| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "encoding/json" |
| 5 | "errors" |
| 6 | "strings" |
| 7 | "testing" |
| 8 | |
| 9 | "reasonix/internal/event" |
| 10 | "reasonix/internal/provider" |
| 11 | "reasonix/internal/tool" |
| 12 | ) |
| 13 | |
| 14 | type policyWindowProvider struct { |
| 15 | sharedWindowTestProvider |
| 16 | policy provider.ContextBudgetPolicy |
| 17 | } |
| 18 | |
| 19 | func (p *policyWindowProvider) ContextBudgetPolicy() provider.ContextBudgetPolicy { return p.policy } |
| 20 | |
| 21 | func TestAdmitOutputBudgetClipsIssue8909AndScreenshot(t *testing.T) { |
| 22 | cases := []struct { |
| 23 | name string |
| 24 | prompt int |
| 25 | requested int |
| 26 | want int |
| 27 | }{ |
| 28 | {name: "issue8909", prompt: 810_882, requested: 354_469, want: 229_502}, |
| 29 | {name: "screenshot", prompt: 917_189, requested: 245_760, want: 123_195}, |
| 30 | } |
| 31 | for _, tc := range cases { |
| 32 | t.Run(tc.name, func(t *testing.T) { |
| 33 | prov := &sharedWindowTestProvider{budget: tc.requested, shared: true} |
| 34 | a := &Agent{ |
| 35 | agentConfig: agentConfig{contextWindow: 1_048_576}, |
| 36 | svc: agentServices{prov: prov}, |
| 37 | sess: sessionRuntime{output: outputBudgetState{outputBudget: tc.requested}}, |
| 38 | } |
| 39 | msgs := []provider.Message{{Role: provider.RoleUser, Content: strings.Repeat("x", 3_000_000)}} |
| 40 | a.setPromptTokenCalibration(tc.prompt, requestCalibrationShapeOf(provider.Request{Messages: msgs})) |
| 41 | adm, err := a.admitOutputBudget(provider.Request{Messages: msgs, MaxTokens: tc.requested}) |
| 42 | if err != nil { |
| 43 | t.Fatal(err) |
| 44 | } |
| 45 | if !adm.Clipped || adm.EffectiveOutputTokens != tc.want || adm.PhysicalRemaining != tc.want { |
| 46 | t.Fatalf("adm=%+v, want clipped %d", adm, tc.want) |
| 47 | } |
| 48 | }) |
| 49 | } |
| 50 | } |
| 51 | |
| 52 | func TestAdmitOutputBudgetUsesOfficialAutoWhenConfigIsZero(t *testing.T) { |
| 53 | prov := &policyWindowProvider{sharedWindowTestProvider: sharedWindowTestProvider{shared: true}, policy: provider.ContextBudgetPolicy{ |
| 54 | WindowMode: provider.ContextWindowShared, |
| 55 | AutoOutputTokens: provider.DeepSeekMaxOutputTokens, |
| 56 | MaxOutputTokens: provider.DeepSeekMaxOutputTokens, |
| 57 | LimitMode: provider.OutputLimitOmitWhenSafe, |
| 58 | }} |
| 59 | a := &Agent{agentConfig: agentConfig{contextWindow: 1_048_576}, svc: agentServices{prov: prov}} |
| 60 | msgs := []provider.Message{{Role: provider.RoleUser, Content: strings.Repeat("x", 3_000_000)}} |
| 61 | a.setPromptTokenCalibration(810_882, requestCalibrationShapeOf(provider.Request{Messages: msgs})) |
| 62 | req := provider.Request{Messages: msgs, MaxTokens: 0} |
| 63 | if err := a.applyAdmissionToRequest(&req); err != nil { |
| 64 | t.Fatal(err) |
| 65 | } |
| 66 | if req.MaxTokens != 229_502 { |
| 67 | t.Fatalf("auto official clip = %d, want 229502", req.MaxTokens) |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | func TestAdmitOutputBudgetOmitsWhenSafeHasRoom(t *testing.T) { |
| 72 | prov := &policyWindowProvider{policy: provider.ContextBudgetPolicy{ |
| 73 | WindowMode: provider.ContextWindowShared, |
| 74 | AutoOutputTokens: provider.DeepSeekMaxOutputTokens, |
