| 1 | package main |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "os" |
| 8 | "path/filepath" |
| 9 | "strings" |
| 10 | "sync" |
| 11 | "testing" |
| 12 | |
| 13 | "reasonix/internal/agent" |
| 14 | "reasonix/internal/control" |
| 15 | "reasonix/internal/event" |
| 16 | "reasonix/internal/provider" |
| 17 | "reasonix/internal/tool" |
| 18 | ) |
| 19 | |
| 20 | type promptResumeCtrl struct { |
| 21 | history []provider.Message |
| 22 | resumed *agent.Session |
| 23 | path string |
| 24 | } |
| 25 | |
| 26 | func (c *promptResumeCtrl) History() []provider.Message { |
| 27 | return append([]provider.Message(nil), c.history...) |
| 28 | } |
| 29 | |
| 30 | func (c *promptResumeCtrl) Resume(s *agent.Session, path string) { |
| 31 | c.resumed = s |
| 32 | c.path = path |
| 33 | } |
| 34 | |
| 35 | func (c *promptResumeCtrl) SetSessionPath(path string) { |
| 36 | c.path = path |
| 37 | } |
| 38 | |
| 39 | func TestSessionWithFreshSystemPromptPreservesLoadedRewriteBaseline(t *testing.T) { |
| 40 | path := filepath.Join(t.TempDir(), "session.jsonl") |
| 41 | s := agent.NewSession("old sys") |
| 42 | s.Add(provider.Message{Role: provider.RoleUser, Content: "first"}) |
| 43 | s.Add(provider.Message{Role: provider.RoleAssistant, ToolCalls: []provider.ToolCall{{ID: "tool-1", Name: "read_file", Arguments: "{}"}}}) |
| 44 | s.Add(provider.Message{Role: provider.RoleTool, ToolCallID: "tool-1", Name: "read_file", Content: strings.Repeat("detail ", 100)}) |
| 45 | s.Add(provider.Message{Role: provider.RoleAssistant, Content: "done"}) |
| 46 | if err := s.Save(path); err != nil { |
| 47 | t.Fatalf("Save base: %v", err) |
| 48 | } |
| 49 | |
| 50 | loaded, err := agent.LoadSession(path) |
| 51 | if err != nil { |
| 52 | t.Fatalf("LoadSession: %v", err) |
| 53 | } |
| 54 | resumed := sessionWithFreshSystemPrompt(loaded, "new sys") |
| 55 | msgs := resumed.Snapshot() |
| 56 | msgs[3].Content = "[elided tool result]" |
| 57 | resumed.Replace(msgs) |
| 58 | if err := resumed.SaveRewrite(path); err != nil { |
| 59 | t.Fatalf("SaveRewrite fresh-system resume: %v", err) |
| 60 | } |
| 61 | |
| 62 | reloaded, err := agent.LoadSession(path) |
| 63 | if err != nil { |
| 64 | t.Fatalf("LoadSession rewritten: %v", err) |
| 65 | } |
| 66 | if got := reloaded.Messages[0].Content; got != "new sys" { |
| 67 | t.Fatalf("system prompt after rewrite = %q, want new sys", got) |
| 68 | } |
| 69 | if got := reloaded.Messages[3].Content; got != "[elided tool result]" { |
| 70 | t.Fatalf("tool result after rewrite = %q, want elided", got) |
| 71 | } |
| 72 | if matches, err := filepath.Glob(filepath.Join(filepath.Dir(path), "*-recovery-*.jsonl")); err != nil || len(matches) != 0 { |
| 73 | t.Fatalf("recovery branches after owned resume rewrite = %v err=%v, want none", matches, err) |
| 74 | } |
| 75 | } |
| 76 | |
| 77 | func TestMigratedSessionWithoutSystemPromptPersistsThroughDesktopSwitch(t *testing.T) { |
