| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "fmt" |
| 6 | "path/filepath" |
| 7 | "strings" |
| 8 | "testing" |
| 9 | |
| 10 | "reasonix/internal/event" |
| 11 | "reasonix/internal/provider" |
| 12 | "reasonix/internal/tool" |
| 13 | ) |
| 14 | |
| 15 | const restartMaintenanceWindow = 1_000_000 |
| 16 | |
| 17 | // largeRestartToolHistory builds a realistic long coding session from capped |
| 18 | // tool results. Reusing the same immutable string keeps the test allocation |
| 19 | // small while every message still contributes to the provider request size. |
| 20 | func largeRestartToolHistory(results int) []provider.Message { |
| 21 | result := strings.Repeat("x", 32*1024) |
| 22 | messages := []provider.Message{ |
| 23 | {Role: provider.RoleSystem, Content: "system"}, |
| 24 | {Role: provider.RoleUser, Content: "long-running task"}, |
| 25 | } |
| 26 | for i := range results { |
| 27 | id := fmt.Sprintf("tool-%d", i) |
| 28 | messages = append(messages, |
| 29 | provider.Message{Role: provider.RoleAssistant, ToolCalls: []provider.ToolCall{{ID: id, Name: "read_file", Arguments: "{}"}}}, |
| 30 | provider.Message{Role: provider.RoleTool, ToolCallID: id, Name: "read_file", Content: result}, |
| 31 | ) |
| 32 | } |
| 33 | return append(messages, |
| 34 | provider.Message{Role: provider.RoleUser, Content: "write the handover now"}, |
| 35 | provider.Message{Role: provider.RoleAssistant, Content: "working"}, |
| 36 | ) |
| 37 | } |
| 38 | |
| 39 | func newRestartMaintenanceAgent(path string, messages []provider.Message, sink event.Sink) *Agent { |
| 40 | // Explicit 0.80 matches the v1.22.0 user scenario and the plan's regression |
| 41 | // fixture. New configs default to 0.80; explicit legacy ratios remain unchanged. |
| 42 | return New(nil, tool.NewRegistry(), &Session{Messages: append([]provider.Message(nil), messages...)}, Options{ |
| 43 | ContextWindow: restartMaintenanceWindow, |
| 44 | CompactRatio: 0.80, |
| 45 | RecentKeep: 2, |
| 46 | SessionPath: path, |
| 47 | WorkspaceID: "workspace", |
| 48 | ModelRef: "provider/model", |
| 49 | }, sink) |
| 50 | } |
| 51 | |
| 52 | func appliedMaintenanceEvents(sink *recordSink) int { |
| 53 | count := 0 |
| 54 | for _, got := range sink.kinds(event.ContextMaintenanceEvent) { |
| 55 | if got.Maintenance != nil && got.Maintenance.Status == "applied" { |
| 56 | count++ |
| 57 | } |
| 58 | } |
| 59 | return count |
| 60 | } |
| 61 | |
| 62 | // 80 capped tool results put the request just above 60% of a one-million-token |
| 63 | // window. In v1.22.0 that crossed tool_result_snip_ratio, rewrote the prompt, |
| 64 | // and did so again after reopening. compact_ratio=80% must be the sole trigger. |
| 65 | func TestContextMaintenanceWaitsForCompactRatioAcrossRestart(t *testing.T) { |
| 66 | messages := largeRestartToolHistory(80) |
| 67 | path := filepath.Join(t.TempDir(), "session.jsonl") |
| 68 | |
| 69 | firstSink := &recordSink{} |
| 70 | first := newRestartMaintenanceAgent(path, messages, firstSink) |
| 71 | before := first.ContextMaintenanceSnapshot() |
| 72 | if before.ProjectedTokens < 600_000 || before.ProjectedTokens >= before.FoldTrigger { |
| 73 | t.Fatalf("fixture prompt/trigger = %d/%d, want prompt in [600000, fold)", before.ProjectedTokens, before.FoldTrigger) |
| 74 | } |
| 75 | if _, err := first.contextManager().Prepare(context.Background(), ContextPreparePolicy{Trigger: CompactionTriggerPressure}); err != nil { |
| 76 | t.Fatal(err) |
| 77 | } |
| 78 | if got := first.currentProjectionVersion(); got != 0 { |
| 79 | t.Fatalf("60%% prompt installed projection version %d below compact_ratio", got) |
| 80 | } |
| 81 | if got := appliedMaintenanceEvents(firstSink); got != 0 { |
| 82 | t.Fatalf("60%% prompt emitted %d applied maintenance events", got) |
| 83 | } |
| 84 | |
| 85 | reopenedSink := &recordSink{} |