| 75 | MaxOutputTokens: provider.DeepSeekMaxOutputTokens, |
| 76 | LimitMode: provider.OutputLimitOmitWhenSafe, |
| 77 | }} |
| 78 | a := &Agent{agentConfig: agentConfig{contextWindow: 1_048_576}, svc: agentServices{prov: prov}} |
| 79 | req := provider.Request{Messages: []provider.Message{{Role: provider.RoleUser, Content: "hi"}}} |
| 80 | if err := a.applyAdmissionToRequest(&req); err != nil { |
| 81 | t.Fatal(err) |
| 82 | } |
| 83 | if req.MaxTokens != 0 { |
| 84 | t.Fatalf("safe omit injected %d", req.MaxTokens) |
| 85 | } |
| 86 | } |
| 87 | |
| 88 | func TestAdmitOutputBudgetAlwaysSendsOpenCodeLimit(t *testing.T) { |
| 89 | prov := &policyWindowProvider{policy: provider.ContextBudgetPolicy{ |
| 90 | WindowMode: provider.ContextWindowShared, |
| 91 | AutoOutputTokens: 131_072, |
| 92 | MaxOutputTokens: 131_072, |
| 93 | LimitMode: provider.OutputLimitAlways, |
| 94 | }} |
| 95 | a := &Agent{agentConfig: agentConfig{contextWindow: 1_048_576}, svc: agentServices{prov: prov}} |
| 96 | req := provider.Request{Messages: []provider.Message{{Role: provider.RoleUser, Content: "hi"}}, MaxTokens: 0} |
| 97 | if err := a.applyAdmissionToRequest(&req); err != nil { |
| 98 | t.Fatal(err) |
| 99 | } |
| 100 | if req.MaxTokens != 131_072 { |
| 101 | t.Fatalf("OpenCode always send = %d, want 131072", req.MaxTokens) |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | func TestAdmitOutputBudgetNegativeOmitsInsteadOfInject(t *testing.T) { |
| 106 | prov := &policyWindowProvider{policy: provider.ContextBudgetPolicy{ |
| 107 | WindowMode: provider.ContextWindowShared, |
| 108 | AutoOutputTokens: provider.DeepSeekMaxOutputTokens, |
| 109 | LimitMode: provider.OutputLimitOmitWhenSafe, |
| 110 | }} |
| 111 | a := &Agent{agentConfig: agentConfig{contextWindow: 1_048_576}, svc: agentServices{prov: prov}} |
| 112 | msgs := []provider.Message{{Role: provider.RoleUser, Content: strings.Repeat("x", 3_000_000)}} |
| 113 | a.setPromptTokenCalibration(810_882, requestCalibrationShapeOf(provider.Request{Messages: msgs})) |
| 114 | _, err := a.admitOutputBudget(provider.Request{Messages: msgs, MaxTokens: -1}) |
| 115 | if !errors.Is(err, ErrCompactionRequired) { |
| 116 | t.Fatalf("negative omit err = %v, want compaction", err) |
| 117 | } |
| 118 | } |
| 119 | |
| 120 | func TestCompactRatioIndependentOfOutputBudgetAndLearnedWindow(t *testing.T) { |
| 121 | a := &Agent{ |
| 122 | svc: agentServices{prov: &sharedWindowTestProvider{budget: 131_072, shared: true}}, |
| 123 | agentConfig: agentConfig{contextWindow: 128_000, compactRatio: defaultCompactRatio}, |
| 124 | } |
| 125 | want := int(float64(128_000) * defaultCompactRatio) |
| 126 | if got := a.compactTrigger(); got != want { |
| 127 | t.Fatalf("trigger = %d, want %d", got, want) |
| 128 | } |
| 129 | a.sess.output.learned.Store(&learnedContextBudget{windowTokens: 64_000}) |
| 130 | if got := a.compactTrigger(); got != int(float64(64_000)*defaultCompactRatio) { |
| 131 | t.Fatalf("learned trigger = %d", got) |
| 132 | } |
| 133 | a.sess.output.learned.Store(&learnedContextBudget{windowTokens: 64_000}) |
| 134 | if a.compactRatio != defaultCompactRatio { |
| 135 | t.Fatal("compact_ratio must stay user-owned") |
| 136 | } |
| 137 | } |
| 138 | |