| 78 | isolateDesktopUserDirs(t) |
| 79 | src := t.TempDir() |
| 80 | dest := t.TempDir() |
| 81 | const legacy = `{"role":"user","content":"recovered after downgrade"} |
| 82 | {"role":"assistant","content":"legacy answer"} |
| 83 | ` |
| 84 | if err := os.WriteFile(filepath.Join(src, "desktop-legacy.jsonl"), []byte(legacy), 0o644); err != nil { |
| 85 | t.Fatalf("write legacy session: %v", err) |
| 86 | } |
| 87 | if n, err := agent.MigrateLegacySessions(src, dest, nil); err != nil || n != 1 { |
| 88 | t.Fatalf("MigrateLegacySessions: n=%d err=%v", n, err) |
| 89 | } |
| 90 | |
| 91 | path := filepath.Join(dest, "desktop-legacy.jsonl") |
| 92 | loaded, err := agent.LoadSession(path) |
| 93 | if err != nil { |
| 94 | t.Fatalf("LoadSession migrated: %v", err) |
| 95 | } |
| 96 | const freshSystem = "current deterministic system prompt" |
| 97 | prov := &capturingProvider{} |
| 98 | exec := agent.New(prov, tool.NewRegistry(), agent.NewSession(freshSystem), agent.Options{}, event.Discard) |
| 99 | ctrl := control.New(control.Options{ |
| 100 | Runner: exec, |
| 101 | Executor: exec, |
| 102 | SystemPrompt: freshSystem, |
| 103 | SessionDir: dest, |
| 104 | SessionPath: path, |
| 105 | Label: "migrated", |
| 106 | Sink: event.Discard, |
| 107 | }) |
| 108 | defer ctrl.Close() |
| 109 | resumeLoadedSessionAndGoal(ctrl, loaded, path, "") |
| 110 | |
| 111 | if history := ctrl.History(); len(history) == 0 || history[0].Role != provider.RoleSystem || history[0].Content != freshSystem { |
| 112 | t.Fatalf("resumed history does not start with the fresh system prompt: %+v", history) |
| 113 | } |
| 114 | if err := ctrl.RunTurn(context.Background(), "new desktop turn"); err != nil { |
| 115 | t.Fatalf("RunTurn: %v", err) |
| 116 | } |
| 117 | |
| 118 | active := &WorkspaceTab{ |
| 119 | ID: "legacy", |
| 120 | Ctrl: ctrl, |
| 121 | Scope: "global", |
| 122 | SessionPath: path, |
| 123 | Ready: true, |
| 124 | disabledMCP: map[string]ServerView{}, |
| 125 | } |
| 126 | target := &WorkspaceTab{ |
| 127 | ID: "target", |
| 128 | Scope: "global", |
| 129 | Ready: true, |
| 130 | disabledMCP: map[string]ServerView{}, |
| 131 | } |
| 132 | app := &App{ |
| 133 | tabs: map[string]*WorkspaceTab{"legacy": active, "target": target}, |
| 134 | tabOrder: []string{"legacy", "target"}, |
| 135 | activeTabID: "legacy", |
| 136 | } |
| 137 | if err := app.SetActiveTab("target"); err != nil { |
| 138 | t.Fatalf("SetActiveTab: %v", err) |
| 139 | } |
| 140 | |
| 141 | reloaded, err := agent.LoadSession(path) |
| 142 | if err != nil { |
| 143 | t.Fatalf("LoadSession after switch: %v", err) |
| 144 | } |
| 145 | got := reloaded.Snapshot() |
| 146 | if len(got) != 5 { |
| 147 | t.Fatalf("reloaded message count = %d, want 5: %+v", len(got), got) |
| 148 | } |
| 149 | if got[0].Role != provider.RoleSystem || got[0].Content != freshSystem { |
| 150 | t.Fatalf("reloaded system prompt = %+v, want %q", got[0], freshSystem) |
