@devflow-tools/benchmark 0.12.6 → 0.13.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/accuracy-queries.d.ts +11 -0
- package/dist/accuracy-queries.d.ts.map +1 -0
- package/dist/accuracy-queries.js +63 -0
- package/dist/accuracy-queries.js.map +1 -0
- package/dist/benchmark-engine.d.ts +26 -0
- package/dist/benchmark-engine.d.ts.map +1 -0
- package/dist/benchmark-engine.js +184 -0
- package/dist/benchmark-engine.js.map +1 -0
- package/dist/debug-buildcontext.d.ts +2 -0
- package/dist/debug-buildcontext.d.ts.map +1 -0
- package/dist/debug-buildcontext.js +28 -0
- package/dist/debug-buildcontext.js.map +1 -0
- package/dist/index.d.ts +3 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +2 -0
- package/dist/index.js.map +1 -0
- package/dist/run-accuracy-tests.d.ts +2 -0
- package/dist/run-accuracy-tests.d.ts.map +1 -0
- package/dist/run-accuracy-tests.js +75 -0
- package/dist/run-accuracy-tests.js.map +1 -0
- package/package.json +4 -4
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
export interface AccuracyTestCase {
|
|
2
|
+
query: string;
|
|
3
|
+
engine: 'codegraph' | 'knowledge' | 'memory';
|
|
4
|
+
expectFiles?: string[];
|
|
5
|
+
expectSymbols?: string[];
|
|
6
|
+
expectSource?: string;
|
|
7
|
+
minResults: number;
|
|
8
|
+
skipReason?: string;
|
|
9
|
+
}
|
|
10
|
+
export declare const ACCURACY_TESTS: AccuracyTestCase[];
|
|
11
|
+
//# sourceMappingURL=accuracy-queries.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"accuracy-queries.d.ts","sourceRoot":"","sources":["../src/accuracy-queries.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,WAAW,GAAG,WAAW,GAAG,QAAQ,CAAC;IAC7C,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,eAAO,MAAM,cAAc,EAAE,gBAAgB,EA6D5C,CAAC"}
|
|
@@ -0,0 +1,63 @@
|
|
|
1
|
+
export const ACCURACY_TESTS = [
|
|
2
|
+
{
|
|
3
|
+
query: 'api key auth guard',
|
|
4
|
+
engine: 'codegraph',
|
|
5
|
+
expectSymbols: ['ApiKeyGuard', 'AuthModule'],
|
|
6
|
+
expectFiles: ['api-key.guard.ts'],
|
|
7
|
+
minResults: 3,
|
|
8
|
+
},
|
|
9
|
+
{
|
|
10
|
+
query: 'post-tool-use hook',
|
|
11
|
+
engine: 'codegraph',
|
|
12
|
+
expectFiles: ['post-tool-use.ts'],
|
|
13
|
+
expectSymbols: ['handlePostToolUse'],
|
|
14
|
+
minResults: 1,
|
|
15
|
+
},
|
|
16
|
+
{
|
|
17
|
+
query: 'How to use React useEffect?',
|
|
18
|
+
engine: 'knowledge',
|
|
19
|
+
expectSource: 'react',
|
|
20
|
+
minResults: 5,
|
|
21
|
+
},
|
|
22
|
+
{
|
|
23
|
+
query: 'Vue 3 composition API setup',
|
|
24
|
+
engine: 'knowledge',
|
|
25
|
+
expectSource: 'vue',
|
|
26
|
+
minResults: 3,
|
|
27
|
+
},
|
|
28
|
+
{
|
|
29
|
+
query: 'NestJS dependency injection providers',
|
|
30
|
+
engine: 'knowledge',
|
|
31
|
+
expectSource: 'nest',
|
|
32
|
+
minResults: 3,
|
|
33
|
+
},
|
|
34
|
+
{
|
|
35
|
+
query: 'dependency graph export subgraph',
|
|
36
|
+
engine: 'codegraph',
|
|
37
|
+
minResults: 3,
|
|
38
|
+
},
|
|
39
|
+
{
|
|
40
|
+
query: 'memory extraction pipeline',
|
|
41
|
+
engine: 'codegraph',
|
|
42
|
+
expectFiles: ['memory-engine.ts'],
|
|
43
|
+
minResults: 2,
|
|
44
|
+
},
|
|
45
|
+
{
|
|
46
|
+
query: 'benchmark test and performance measurement',
|
|
47
|
+
engine: 'codegraph',
|
|
48
|
+
expectFiles: ['benchmark-view.tsx'],
|
|
49
|
+
expectSymbols: ['BenchmarkTask'],
|
|
50
|
+
minResults: 3,
|
|
51
|
+
},
|
|
52
|
+
{
|
|
53
|
+
query: '项目整体架构 MCP 工具',
|
|
54
|
+
engine: 'codegraph',
|
|
55
|
+
minResults: 3,
|
|
56
|
+
},
|
|
57
|
+
{
|
|
58
|
+
engine: 'memory',
|
|
59
|
+
query: '',
|
|
60
|
+
minResults: 5,
|
|
61
|
+
},
|
|
62
|
+
];
|
|
63
|
+