| 86 | reopened := newRestartMaintenanceAgent(path, messages, reopenedSink) |
| 87 | if _, err := reopened.contextManager().Prepare(context.Background(), ContextPreparePolicy{Trigger: CompactionTriggerPressure}); err != nil { |
| 88 | t.Fatal(err) |
| 89 | } |
| 90 | if got := reopened.currentProjectionVersion(); got != 0 { |
| 91 | t.Fatalf("reopened 60%% prompt installed projection version %d", got) |
| 92 | } |
| 93 | if got := appliedMaintenanceEvents(reopenedSink); got != 0 { |
| 94 | t.Fatalf("reopened 60%% prompt emitted %d applied maintenance events", got) |
| 95 | } |
| 96 | } |
| 97 | |
| 98 | // Once the prompt really crosses 80%, one summary checkpoint is installed. |
| 99 | // Reopening must restore that sidecar without re-summarizing or replaying events. |
| 100 | func TestContextMaintenanceRestoresAppliedProjectionWithoutReapplying(t *testing.T) { |
| 101 | messages := largeRestartToolHistory(110) |
| 102 | path := filepath.Join(t.TempDir(), "session.jsonl") |
| 103 | |
| 104 | // A deterministic fake provider so the single summary transaction can land. |
| 105 | firstSink := &recordSink{} |
| 106 | first := New(&fakeProvider{reply: "structured resume briefing"}, tool.NewRegistry(), |
| 107 | &Session{Messages: append([]provider.Message(nil), messages...)}, Options{ |
| 108 | ContextWindow: restartMaintenanceWindow, |
| 109 | CompactRatio: 0.80, |
| 110 | RecentKeep: 2, |
| 111 | SessionPath: path, |
| 112 | WorkspaceID: "workspace", |
| 113 | ModelRef: "provider/model", |
| 114 | }, firstSink) |
| 115 | before := first.ContextMaintenanceSnapshot() |
| 116 | if before.ProjectedTokens < before.FoldTrigger { |
| 117 | t.Fatalf("fixture prompt/trigger = %d/%d, want prompt at or above fold", before.ProjectedTokens, before.FoldTrigger) |
| 118 | } |
| 119 | if _, err := first.contextManager().Prepare(context.Background(), ContextPreparePolicy{Trigger: CompactionTriggerPressure}); err != nil { |
| 120 | t.Fatal(err) |
| 121 | } |
| 122 | if got := first.currentProjectionVersion(); got != 1 { |
| 123 | t.Fatalf("first maintenance projection version = %d, want 1", got) |
| 124 | } |
| 125 | if got := appliedMaintenanceEvents(firstSink); got != 1 { |
| 126 | t.Fatalf("first maintenance applied events = %d, want 1", got) |
| 127 | } |
| 128 | // Content-driven checkpoint should materially shrink the visible request. |
| 129 | after := first.ContextMaintenanceSnapshot() |
| 130 | if after.ProjectedTokens > after.HardInputCeiling/2 && after.ProjectedTokens > 300_000 { |
| 131 | t.Fatalf("checkpoint projected tokens = %d, want content-driven low occupancy", after.ProjectedTokens) |
| 132 | } |
| 133 | |
| 134 | reopenedSink := &recordSink{} |
| 135 | reopened := New(nil, tool.NewRegistry(), |
| 136 | &Session{Messages: append([]provider.Message(nil), messages...)}, Options{ |
| 137 | ContextWindow: restartMaintenanceWindow, |
| 138 | CompactRatio: 0.80, |
| 139 | RecentKeep: 2, |
| 140 | SessionPath: path, |
| 141 | WorkspaceID: "workspace", |
| 142 | ModelRef: "provider/model", |
| 143 | }, reopenedSink) |
| 144 | if got := reopened.currentProjectionVersion(); got != 1 { |
| 145 | t.Fatalf("restored projection version = %d, want 1", got) |
| 146 | } |
| 147 | if got := reopened.ContextMaintenanceSnapshot().CheckpointState; got != "restored" { |
| 148 | t.Fatalf("checkpoint state = %q, want restored", got) |
| 149 | } |
| 150 | if _, err := reopened.contextManager().Prepare(context.Background(), ContextPreparePolicy{Trigger: CompactionTriggerPressure}); err != nil { |
| 151 | t.Fatal(err) |
| 152 | } |
| 153 | if got := reopened.currentProjectionVersion(); got != 1 { |
| 154 | t.Fatalf("reopen advanced unchanged projection to version %d", got) |
| 155 | } |
| 156 | if got := appliedMaintenanceEvents(reopenedSink); got != 0 { |
| 157 | t.Fatalf("reopen reapplied maintenance %d times", got) |
| 158 | } |
| 159 | } |
| 160 |