| 151 | } |
| 152 | if got[3].Role != provider.RoleUser || got[3].Content != "new desktop turn" { |
| 153 | t.Fatalf("reloaded new user turn = %+v", got[3]) |
| 154 | } |
| 155 | if got[4].Role != provider.RoleAssistant || got[4].Content != "ok" { |
| 156 | t.Fatalf("reloaded assistant turn = %+v", got[4]) |
| 157 | } |
| 158 | if matches, err := filepath.Glob(filepath.Join(dest, "*-recovery-*.jsonl")); err != nil || len(matches) != 0 { |
| 159 | t.Fatalf("recovery branches after migrated switch = %v err=%v, want none", matches, err) |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | func TestParallelDesktopTabsPersistCompleteTranscriptsAcrossReload(t *testing.T) { |
| 164 | isolateDesktopUserDirs(t) |
| 165 | const ( |
| 166 | tabCount = 12 |
| 167 | turnsPerTab = 6 |
| 168 | systemText = "stable desktop system prompt" |
| 169 | ) |
| 170 | dir := t.TempDir() |
| 171 | tabs := make(map[string]*WorkspaceTab, tabCount) |
| 172 | tabOrder := make([]string, 0, tabCount) |
| 173 | controllers := make([]*control.Controller, 0, tabCount) |
| 174 | |
| 175 | for tabIndex := 0; tabIndex < tabCount; tabIndex++ { |
| 176 | id := fmt.Sprintf("parallel-%02d", tabIndex) |
| 177 | path := filepath.Join(dir, id+".jsonl") |
| 178 | prov := &capturingProvider{} |
| 179 | exec := agent.New(prov, tool.NewRegistry(), agent.NewSession(systemText), agent.Options{}, event.Discard) |
| 180 | ctrl := control.New(control.Options{ |
| 181 | Runner: exec, |
| 182 | Executor: exec, |
| 183 | SystemPrompt: systemText, |
| 184 | SessionDir: dir, |
| 185 | SessionPath: path, |
| 186 | Label: id, |
| 187 | Sink: event.Discard, |
| 188 | }) |
| 189 | controllers = append(controllers, ctrl) |
| 190 | tabOrder = append(tabOrder, id) |
| 191 | tabs[id] = &WorkspaceTab{ |
| 192 | ID: id, |
| 193 | Ctrl: ctrl, |
| 194 | Scope: "global", |
| 195 | SessionPath: path, |
| 196 | Ready: true, |
| 197 | disabledMCP: map[string]ServerView{}, |
| 198 | } |
| 199 | } |
| 200 | app := &App{ |
| 201 | tabs: tabs, |
| 202 | tabOrder: tabOrder, |
| 203 | activeTabID: tabOrder[0], |
| 204 | } |
| 205 | |
| 206 | start := make(chan struct{}) |
| 207 | errs := make(chan error, tabCount+1) |
| 208 | var wg sync.WaitGroup |
| 209 | for tabIndex, ctrl := range controllers { |
| 210 | tabIndex, ctrl := tabIndex, ctrl |
| 211 | wg.Add(1) |
| 212 | go func() { |
| 213 | defer wg.Done() |
| 214 | <-start |
| 215 | for turn := 0; turn < turnsPerTab; turn++ { |
| 216 | input := fmt.Sprintf("tab-%02d-turn-%02d", tabIndex, turn) |
| 217 | if err := ctrl.RunTurn(context.Background(), input); err != nil { |
| 218 | errs <- fmt.Errorf("%s: %w", input, err) |
| 219 | return |
| 220 | } |
| 221 | } |
| 222 | }() |
| 223 | } |
| 224 | wg.Add(1) |
| 225 | go func() { |
| 226 | defer wg.Done() |
| 227 | <-start |
| 228 | for round := 0; round < 3; round++ { |
| 229 | for _, id := range tabOrder { |
| 230 | if err := app.SetActiveTab(id); err != nil { |