//# sourceMappingURL=accuracy-queries.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"accuracy-queries.js","sourceRoot":"","sources":["../src/accuracy-queries.ts"],"names":[],"mappings":"AAUA,MAAM,CAAC,MAAM,cAAc,GAAuB;IAChD;QACE,KAAK,EAAE,oBAAoB;QAC3B,MAAM,EAAE,WAAW;QACnB,aAAa,EAAE,CAAC,aAAa,EAAE,YAAY,CAAC;QAC5C,WAAW,EAAE,CAAC,kBAAkB,CAAC;QACjC,UAAU,EAAE,CAAC;KACd;IACD;QACE,KAAK,EAAE,oBAAoB;QAC3B,MAAM,EAAE,WAAW;QACnB,WAAW,EAAE,CAAC,kBAAkB,CAAC;QACjC,aAAa,EAAE,CAAC,mBAAmB,CAAC;QACpC,UAAU,EAAE,CAAC;KACd;IACD;QACE,KAAK,EAAE,6BAA6B;QACpC,MAAM,EAAE,WAAW;QACnB,YAAY,EAAE,OAAO;QACrB,UAAU,EAAE,CAAC;KACd;IACD;QACE,KAAK,EAAE,6BAA6B;QACpC,MAAM,EAAE,WAAW;QACnB,YAAY,EAAE,KAAK;QACnB,UAAU,EAAE,CAAC;KACd;IACD;QACE,KAAK,EAAE,uCAAuC;QAC9C,MAAM,EAAE,WAAW;QACnB,YAAY,EAAE,MAAM;QACpB,UAAU,EAAE,CAAC;KACd;IACD;QACE,KAAK,EAAE,kCAAkC;QACzC,MAAM,EAAE,WAAW;QACnB,UAAU,EAAE,CAAC;KACd;IACD;QACE,KAAK,EAAE,4BAA4B;QACnC,MAAM,EAAE,WAAW;QACnB,WAAW,EAAE,CAAC,kBAAkB,CAAC;QACjC,UAAU,EAAE,CAAC;KACd;IACD;QACE,KAAK,EAAE,4CAA4C;QACnD,MAAM,EAAE,WAAW;QACnB,WAAW,EAAE,CAAC,oBAAoB,CAAC;QACnC,aAAa,EAAE,CAAC,eAAe,CAAC;QAChC,UAAU,EAAE,CAAC;KACd;IACD;QACE,KAAK,EAAE,eAAe;QACtB,MAAM,EAAE,WAAW;QACnB,UAAU,EAAE,CAAC;KACd;IACD;QACE,MAAM,EAAE,QAAQ;QAChB,KAAK,EAAE,EAAE;QACT,UAAU,EAAE,CAAC;KACd;CACF,CAAC"}
|
|
@@ -0,0 +1,26 @@
|
|
|
1
|
+
import type { BenchmarkTask, TaskResult, BenchmarkReport, ReportMeta, BenchmarkScenario, BenchmarkRunResult, ComparisonResult, BenchmarkComparisonResult } from "@devflow-tools/sdk";
|
|
2
|
+
export interface BenchmarkOptions {
|
|
3
|
+
taskIds?: string[];
|
|
4
|
+
rounds?: number;
|
|
5
|
+
}
|
|
6
|
+
export type TaskRunner = (task: BenchmarkTask) => Promise<TaskResult>;
|
|
7
|
+
export declare class BenchmarkEngine {
|
|
8
|
+
private reports;
|
|
9
|
+
private tasks;
|
|
10
|
+
private scenarios;
|
|
11
|
+
addTask(task: BenchmarkTask): Promise<void>;
|
|
12
|
+
runOne(taskId: string, runner: TaskRunner): Promise<TaskResult>;
|
|
13
|
+
runAll(runner: TaskRunner, options?: BenchmarkOptions): Promise<BenchmarkReport>;
|
|
14
|
+
private computeSummary;
|
|
15
|
+
getReport(runId: string): Promise<BenchmarkReport | null>;
|
|
16
|
+
listReports(): Promise<ReportMeta[]>;
|
|
17
|
+
getTasks(): BenchmarkTask[];
|
|
18
|
+
runScenario(scenario: BenchmarkScenario, mode: "baseline" | "devflow"): Promise<BenchmarkRunResult>;
|
|
19
|
+
private runBaseline;
|
|
20
|
+
private runDevFlow;
|
|
21
|
+
compareResults(baseline: BenchmarkRunResult, devflow: BenchmarkRunResult): ComparisonResult;
|
|
22
|
+
registerScenario(scenario: BenchmarkScenario): void;
|
|
23
|
+
runScenarioPair(scenarioId: string, baselineRunner: (prompt: string) => Promise<Omit<BenchmarkRunResult, "scenarioId" | "mode" | "timestamp">>, devflowRunner: (prompt: string) => Promise<Omit<BenchmarkRunResult, "scenarioId" | "mode" | "timestamp">>): Promise<BenchmarkComparisonResult>;
|
|
24
|
+
runScenarioBatch(scenarioIds: string[], baselineRunner: (prompt: string) => Promise<Omit<BenchmarkRunResult, "scenarioId" | "mode" | "timestamp">>, devflowRunner: (prompt: string) => Promise<Omit<BenchmarkRunResult, "scenarioId" | "mode" | "timestamp">>): Promise<BenchmarkComparisonResult[]>;
|
|
25
|
+
}
|
|
26
|
+
//# sourceMappingURL=benchmark-engine.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark-engine.d.ts","sourceRoot":"","sources":["../src/benchmark-engine.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,aAAa,EACb,UAAU,EACV,eAAe,EACf,UAAU,EACV,iBAAiB,EACjB,kBAAkB,EAClB,gBAAgB,EAChB,yBAAyB,EAE1B,MAAM,oBAAoB,CAAC;AAY5B,MAAM,WAAW,gBAAgB;IAC/B,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,MAAM,UAAU,GAAG,CAAC,IAAI,EAAE,aAAa,KAAK,OAAO,CAAC,UAAU,CAAC,CAAC;AAEtE,qBAAa,eAAe;IAC1B,OAAO,CAAC,OAAO,CAAsC;IACrD,OAAO,CAAC,KAAK,CAAuC;IACpD,OAAO,CAAC,SAAS,CAAwC;IAEnD,OAAO,CAAC,IAAI,EAAE,aAAa,GAAG,OAAO,CAAC,IAAI,CAAC;IAS3C,MAAM,CACV,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,UAAU,GACjB,OAAO,CAAC,UAAU,CAAC;IAMhB,MAAM,CACV,MAAM,EAAE,UAAU,EAClB,OAAO,CAAC,EAAE,gBAAgB,GACzB,OAAO,CAAC,eAAe,CAAC;IAwB3B,OAAO,CAAC,cAAc;IAuBhB,SAAS,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,eAAe,GAAG,IAAI,CAAC;IAIzD,WAAW,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;IAQ1C,QAAQ,IAAI,aAAa,EAAE;IAIrB,WAAW,CACf,QAAQ,EAAE,iBAAiB,EAC3B,IAAI,EAAE,UAAU,GAAG,SAAS,GAC3B,OAAO,CAAC,kBAAkB,CAAC;YAyBhB,WAAW;YAgBX,UAAU;IAgBxB,cAAc,CACZ,QAAQ,EAAE,kBAAkB,EAC5B,OAAO,EAAE,kBAAkB,GAC1B,gBAAgB;IAmBnB,gBAAgB,CAAC,QAAQ,EAAE,iBAAiB,GAAG,IAAI;IAI7C,eAAe,CACnB,UAAU,EAAE,MAAM,EAClB,cAAc,EAAE,CACd,MAAM,EAAE,MAAM,KACX,OAAO,CAAC,IAAI,CAAC,kBAAkB,EAAE,YAAY,GAAG,MAAM,GAAG,WAAW,CAAC,CAAC,EAC3E,aAAa,EAAE,CACb,MAAM,EAAE,MAAM,KACX,OAAO,CAAC,IAAI,CAAC,kBAAkB,EAAE,YAAY,GAAG,MAAM,GAAG,WAAW,CAAC,CAAC,GAC1E,OAAO,CAAC,yBAAyB,CAAC;IAiD/B,gBAAgB,CACpB,WAAW,EAAE,MAAM,EAAE,EACrB,cAAc,EAAE,CACd,MAAM,EAAE,MAAM,KACX,OAAO,CAAC,IAAI,CAAC,kBAAkB,EAAE,YAAY,GAAG,MAAM,GAAG,WAAW,CAAC,CAAC,EAC3E,aAAa,EAAE,CACb,MAAM,EAAE,MAAM,KACX,OAAO,CAAC,IAAI,CAAC,kBAAkB,EAAE,YAAY,GAAG,MAAM,GAAG,WAAW,CAAC,CAAC,GAC1E,OAAO,CAAC,yBAAyB,EAAE,CAAC;CAQxC"}