| 139 | func TestGuardedSummaryUsesSharedPolicyWhenAutoBudgetIsZero(t *testing.T) { |
| 140 | prov := &policyWindowProvider{policy: provider.ContextBudgetPolicy{WindowMode: provider.ContextWindowShared, LimitMode: provider.OutputLimitOmitWhenSafe}} |
| 141 | a := &Agent{agentConfig: agentConfig{contextWindow: 100_000}, svc: agentServices{prov: prov, sink: event.Discard}} |
| 142 | fold := []provider.Message{{Role: provider.RoleUser, Content: strings.Repeat("字", 80_000)}} |
| 143 | if budget := a.summaryInputBudget(""); budget <= 0 { |
| 144 | t.Fatalf("shared auto-zero summary budget = %d", budget) |
| 145 | } |
| 146 | if got := a.guardedSummaryInputTokens(fold); got <= 0 { |
| 147 | t.Fatalf("guarded tokens = %d", got) |
| 148 | } |
| 149 | } |
| 150 | |
| 151 | func TestLearnedWindowIsolatedPerAgent(t *testing.T) { |
| 152 | first := &Agent{agentConfig: agentConfig{contextWindow: 1_048_576}} |
| 153 | second := &Agent{agentConfig: agentConfig{contextWindow: 1_048_576}} |
| 154 | first.learnContextBudget(200_000, 0, false) |
| 155 | if first.effectiveContextWindow() != 200_000 { |
| 156 | t.Fatalf("first window = %d", first.effectiveContextWindow()) |
| 157 | } |
| 158 | if second.effectiveContextWindow() != 1_048_576 { |
| 159 | t.Fatalf("second agent inherited learned window %d", second.effectiveContextWindow()) |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | func TestZeroConfigWindowUsesLearned(t *testing.T) { |
| 164 | a := &Agent{} |
| 165 | a.learnContextBudget(262_144, 0, false) |
| 166 | if a.effectiveContextWindow() != 262_144 { |
| 167 | t.Fatalf("zero config window = %d", a.effectiveContextWindow()) |
| 168 | } |
| 169 | } |
| 170 | |
| 171 | func TestForkCaptureForwardsContextBudgetPolicy(t *testing.T) { |
| 172 | t.Setenv("REASONIX_EXPERIMENT_FORK_CAPTURE_DIR", t.TempDir()) |
| 173 | inner := &policyWindowProvider{policy: provider.ContextBudgetPolicy{ |
| 174 | WindowMode: provider.ContextWindowShared, AutoOutputTokens: 384_000, LimitMode: provider.OutputLimitOmitWhenSafe, |
| 175 | }} |
| 176 | a := New(inner, tool.NewRegistry(), NewSession(""), Options{}, event.Discard) |
| 177 | got := provider.ResolveContextBudgetPolicy(a.svc.prov) |
| 178 | if got.AutoOutputTokens != 384_000 || got.WindowMode != provider.ContextWindowShared { |
| 179 | t.Fatalf("wrapped policy = %+v", got) |
| 180 | } |
| 181 | } |
| 182 | |
| 183 | func TestFreezeProviderRequestOwnsNestedServerSearchData(t *testing.T) { |
| 184 | req := provider.Request{Messages: []provider.Message{{ |
| 185 | Role: provider.RoleAssistant, |
| 186 | ServerSearch: []provider.ServerSearchCall{{ |
| 187 | ID: "search-1", |
| 188 | Results: []provider.ServerSearchHit{{Title: "original", URL: "https://example.test/original"}}, |
| 189 | Raw: json.RawMessage(`{"value":"original"}`), |
| 190 | }}, |
| 191 | }}} |
| 192 | frozen := freezeProviderRequest(req) |
| 193 | req.Messages[0].ServerSearch[0].Results[0].Title = "mutated" |
| 194 | req.Messages[0].ServerSearch[0].Raw[0] = '[' |
| 195 | |
| 196 | got := frozen.Messages[0].ServerSearch[0] |
| 197 | if got.Results[0].Title != "original" || string(got.Raw) != `{"value":"original"}` { |
| 198 | t.Fatalf("frozen server search shared mutable data: %+v", got) |
| 199 | } |
| 200 | } |
| 201 |