| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "fmt" |
| 6 | "strings" |
| 7 | "testing" |
| 8 | |
| 9 | "reasonix/internal/event" |
| 10 | "reasonix/internal/provider" |
| 11 | "reasonix/internal/tool" |
| 12 | ) |
| 13 | |
| 14 | // Issue #7935 used to install prune projections every time many tool results |
| 15 | // crossed a snip threshold. Automatic maintenance is now summary-only and |
| 16 | // generation-scoped: below compact_ratio nothing happens; at the threshold |
| 17 | // exactly one summary lands; appends before the next threshold do not re-pay. |
| 18 | func TestIssue7935Maintains206ToolResultsOnceAndOnlyProcessesNewTail(t *testing.T) { |
| 19 | const staleResults = 40 |
| 20 | big := strings.Repeat("line\n", 200) |
| 21 | messages := []provider.Message{ |
| 22 | {Role: provider.RoleSystem, Content: "system"}, |
| 23 | {Role: provider.RoleUser, Content: "first task"}, |
| 24 | } |
| 25 | for i := range staleResults { |
| 26 | id := fmt.Sprintf("old-%d", i) |
| 27 | messages = append(messages, |
| 28 | provider.Message{Role: provider.RoleAssistant, ToolCalls: []provider.ToolCall{{ID: id, Name: "read_file", Arguments: "{}"}}}, |
| 29 | provider.Message{Role: provider.RoleTool, ToolCallID: id, Name: "read_file", Content: big}, |
| 30 | ) |
| 31 | } |
| 32 | messages = append(messages, |
| 33 | provider.Message{Role: provider.RoleUser, Content: "recent question"}, |
| 34 | provider.Message{Role: provider.RoleAssistant, Content: "recent answer"}, |
| 35 | ) |
| 36 | sess := &Session{Messages: messages} |
| 37 | sink := &recordSink{} |
| 38 | a := New(&fakeProvider{reply: "structured history digest"}, tool.NewRegistry(), sess, Options{ |
| 39 | ContextWindow: 8_000, |
| 40 | CompactRatio: 0.80, |
| 41 | RecentKeep: 2, |
| 42 | ArchiveDir: t.TempDir(), |
| 43 | }, sink) |
| 44 | |
| 45 | // Below the fold trigger nothing may be rewritten. |
| 46 | prepareForObservedUsage(a, context.Background(), &provider.Usage{PromptTokens: 5_000}) |
| 47 | if got := a.currentProjectionVersion(); got != 0 { |
| 48 | t.Fatalf("history was rewritten below the fold trigger: projection version %d", got) |
| 49 | } |
| 50 | |
| 51 | // Crossing the sole trigger installs one summary checkpoint, never prune. |
| 52 | prepareForObservedUsage(a, context.Background(), &provider.Usage{PromptTokens: 7_000}) |
| 53 | firstVersion := a.currentProjectionVersion() |
| 54 | if firstVersion != 1 { |
| 55 | t.Fatalf("first maintenance version = %d, want 1", firstVersion) |
| 56 | } |
| 57 | if got := countToolResultsWithPrefix(a.modelVisibleMessages(), prunedMarker); got != 0 { |
| 58 | t.Fatalf("automatic prune markers installed = %d, want 0", got) |
| 59 | } |
| 60 | status := a.ContextMaintenanceSnapshot() |
| 61 | if status.LastReceipt == nil || status.LastReceipt.Action != "summary" || status.LastReceipt.Status != "applied" { |
| 62 | t.Fatalf("maintenance snapshot receipt = %+v", status.LastReceipt) |
| 63 | } |
| 64 | for i, msg := range sess.Snapshot() { |
| 65 | if msg.Role == provider.RoleTool && msg.Content != big { |
| 66 | t.Fatalf("canonical tool result %d was rewritten", i) |
| 67 | } |
| 68 | } |
| 69 | |
| 70 | // Appends that stay under the next threshold must not re-summarize. |
| 71 | sess.Add(provider.Message{Role: provider.RoleAssistant, ToolCalls: []provider.ToolCall{{ID: "new", Name: "read_file", Arguments: "{}"}}}) |
| 72 | sess.Add(provider.Message{Role: provider.RoleTool, ToolCallID: "new", Name: "read_file", Content: "small"}) |
| 73 | sess.Add(provider.Message{Role: provider.RoleUser, Content: "continue"}) |
| 74 | sess.Add(provider.Message{Role: provider.RoleAssistant, Content: "ok"}) |
| 75 | prepareForObservedUsage(a, context.Background(), &provider.Usage{PromptTokens: 1_500}) |
| 76 | if got := a.currentProjectionVersion(); got != firstVersion { |
| 77 | t.Fatalf("below-threshold append advanced version to %d, want %d", got, firstVersion) |
| 78 | } |
| 79 | |
| 80 | var applied int |
| 81 | for _, got := range sink.kinds(event.ContextMaintenanceEvent) { |
| 82 | if got.Maintenance != nil && got.Maintenance.Status == "applied" && got.Maintenance.Action == "summary" { |
| 83 | applied++ |
| 84 | } |
| 85 | } |
| 86 | if applied != 1 { |
| 87 | t.Fatalf("summary applied events = %d, want exactly 1", applied) |
| 88 | } |
| 89 | } |
| 90 | |
| 91 | func countToolResultsWithPrefix(messages []provider.Message, prefix string) int { |
| 92 | var count int |
| 93 | for _, msg := range messages { |
| 94 | if msg.Role == provider.RoleTool && strings.HasPrefix(msg.Content, prefix) { |
| 95 | count++ |
| 96 | } |
| 97 | } |
| 98 | return count |
| 99 | } |
| 100 |