|
|
@@ -0,0 +1,184 @@
|
|
|
1
|
+
const BUILTIN_TASKS = [
|
|
2
|
+
{ id: "bug-1", type: "bugfix", desc: "修复状态未更新导致的白屏", expectedFiles: 3 },
|
|
3
|
+
{ id: "bug-2", type: "bugfix", desc: "修复 API 错误码未处理导致崩溃", expectedFiles: 2 },
|
|
4
|
+
{ id: "bug-3", type: "bugfix", desc: "修复条件渲染逻辑反转", expectedFiles: 1 },
|
|
5
|
+
{ id: "feat-1", type: "feature", desc: "添加用户搜索功能", expectedEndpoints: 1 },
|
|
6
|
+
{ id: "feat-2", type: "feature", desc: "添加分页组件", expectedComponents: 2 },
|
|
7
|
+
{ id: "review-1", type: "review", desc: "Review 包含 N+1 查询的 PR", expectedIssues: 3 },
|
|
8
|
+
{ id: "review-2", type: "review", desc: "Review 包含 XSS 风险的 PR", expectedIssues: 2 },
|
|
9
|
+
];
|
|
10
|
+
export class BenchmarkEngine {
|
|
11
|
+
reports = new Map();
|
|
12
|
+
tasks = [...BUILTIN_TASKS];
|
|
13
|
+
scenarios = new Map();
|
|
14
|
+
async addTask(task) {
|
|
15
|
+
const existing = this.tasks.findIndex((t) => t.id === task.id);
|
|
16
|
+
if (existing >= 0) {
|
|
17
|
+
this.tasks[existing] = task;
|
|
18
|
+
}
|
|
19
|
+
else {
|
|
20
|
+
this.tasks.push(task);
|
|
21
|
+
}
|
|
22
|
+
}
|
|
23
|
+
async runOne(taskId, runner) {
|
|
24
|
+
const task = this.tasks.find((t) => t.id === taskId);
|
|
25
|
+
if (!task)
|
|
26
|
+
throw new Error(`Task "${taskId}" not found`);
|
|
27
|
+
return runner(task);
|
|
28
|
+
}
|
|
29
|
+
async runAll(runner, options) {
|
|
30
|
+
const targets = options?.taskIds
|
|
31
|
+
? this.tasks.filter((t) => options.taskIds.includes(t.id))
|
|
32
|
+
: this.tasks;
|
|
33
|
+
const runId = `bench:${Date.now()}:${Math.random().toString(36).slice(2, 7)}`;
|
|
34
|
+
const taskResults = [];
|
|
35
|
+
for (const task of targets) {
|
|
36
|
+
const result = await runner(task);
|
|
37
|
+
taskResults.push(result);
|
|
38
|
+
}
|
|
39
|
+
const report = {
|
|
40
|
+
runId,
|
|
41
|
+
timestamp: Date.now(),
|
|
42
|
+
summary: this.computeSummary(taskResults),
|
|
43
|
+
tasks: taskResults,
|
|
44
|
+
};
|
|
45
|
+
this.reports.set(runId, report);
|
|
46
|
+
return report;
|
|
47
|
+
}
|
|
48
|
+
computeSummary(results) {
|
|
49
|
+
if (results.length === 0) {
|
|
50
|
+
return {
|
|
51
|
+
avgTokenReduction: "0%",
|
|
52
|
+
avgTimeSaved: "0%",
|
|
53
|
+
avgAccuracyGain: "0%",
|
|
54
|
+
avgConfidenceGain: "0%",
|
|
55
|
+
};
|
|
56
|
+
}
|
|
57
|
+
const totalReduction = results.reduce((s, r) => s + r.delta.tokenReduction, 0);
|
|
58
|
+
const totalTimeSaved = results.reduce((s, r) => s + r.delta.timeSaved, 0);
|
|
59
|
+
const totalAccuracy = results.reduce((s, r) => s + r.delta.accuracy, 0);
|
|
60
|
+
const n = results.length;
|
|
61
|
+
return {
|
|
62
|
+
avgTokenReduction: `${(totalReduction / n * 100).toFixed(0)}%`,
|
|
63
|
+
avgTimeSaved: `↓${(totalTimeSaved / n * 100).toFixed(0)}%`,
|
|
64
|
+
avgAccuracyGain: `+${(totalAccuracy / n * 100).toFixed(0)}%`,
|
|
65
|
+
avgConfidenceGain: `+${(Math.min(totalAccuracy / n + 0.05, 0.99) * 100).toFixed(0)}%`,
|
|
66
|
+
};
|
|
67
|
+
}
|
|
68
|
+
async getReport(runId) {
|
|
69
|
+
return this.reports.get(runId) ?? null;
|
|
70
|
+
}
|
|
71
|
+
async listReports() {
|
|
72
|
+
return Array.from(this.reports.values()).map((r) => ({
|
|
73
|
+
runId: r.runId,
|
|
74
|
+
timestamp: r.timestamp,
|
|
75
|
+
taskCount: r.tasks.length,
|
|
76
|
+
}));
|
|
77
|
+
}
|
|
78
|
+
getTasks() {
|
|
79
|
+
return [...this.tasks];
|
|
80
|
+
}
|
|
81
|
+
async runScenario(scenario, mode) {
|
|
82
|
+
const startTime = Date.now();
|
|
83
|
+
const prepStart = startTime;
|
|
84
|
+
const result = mode === "baseline"
|
|
85
|
+
? await this.runBaseline(scenario)
|
|
86
|
+
: await this.runDevFlow(scenario);
|
|
87
|
+
const prepTimeMs = Date.now() - prepStart;
|
|
88
|
+
const durationMs = Date.now() - startTime;
|
|
89
|
+
return {
|
|
90
|
+
scenarioId: scenario.id,
|
|
91
|
+
mode,
|
|
92
|
+
prepTimeMs,
|
|
93
|
+
tokenCount: result.tokenCount,
|
|
94
|
+