| 231 | errs <- fmt.Errorf("switch to %s: %w", id, err) |
| 232 | return |
| 233 | } |
| 234 | } |
| 235 | } |
| 236 | }() |
| 237 | |
| 238 | close(start) |
| 239 | wg.Wait() |
| 240 | close(errs) |
| 241 | for err := range errs { |
| 242 | t.Error(err) |
| 243 | } |
| 244 | if t.Failed() { |
| 245 | for _, ctrl := range controllers { |
| 246 | ctrl.Close() |
| 247 | } |
| 248 | return |
| 249 | } |
| 250 | |
| 251 | for _, ctrl := range controllers { |
| 252 | if err := ctrl.Snapshot(); err != nil { |
| 253 | t.Fatalf("final snapshot %s: %v", ctrl.Label(), err) |
| 254 | } |
| 255 | ctrl.Close() |
| 256 | } |
| 257 | |
| 258 | for tabIndex, id := range tabOrder { |
| 259 | path := tabs[id].SessionPath |
| 260 | reloaded, err := agent.LoadSession(path) |
| 261 | if err != nil { |
| 262 | t.Fatalf("LoadSession %s: %v", id, err) |
| 263 | } |
| 264 | msgs := reloaded.Snapshot() |
| 265 | wantMessages := 1 + turnsPerTab*2 |
| 266 | if len(msgs) != wantMessages { |
| 267 | t.Fatalf("%s message count = %d, want %d: %+v", id, len(msgs), wantMessages, msgs) |
| 268 | } |
| 269 | if msgs[0].Role != provider.RoleSystem || msgs[0].Content != systemText { |
| 270 | t.Fatalf("%s system message = %+v", id, msgs[0]) |
| 271 | } |
| 272 | for turn := 0; turn < turnsPerTab; turn++ { |
| 273 | user := msgs[1+turn*2] |
| 274 | assistant := msgs[2+turn*2] |
| 275 | wantUser := fmt.Sprintf("tab-%02d-turn-%02d", tabIndex, turn) |
| 276 | if user.Role != provider.RoleUser || user.Content != wantUser { |
| 277 | t.Fatalf("%s turn %d user = %+v, want %q", id, turn, user, wantUser) |
| 278 | } |
| 279 | if assistant.Role != provider.RoleAssistant || assistant.Content != "ok" { |
| 280 | t.Fatalf("%s turn %d assistant = %+v, want ok", id, turn, assistant) |
| 281 | } |
| 282 | } |
| 283 | } |
| 284 | } |
| 285 | |
| 286 | func TestResumeWithFreshSystemPromptPreservesLoadedRewriteBaseline(t *testing.T) { |
| 287 | path := filepath.Join(t.TempDir(), "session.jsonl") |
| 288 | s := agent.NewSession("old sys") |
| 289 | s.Add(provider.Message{Role: provider.RoleUser, Content: "first"}) |
| 290 | s.Add(provider.Message{Role: provider.RoleAssistant, ToolCalls: []provider.ToolCall{{ID: "tool-1", Name: "read_file", Arguments: "{}"}}}) |
| 291 | s.Add(provider.Message{Role: provider.RoleTool, ToolCallID: "tool-1", Name: "read_file", Content: strings.Repeat("detail ", 100)}) |
| 292 | s.Add(provider.Message{Role: provider.RoleAssistant, Content: "done"}) |
| 293 | if err := s.Save(path); err != nil { |
| 294 | t.Fatalf("Save base: %v", err) |
| 295 | } |
| 296 | |
| 297 | loaded, err := agent.LoadSession(path) |
| 298 | if err != nil { |
| 299 | t.Fatalf("LoadSession: %v", err) |
| 300 | } |
| 301 | ctrl := &promptResumeCtrl{history: []provider.Message{{Role: provider.RoleSystem, Content: "new sys"}}} |
| 302 | resumeWithFreshSystemPrompt(ctrl, loaded.Snapshot(), path) |
| 303 | if ctrl.resumed == nil { |
