memory.Close())
72 records, usageErr := buildReadUsage(client.UsagePath())
73 err = errors.Join(err, usageErr, buildWriteReport(out, appended, stop, records, sizes))
74 if stop != "" {
75 err = errors.Join(err, fmt.Errorf("rebuild stopped: %s", stop))
76 }
77
78 if err != nil {
79 return "", fmt.Errorf("finish rebuilt memory: %w", err)
80 }
81
82 err = rebuildCopy(agent.dir, root)
83 if err != nil {
84 return "", fmt.Errorf("copy other agent files: %w", err)
85 }
86
87 usage, err := os.ReadFile(client.UsagePath())
88 if errors.Is(err, os.ErrNotExist) {
89 err = nil
90 }
91
92 if err != nil {
93 return "", fmt.Errorf("read rebuild usage: %w", err)
94 }
95
96 if len(usage) > 0 {
97 err = rebuildAddUsage(filepath.Join(root, "usage.jsonl"), usage)
98 if err != nil {
99 return "", fmt.Errorf("append rebuild usage: %w", err)
100 }
101 }
102
103 if ctx.Err() != nil {
104 return "", ctx.Err()
105 }
106
107 backup = agent.dir + "-backup-" + time.Now().Format("20060102-150405.000000000")
108 err = os.Rename(agent.dir, backup)
109 if err != nil {
110 return "", fmt.Errorf("back up agent: %w", err)
111 }
112
113 err = os.Rename(root, agent.dir)
114 if err != nil {
115 restoreErr := os.Rename(backup, agent.dir)
116 if restoreErr != nil {
117 err = errors.Join(err, fmt.Errorf("restore original agent: %w", restoreErr))
118 }
119
120 return "", fmt.Errorf("install rebuilt agent: %w", err)
121 }
122
123 return backup, nil
124 }
125
126 // rebuildIdle refuses an agent with unfinished subagents, including ones that a
127 // daemon restart must resume. A missing work file lists no work.
128 func rebuildIdle(root string) error {
129 data, err := os.ReadFile(filepath.Join(root, subagentWorkFileName))
130 if errors.Is(err, os.ErrNotExist) {
131 return nil
132 }
133
134 if err != nil {
135 return fmt.Errorf("read running subagents: %w", err)
136 }
137
138 var work SubagentWorkFile
139 err = json.Unmarshal(data, &work)
140 if err != nil {
141 return fmt.Errorf("decode running subagents: %w", err)
142 }
143
144 if len(work.Running) > 0 {
145 return errors.New("rebuild refused: subagents have unfinished work")
146 }
147
148 return nil
149 }
150
151 // rebuildReadLog reads and validates the day files without a write or an index
152 // repair. Even an incomplete final line is an error, not a repair.
153 func rebuildReadLog(ctx context.Context, root string) ([]DatedMessage, error) {
154 paths, err := filepath.Glob(filepath.Join(root, "main", "*.jsonl"))
155 if err != nil {
156 return nil, fmt.Errorf("list message files: %w", err)
157 }
158
159 slices.Sort(paths)
160 var entries []DatedMessage
161 for _, path := range paths {
162 var day uint32
163
164 day, err = journalDayNumber(strings.TrimSuffix(filepath.Base(path), ".jsonl"))
165 if err != nil {
166 return nil, fmt.Errorf("parse message day: %w", err)
167 }
168
169 err = journalScanFile(ctx, path, day, 0, false, func(line []byte, _ Position) error {
170 var message Message
171 err = json.Unmarshal(line, &message)
172 if err != nil {
173 return fmt.Errorf("decode message: %w", err)
174 }
175
176 err = logValidateMessage(message)
177 if err != nil {
178 return fmt.Errorf("validate message: %w", err)
179 }
180
181 if message.ID != len(entries) {
182 return fmt.Errorf("expected message %d, found %d", len(entries), message.ID)
183 }
184
185 entries = append(entries, DatedMessage{
186 Kind: message.Kind,
187 Text: message.Text,
188 Date: message.Date,
189 })
190
191 return nil
192 })
193 if err != nil {
194 return nil, fmt.Errorf("read message file: %w", err)
195 }
196 }
197
198 return entries, nil
199 }
200
201 // rebuildMerge preserves the input order for equal dates, with old messages
202 // before added messages. Each date must be explicit and valid.
203 func rebuildMerge(old, added []DatedMessage) ([]DatedMessage, error) {
204 entries := append(slices.Clone(old), added...)
205 dates := make(map[string]time.Time)
206 for _, entry := range entries {
207 err := logValidateMessage(Message{
208 Kind: entry.Kind,
209 Text: entry.Text,
210 Size: len(entry.Text),
211 Date: entry.Date,
212 })
213 if err != nil {
214 return nil, fmt.Errorf("validate dated message: %w", err)
215 }
216
217 var date time.Time
218 date, err = time.Parse(time.RFC3339Nano, entry.Date)
219 if err != nil {
220 return nil, fmt.Errorf("parse merge date: %w", err)
221 }
222
223 dates[entry.Date] = date
224 }
225
226 slices.SortStableFunc(entries, func(a, b DatedMessage) int {
227 return dates[a.Date].Compare(dates[b.Date])
228 })
229
230 return entries, nil
231 }
232
233 // rebuildCopy copies the other agent directories and regular files. It excludes
234 // the memory and its lock, and refuses all other file types.
235 func rebuildCopy(source, target string) error {
236 err := filepath.WalkDir(source, func(path string, entry fs.DirEntry, err error) error {
237 if err != nil {
238 return fmt.Errorf("read agent entry: %w", err)
239 }
240
241 relative, err := filepath.Rel(source, path)
242 if err != nil {
243 return fmt.Errorf("find agent relative path: %w", err)
244 }
245
246 if relative == "." {
247 return nil
248 }
249
250 switch relative {
251 case "main", "tree", "index":
252 return filepath.SkipDir
253 case "lock", "view.json":
254 return nil
255 }
256
257 destination := filepath.Join(target, relative)
258 switch {
259 case entry.IsDir():
260 err = os.Mkdir(destination, 0o700)
261 if err != nil {
262 return fmt.Errorf("create agent directory: %w", err)
263 }
264 case entry.Type().IsRegular():
265 var data []byte
266 data, err = os.ReadFile(path)
267 if err != nil {
268 return fmt.Errorf("read agent file: %w", err)
269 }
270
271 err = os.WriteFile(destination, data, 0o600)
272 if err != nil {
273 return fmt.Errorf("copy agent file: %w", err)
274 }
275 default:
276 return fmt.Errorf("copy agent: unsupported file type for %s", relative)
277 }
278
279 return nil
280 })
281 if err != nil {
282 return fmt.Errorf("copy agent files: %w", err)
283 }
284
285 return nil
286 }
287
288 // rebuildAddUsage keeps both the original usage and the calls of this run.
289 func rebuildAddUsage(path string, data []byte) (err error) {
290 file, err := os.OpenFile(path, os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0o600)
291 if err != nil {
292 return fmt.Errorf("open rebuilt usage: %w", err)
293 }
294
295 defer func() {
296 closeErr := file.Close()
297 if closeErr != nil {
298 err = errors.Join(err, fmt.Errorf("close rebuilt usage: %w", closeErr))
299 }
300 }()
301
302 _, err = file.Write(data)
303 if err != nil {
304 return fmt.Errorf("append rebuilt usage: %w", err)
305 }
306
307 return nil
308 }