firstPassSuccess: result.firstPassSuccess,
|
|
95
|
+
reworkCount: result.reworkCount,
|
|
96
|
+
durationMs,
|
|
97
|
+
notes: result.notes,
|
|
98
|
+
timestamp: startTime,
|
|
99
|
+
};
|
|
100
|
+
}
|
|
101
|
+
async runBaseline(_scenario) {
|
|
102
|
+
return {
|
|
103
|
+
tokenCount: 0,
|
|
104
|
+
firstPassSuccess: false,
|
|
105
|
+
reworkCount: 0,
|
|
106
|
+
notes: "No runner provided — benchmark was not executed",
|
|
107
|
+
};
|
|
108
|
+
}
|
|
109
|
+
async runDevFlow(_scenario) {
|
|
110
|
+
return {
|
|
111
|
+
tokenCount: 0,
|
|
112
|
+
firstPassSuccess: false,
|
|
113
|
+
reworkCount: 0,
|
|
114
|
+
notes: "No runner provided — benchmark was not executed",
|
|
115
|
+
};
|
|
116
|
+
}
|
|
117
|
+
compareResults(baseline, devflow) {
|
|
118
|
+
return {
|
|
119
|
+
timeSavedRate: baseline.durationMs > 0
|
|
120
|
+
? (baseline.durationMs - devflow.durationMs) / baseline.durationMs
|
|
121
|
+
: 0,
|
|
122
|
+
tokenReductionRate: baseline.tokenCount > 0
|
|
123
|
+
? (baseline.tokenCount - devflow.tokenCount) / baseline.tokenCount
|
|
124
|
+
: 0,
|
|
125
|
+
firstPassLift: devflow.firstPassSuccess ? 1 : 0,
|
|
126
|
+
reworkReductionRate: baseline.reworkCount > 0
|
|
127
|
+
? (baseline.reworkCount - devflow.reworkCount) /
|
|
128
|
+
baseline.reworkCount
|
|
129
|
+
: 0,
|
|
130
|
+
};
|
|
131
|
+
}
|
|
132
|
+
registerScenario(scenario) {
|
|
133
|
+
this.scenarios.set(scenario.id, scenario);
|
|
134
|
+
}
|
|
135
|
+
async runScenarioPair(scenarioId, baselineRunner, devflowRunner) {
|
|
136
|
+
const scenario = this.scenarios.get(scenarioId);
|
|
137
|
+
if (!scenario)
|
|
138
|
+
throw new Error(`Scenario "${scenarioId}" not found`);
|
|
139
|
+
const prompt = scenario.desc;
|
|
140
|
+
const baselineResult = await baselineRunner(prompt);
|
|
141
|
+
const devflowResult = await devflowRunner(prompt);
|
|
142
|
+
const baseline = {
|
|
143
|
+
...baselineResult,
|
|
144
|
+
scenarioId,
|
|
145
|
+
mode: "baseline",
|
|
146
|
+
timestamp: Date.now(),
|
|
147
|
+
};
|
|
148
|
+
const devflow = {
|
|
149
|
+
...devflowResult,
|
|
150
|
+
scenarioId,
|
|
151
|
+
mode: "devflow",
|
|
152
|
+
timestamp: Date.now(),
|
|
153
|
+
};
|
|
154
|
+
const comparison = {
|
|
155
|
+
timeSavedRate: baseline.durationMs > 0
|
|
156
|
+
? (baseline.durationMs - devflow.durationMs) / baseline.durationMs
|
|
157
|
+
: 0,
|
|
158
|
+
tokenReductionRate: baseline.tokenCount > 0
|
|
159
|
+
? (baseline.tokenCount - devflow.tokenCount) / baseline.tokenCount
|
|
160
|
+
: 0,
|
|
161
|
+
firstPassLift: (devflow.firstPassSuccess ? 1 : 0) -
|
|
162
|
+
(baseline.firstPassSuccess ? 1 : 0),
|
|
163
|
+
reworkReductionRate: baseline.reworkCount > 0
|
|
164
|
+
? (baseline.reworkCount - devflow.reworkCount) /
|
|
165
|
+
baseline.reworkCount
|
|
166
|
+
: 0,
|
|
167
|
+
};
|
|
168
|
+
return {
|
|
169
|
+
scenarioId,
|
|
170
|
+
baseline,
|
|
171
|
+
devflow,
|
|
172
|
+
comparison,
|
|
173
|
+
};
|
|
174
|
+
}
|
|
175
|
+
async runScenarioBatch(scenarioIds, baselineRunner, devflowRunner) {
|
|
176
|
+
const results = [];
|
|
177
|
+
for (const id of scenarioIds) {
|
|
178
|
+
const result = await this.runScenarioPair(id, baselineRunner, devflowRunner);
|
|
179
|
+
results.push(result);
|
|
180
|
+
}
|
|
181
|
+
return results;
|
|
182
|
+
}
|
|
183
|
+
}
|
|
184
|
+
//# sourceMappingURL=benchmark-engine.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"benchmark-engine.js","sourceRoot":"","sources":["../src/benchmark-engine.ts"],"names":[],"mappings":"AAYA,MAAM,aAAa,GAAoB;IACrC,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,cAAc,EAAE,aAAa,EAAE,CAAC,EAAE;IACvE,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,mBAAmB,EAAE,aAAa,EAAE,CAAC,EAAE;IAC5E,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,YAAY,EAAE,aAAa,EAAE,CAAC,EAAE;IACrE,EAAE,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,iBAAiB,EAAE,CAAC,EAAE;IACzE,EAAE,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,QAAQ,EAAE,kBAAkB,EAAE,CAAC,EAAE;IACxE,EAAE,EAAE,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,sBAAsB,EAAE,cAAc,EAAE,CAAC,EAAE;IACnF,EAAE,EAAE,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,sBAAsB,EAAE