| 304 | t.Fatalf("Resume was not called") |
| 305 | } |
| 306 | |
| 307 | msgs := ctrl.resumed.Snapshot() |
| 308 | msgs[3].Content = "[elided tool result]" |
| 309 | ctrl.resumed.Replace(msgs) |
| 310 | if err := ctrl.resumed.SaveRewrite(path); err != nil { |
| 311 | t.Fatalf("SaveRewrite resumed history: %v", err) |
| 312 | } |
| 313 | |
| 314 | if got := ctrl.path; got != path { |
| 315 | t.Fatalf("resume path = %q, want %q", got, path) |
| 316 | } |
| 317 | reloaded, err := agent.LoadSession(path) |
| 318 | if err != nil { |
| 319 | t.Fatalf("LoadSession rewritten: %v", err) |
| 320 | } |
| 321 | if got := reloaded.Messages[0].Content; got != "new sys" { |
| 322 | t.Fatalf("system prompt after rewrite = %q, want new sys", got) |
| 323 | } |
| 324 | if got := reloaded.Messages[3].Content; got != "[elided tool result]" { |
| 325 | t.Fatalf("tool result after rewrite = %q, want elided", got) |
| 326 | } |
| 327 | if matches, err := filepath.Glob(filepath.Join(filepath.Dir(path), "*-recovery-*.jsonl")); err != nil || len(matches) != 0 { |
| 328 | t.Fatalf("recovery branches after resume rewrite = %v err=%v, want none", matches, err) |
| 329 | } |
| 330 | } |
| 331 | |
| 332 | func TestResumeWithFreshSystemPromptRejectsStaleCarriedHistoryBaseline(t *testing.T) { |
| 333 | path := filepath.Join(t.TempDir(), "session.jsonl") |
| 334 | current := agent.NewSession("old sys") |
| 335 | current.Add(provider.Message{Role: provider.RoleUser, Content: "first"}) |
| 336 | current.Add(provider.Message{Role: provider.RoleAssistant, Content: "one"}) |
| 337 | current.Add(provider.Message{Role: provider.RoleUser, Content: "disk second"}) |
| 338 | current.Add(provider.Message{Role: provider.RoleAssistant, Content: "disk two"}) |
| 339 | if err := current.Save(path); err != nil { |
| 340 | t.Fatalf("Save current: %v", err) |
| 341 | } |
| 342 | |
| 343 | stale := []provider.Message{ |
| 344 | {Role: provider.RoleSystem, Content: "old sys"}, |
| 345 | {Role: provider.RoleUser, Content: "first"}, |
| 346 | {Role: provider.RoleAssistant, Content: "one"}, |
| 347 | } |
| 348 | ctrl := &promptResumeCtrl{history: []provider.Message{{Role: provider.RoleSystem, Content: "new sys"}}} |
| 349 | resumeWithFreshSystemPrompt(ctrl, stale, path) |
| 350 | if ctrl.resumed == nil { |
| 351 | t.Fatalf("Resume was not called") |
| 352 | } |
| 353 | if err := ctrl.resumed.SaveRewrite(path); !errors.Is(err, agent.ErrSessionSnapshotConflict) { |
| 354 | t.Fatalf("SaveRewrite stale carried history err = %v, want ErrSessionSnapshotConflict", err) |
| 355 | } |
| 356 | |
| 357 | reloaded, err := agent.LoadSession(path) |
| 358 | if err != nil { |
| 359 | t.Fatalf("LoadSession current: %v", err) |
| 360 | } |
| 361 | if got := reloaded.Messages[len(reloaded.Messages)-1].Content; got != "disk two" { |
| 362 | t.Fatalf("original tail after stale resume rewrite = %q, want disk two", got) |
| 363 | } |
| 364 | } |
| 365 |