,cAAc,EAAE,CAAC,EAAE;CACpF,CAAC;AASF,MAAM,OAAO,eAAe;IAClB,OAAO,GAAG,IAAI,GAAG,EAA2B,CAAC;IAC7C,KAAK,GAAoB,CAAC,GAAG,aAAa,CAAC,CAAC;IAC5C,SAAS,GAAG,IAAI,GAAG,EAA6B,CAAC;IAEzD,KAAK,CAAC,OAAO,CAAC,IAAmB;QAC/B,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/D,IAAI,QAAQ,IAAI,CAAC,EAAE,CAAC;YAClB,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,IAAI,CAAC;QAC9B,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACxB,CAAC;IACH,CAAC;IAED,KAAK,CAAC,MAAM,CACV,MAAc,EACd,MAAkB;QAElB,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,MAAM,CAAC,CAAC;QACrD,IAAI,CAAC,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,SAAS,MAAM,aAAa,CAAC,CAAC;QACzD,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC;IACtB,CAAC;IAED,KAAK,CAAC,MAAM,CACV,MAAkB,EAClB,OAA0B;QAE1B,MAAM,OAAO,GAAG,OAAO,EAAE,OAAO;YAC9B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,OAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAC3D,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC;QAEf,MAAM,KAAK,GAAG,SAAS,IAAI,CAAC,GAAG,EAAE,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,CAAC;QAC9E,MAAM,WAAW,GAAiB,EAAE,CAAC;QAErC,KAAK,MAAM,IAAI,IAAI,OAAO,EAAE,CAAC;YAC3B,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;YAClC,WAAW,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAC3B,CAAC;QAED,MAAM,MAAM,GAAoB;YAC9B,KAAK;YACL,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;YACrB,OAAO,EAAE,IAAI,CAAC,cAAc,CAAC,WAAW,CAAC;YACzC,KAAK,EAAE,WAAW;SACnB,CAAC;QAEF,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QAChC,OAAO,MAAM,CAAC;IAChB,CAAC;IAEO,cAAc,CAAC,OAAqB;QAC1C,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACzB,OAAO;gBACL,iBAAiB,EAAE,IAAI;gBACvB,YAAY,EAAE,IAAI;gBAClB,eAAe,EAAE,IAAI;gBACrB,iBAAiB,EAAE,IAAI;aACxB,CAAC;QACJ,CAAC;QAED,MAAM,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC;QAC/E,MAAM,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QAC1E,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;QACxE,MAAM,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;QAEzB,OAAO;YACL,iBAAiB,EAAE,GAAG,CAAC,cAAc,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;YAC9D,YAAY,EAAE,IAAI,CAAC,cAAc,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;YAC1D,eAAe,EAAE,IAAI,CAAC,aAAa,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;YAC5D,iBAAiB,EAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,aAAa,GAAG,CAAC,GAAG,IAAI,EAAE,IAAI,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;SACtF,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,SAAS,CAAC,KAAa;QAC3B,OAAO,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC;IACzC,CAAC;IAED,KAAK,CAAC,WAAW;QACf,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACnD,KAAK,EAAE,CAAC,CAAC,KAAK;YACd,SAAS,EAAE,CAAC,CAAC,SAAS;YACtB,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,MAAM;SAC1B,CAAC,CAAC,CAAC;IACN,CAAC;IAED,QAAQ;QACN,OAAO,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;IACzB,CAAC;IAED,KAAK,CAAC,WAAW,CACf,QAA2B,EAC3B,IAA4B;QAE5B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,SAAS,GAAG,SAAS,CAAC;QAE5B,MAAM,MAAM,GACV,IAAI,KAAK,UAAU;YACjB,CAAC,CAAC,MAAM,IAAI,CAAC,WAAW,CAAC,QAAQ,CAAC;YAClC,CAAC,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC;QAEtC,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QAC1C,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,CAAC;QAE1C,OAAO;YACL,UAAU,EAAE,QAAQ,CAAC,EAAE;YACvB,IAAI;YACJ,UAAU;YACV,UAAU,EAAE,MAAM,CAAC,UAAU;YAC7B,gBAAgB,EAAE,MAAM,CAAC,gBAAgB;YACzC,WAAW,EAAE,MAAM,CAAC,WAAW;YAC/B,UAAU;YACV,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,SAAS,EAAE,SAAS;SACrB,CAAC;IACJ,CAAC;IAEO,KAAK,CAAC,WAAW,CACvB,SAA4B;QAO5B,OAAO;YACL,UAAU,EAAE,CAAC;YACb,gBAAgB,EAAE,KAAK;YACvB,WAAW,EAAE,CAAC;YACd,KAAK,EAAE,iDAAiD;SACzD,CAAC;IACJ,CAAC;IAEO,KAAK,CAAC,UAAU,CACtB,SAA4B;QAO5B,OAAO;YACL,UAAU,EAAE,CAAC;YACb,gBAAgB,EAAE,KAAK;YACvB,WAAW,EAAE,CAAC;YACd,KAAK,EAAE,iDAAiD;SACzD,CAAC;IACJ,CAAC;IAED,cAAc,CACZ,QAA4B,EAC5B,OAA2B;QAE3B,OAAO;YACL,aAAa,EACX,QAAQ,CAAC,UAAU,GAAG,CAAC;gBACrB,CAAC,CAAC,CAAC,QAAQ,CAAC,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC,GAAG,QAAQ,CAAC,UAAU;gBAClE,CAAC,CAAC,CAAC;YACP,kBAAkB,EAChB,QAAQ,CAAC,UAAU,GAAG,CAAC;gBACrB,CAAC,CAAC,CAAC,QAAQ,CAAC,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC,GAAG,QAAQ,CAAC,UAAU;gBAClE,CAAC,CAAC,CAAC;YACP,aAAa,EAAE,OAAO,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC/C,mBAAmB,EACjB,QAAQ,CAAC,WAAW,GAAG,CAAC;gBACtB,CAAC,CAAC,CAAC,QAAQ,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC;oBAC5C,QAAQ,CAAC,WAAW;gBACtB,CAAC,CAAC,CAAC;SACR,CAAC;IACJ,CAAC;IAED,gBAAgB,CAAC,QAA2B;QAC1C,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,CAAC;IAC5C,CAAC;IAED,KAAK,CAAC,eAAe,CACnB,UAAkB,EAClB,cAE2E,EAC3E,aAE2E;QAE3E,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;QAChD,IAAI,CAAC,QAAQ;YAAE,MAAM,IAAI,KAAK,CAAC,aAAa,UAAU,aAAa,CAAC,CAAC;QAErE,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC;QAC7B,MAAM,cAAc,GAAG,MAAM,cAAc,CAAC,MAAM,CAAC,CAAC;QACpD,MAAM,aAAa,GAAG,MAAM,aAAa,CAAC,MAAM,CAAC,CAAC;QAElD,MAAM,QAAQ,GAAuB;YACnC,GAAG,cAAc;YACjB,UAAU;YACV,IAAI,EAAE,UAA2B;YACjC,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACtB,CAAC;QAEF,MAAM,OAAO,GAAuB;YAClC,GAAG,aAAa;YAChB,UAAU;YACV,IAAI,EAAE,SAA0B;YAChC,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACtB,CAAC;QAEF,MAAM,UAAU,GAAqB;YACnC,aAAa,EACX,QAAQ,CAAC,UAAU,GAAG,CAAC;gBACrB,CAAC,CAAC,CAAC,QAAQ,CAAC,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC,GAAG,QAAQ,CAAC,UAAU;gBAClE,CAAC,CAAC,CAAC;YACP,kBAAkB,EAChB,QAAQ,CAAC,UAAU,GAAG,CAAC;gBACrB,CAAC,CAAC,CAAC,QAAQ,CAAC,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC,GAAG,QAAQ,CAAC,UAAU;gBAClE,CAAC,CAAC,CAAC;YACP,aAAa,EACX,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;gBAClC,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACrC,mBAAmB,EACjB,QAAQ,CAAC,WAAW,GAAG,CAAC;gBACtB,CAAC,CAAC,CAAC,QAAQ,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC;oBAC5C,QAAQ,CAAC,WAAW;gBACtB,CAAC,CAAC,CAAC;SACR,CAAC;QAEF,OAAO;YACL,UAAU;YACV,QAAQ;YACR,OAAO;YACP,UAAU;SACX,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,gBAAgB,CACpB,WAAqB,EACrB,cAE2E,EAC3E,aAE2E;QAE3E,MAAM,OAAO,GAAgC,EAAE,CAAC;QAChD,KAAK,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC;YAC7B,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,eAAe,CAAC,EAAE,EAAE,cAAc,EAAE,aAAa,CAAC,CAAC;YAC7E,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACvB,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;CACF"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"debug-buildcontext.d.ts","sourceRoot":"","sources":["../src/debug-buildcontext.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
import CGModule from "@colbymchenry/codegraph";
|
|
2
|
+
async function main() {
|
|
3
|
+
const CG = CGModule.default ?? CGModule;
|
|
4
|
+
const cg = CG.openSync(process.cwd());
|
|
5
|
+
const result = await cg.buildContext("post-tool-use hook", {
|
|
6
|
+
searchLimit: 5,
|
|
7
|
+
traversalDepth: 2,
|
|
8
|
+
maxNodes: 10,
|
|
9
|
+
includeCode: false,
|
|
10
|
+
format: "json",
|
|
11
|
+
});
|
|
12
|
+
const obj = JSON.parse(result);
|
|
13
|
+
console.log("Top-level keys:", Object.keys(obj));
|
|
14
|
+
console.log("subgraph exists:", "subgraph" in obj);
|
|
15
|
+
console.log("subgraph:", obj.subgraph ? Object.keys(obj.subgraph) : "N/A");
|
|
16
|
+
if (obj.subgraph?.nodes) {
|
|
17
|
+
console.log("subgraph.nodes is Array:", Array.isArray(obj.subgraph.nodes));
|
|
18
|
+
console.log("subgraph.nodes count:", obj.subgraph.nodes.length);
|
|
19
|
+
if (obj.subgraph.nodes.length > 0) {
|
|
20
|
+
console.log("First node:", JSON.stringify(obj.subgraph.nodes[0]).slice(0, 200));
|
|
21
|
+
}
|
|
22
|
+
}
|
|
23
|
+
console.log("entryPoints:", obj.entryPoints?.length);
|
|
24
|
+
console.log("stats:", obj.stats);
|
|
25
|
+
console.log("summary:", obj.summary);
|
|
26
|
+
}
|
|
27
|
+
main();
|
|
28
|
+
//# sourceMappingURL=debug-buildcontext.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"debug-buildcontext.js","sourceRoot":"","sources":["../src/debug-buildcontext.ts"],"names":[],"mappings":"AAAA,OAAO,QAAQ,MAAM,yBAAyB,CAAC;AAC/C,KAAK,UAAU,IAAI;IACjB,MAAM,EAAE,GAAK,QAAgB,CAAC,OAA2B,IAAI,QAAQ,CAAC;IACtE,MAAM,EAAE,GAAG,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,CAAC;IAEtC,MAAM,MAAM,GAAG,MAAO,EAAU,CAAC,YAAY,CAAC,oBAAoB,EAAE;QAClE,WAAW,EAAE,CAAC;QACd,cAAc,EAAE,CAAC;QACjB,QAAQ,EAAE,EAAE;QACZ,WAAW,EAAE,KAAK;QAClB,MAAM,EAAE,MAAM;KACf,CAAC,CAAC;IAEH,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,MAAgB,CAAC,CAAC;IACzC,OAAO,CAAC,GAAG,CAAC,iBAAiB,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,OAAO,CAAC,GAAG,CAAC,kBAAkB,EAAE,UAAU,IAAI,GAAG,CAAC,CAAC;IACnD,OAAO,CAAC,GAAG,CAAC,WAAW,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC;IAC3E,IAAI,GAAG,CAAC,QAAQ,EAAE,KAAK,EAAE,CAAC;QACxB,OAAO,CAAC,GAAG,CAAC,0BAA0B,EAAE,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC;QAC3E,OAAO,CAAC,GAAG,CAAC,uBAAuB,EAAE,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QAChE,IAAI,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAClC,OAAO,CAAC,GAAG,CAAC,aAAa,EAAE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;QAClF,CAAC;IACH,CAAC;IACD,OAAO,CAAC,GAAG,CAAC,cAAc,EAAE,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC;IACrD,OAAO,CAAC,GAAG,CAAC,QAAQ,EAAE,GAAG,CAAC,KAAK,CAAC,CAAC;IACjC,OAAO,CAAC,GAAG,CAAC,UAAU,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC;AACvC,CAAC;AACD,IAAI,EAAE,CAAC"}
|
package/dist/index.d.ts
ADDED
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,YAAY,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC"}
|
package/dist/index.js
ADDED
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"run-accuracy-tests.d.ts","sourceRoot":"","sources":["../src/run-accuracy-tests.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,75 @@
|
|
|
1
|
+
import { createEngines } from '@devflow-tools/context-engine';
|
|
2
|
+
import { ACCURACY_TESTS } from './accuracy-queries.js';
|
|
3
|
+
async function runTest(test, engines) {
|
|
4
|
+
switch (test.engine) {
|
|
5
|
+
case 'codegraph': {
|
|
6
|
+
const result = await engines.context.buildTaskContext(test.query, { mode: 'deep' });
|
|
7
|
+
const allFiles = [...result.requiredFiles.files, ...result.optionalFiles.files];
|
|
8
|
+
const allSymbols = [...result.requiredFiles.symbols, ...result.optionalFiles.symbols];
|
|
9
|
+
const fileMatch = !test.expectFiles || test.expectFiles.some(f => allFiles.some(af => af.includes(f)));
|
|
10
|
+
const symbolMatch = !test.expectSymbols || test.expectSymbols.some(s => allSymbols.some(as => as.includes(s)));
|
|
11
|
+
const countOk = allFiles.length >= test.minResults;
|
|
12
|
+
const passed = fileMatch && symbolMatch && countOk;
|
|
13
|
+
return {
|
|
14
|
+
passed,
|
|
15
|
+
details: passed
|
|
16
|
+
? `OK: ${allFiles.length} files, ${allSymbols.length} symbols`
|
|
17
|
+
: `FAIL: files=${allFiles.join(', ')}, symbols=${allSymbols.join(', ')}. Expected: files=${test.expectFiles}, symbols=${test.expectSymbols}, min=${test.minResults}`,
|
|
18
|
+
};
|
|
19
|
+
}
|
|
20
|
+
case 'knowledge': {
|
|
21
|
+
const result = await engines.knowledge.search(test.query, { limit: test.minResults * 2 });
|
|
22
|
+
const sourceMatch = !test.expectSource || result.some(r => r.source === test.expectSource);
|
|
23
|
+
const countOk = result.length >= test.minResults;
|
|
24
|
+
const passed = sourceMatch && countOk;
|
|
25
|
+
return {
|
|
26
|
+
passed,
|
|
27
|
+
details: passed
|
|
28
|
+
? `OK: ${result.length} results from ${[...new Set(result.map(r => r.source))].join(', ')}`
|
|
29
|
+
: `FAIL: ${result.length} results. Expected source=${test.expectSource}, min=${test.minResults}`,
|
|
30
|
+
};
|
|
31
|
+
}
|
|
32
|
+
case 'memory': {
|
|
33
|
+
const all = await engines.memory.getAll();
|
|
34
|
+
const conventions = all.project?.conventions ?? [];
|
|
35
|
+
const memCount = Array.isArray(conventions) ? conventions.length : 0;
|
|
36
|
+
const passed = memCount >= test.minResults;
|
|
37
|
+
return { passed, details: passed ? `OK: ${memCount} memories` : `FAIL: ${memCount} memories, min=${test.minResults}` };
|
|
38
|
+
}
|
|
39
|
+
}
|
|
40
|
+
}
|
|
41
|
+
async function main() {
|
|
42
|
+
const projectRoot = process.env.CLAUDE_PROJECT_DIR ?? process.cwd();
|
|
43
|
+
console.log(`Running accuracy tests for ${projectRoot}...\n`);
|
|
44
|
+
const engines = await createEngines(projectRoot);
|
|
45
|
+
let passed = 0;
|
|
46
|
+
let failed = 0;
|
|
47
|
+
for (const test of ACCURACY_TESTS) {
|
|
48
|
+
if (test.skipReason) {
|
|
49
|
+
console.log(`SKIP: "${test.query}" — ${test.skipReason}`);
|
|
50
|
+
continue;
|
|
51
|
+
}
|
|
52
|
+
try {
|
|
53
|
+
const result = await runTest(test, engines);
|
|
54
|
+
if (result.passed) {
|
|
55
|
+
console.log(`PASS: [${test.engine}] "${test.query}" — ${result.details}`);
|
|
56
|
+
passed++;
|
|
57
|
+
}
|
|
58
|
+
else {
|
|
59
|
+
console.log(`FAIL: [${test.engine}] "${test.query}" — ${result.details}`);
|
|
60
|
+
failed++;
|
|
61
|
+
}
|
|
62
|
+
}
|
|
63
|
+
catch (err) {
|
|
64
|
+
console.log(`FAIL: [${test.engine}] "${test.query}" — error: ${err.message}`);
|
|
65
|
+
failed++;
|
|
66
|
+
}
|
|
67
|
+
}
|
|
68
|
+
console.log(`\n${passed}/${passed + failed} passed`);
|
|
69
|
+
process.exit(failed > 0 ? 1 : 0);
|
|
70
|
+
}
|
|
71
|
+
main().catch(err => {
|
|
72
|
+
console.error('FATAL:', err);
|
|
73
|
+
process.exit(1);
|
|
74
|
+
});
|
|
75
|
+
//# sourceMappingURL=run-accuracy-tests.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"run-accuracy-tests.js","sourceRoot":"","sources":["../src/run-accuracy-tests.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,+BAA+B,CAAC;AAC9D,OAAO,EAAE,cAAc,EAAyB,MAAM,uBAAuB,CAAC;AAE9E,KAAK,UAAU,OAAO,CAAC,IAAsB,EAAE,OAAkD;IAC/F,QAAQ,IAAI,CAAC,MAAM,EAAE,CAAC;QACpB,KAAK,WAAW,CAAC,CAAC,CAAC;YACjB,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,OAAO,CAAC,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;YACpF,MAAM,QAAQ,GAAG,CAAC,GAAG,MAAM,CAAC,aAAa,CAAC,KAAK,EAAE,GAAG,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC;YAChF,MAAM,UAAU,GAAG,CAAC,GAAG,MAAM,CAAC,aAAa,CAAC,OAAO,EAAE,GAAG,MAAM,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;YAEtF,MAAM,SAAS,GAAG,CAAC,IAAI,CAAC,WAAW,IAAI,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACvG,MAAM,WAAW,GAAG,CAAC,IAAI,CAAC,aAAa,IAAI,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC/G,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,IAAI,IAAI,CAAC,UAAU,CAAC;YACnD,MAAM,MAAM,GAAG,SAAS,IAAI,WAAW,IAAI,OAAO,CAAC;YACnD,OAAO;gBACL,MAAM;gBACN,OAAO,EAAE,MAAM;oBACb,CAAC,CAAC,OAAO,QAAQ,CAAC,MAAM,WAAW,UAAU,CAAC,MAAM,UAAU;oBAC9D,CAAC,CAAC,eAAe,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,qBAAqB,IAAI,CAAC,WAAW,aAAa,IAAI,CAAC,aAAa,SAAS,IAAI,CAAC,UAAU,EAAE;aACvK,CAAC;QACJ,CAAC;QACD,KAAK,WAAW,CAAC,CAAC,CAAC;YACjB,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,EAAE,EAAE,KAAK,EAAE,IAAI,CAAC,UAAU,GAAG,CAAC,EAAE,CAAC,CAAC;YAC1F,MAAM,WAAW,GAAG,CAAC,IAAI,CAAC,YAAY,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,YAAY,CAAC,CAAC;YAC3F,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,IAAI,IAAI,CAAC,UAAU,CAAC;YACjD,MAAM,MAAM,GAAG,WAAW,IAAI,OAAO,CAAC;YACtC,OAAO;gBACL,MAAM;gBACN,OAAO,EAAE,MAAM;oBACb,CAAC,CAAC,OAAO,MAAM,CAAC,MAAM,iBAAiB,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;oBAC3F,CAAC,CAAC,SAAS,MAAM,CAAC,MAAM,6BAA6B,IAAI,CAAC,YAAY,SAAS,IAAI,CAAC,UAAU,EAAE;aACnG,CAAC;QACJ,CAAC;QACD,KAAK,QAAQ,CAAC,CAAC,CAAC;YACd,MAAM,GAAG,GAAG,MAAM,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YAC1C,MAAM,WAAW,GAAI,GAAW,CAAC,OAAO,EAAE,WAAW,IAAI,EAAE,CAAC;YAC5D,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;YACrE,MAAM,MAAM,GAAG,QAAQ,IAAI,IAAI,CAAC,UAAU,CAAC;YAC3C,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,QAAQ,WAAW,CAAC,CAAC,CAAC,SAAS,QAAQ,kBAAkB,IAAI,CAAC,UAAU,EAAE,EAAE,CAAC;QACzH,CAAC;IACH,CAAC;AACH,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,kBAAkB,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;IACpE,OAAO,CAAC,GAAG,CAAC,8BAA8B,WAAW,OAAO,CAAC,CAAC;IAE9D,MAAM,OAAO,GAAG,MAAM,aAAa,CAAC,WAAW,CAAC,CAAC;IACjD,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,MAAM,GAAG,CAAC,CAAC;IAEf,KAAK,MAAM,IAAI,IAAI,cAAc,EAAE,CAAC;QAClC,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC;YACpB,OAAO,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,KAAK,OAAO,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC;YAC1D,SAAS;QACX,CAAC;QACD,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;YAC5C,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;gBAClB,OAAO,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,MAAM,MAAM,IAAI,CAAC,KAAK,OAAO,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;gBAC1E,MAAM,EAAE,CAAC;YACX,CAAC;iBAAM,CAAC;gBACN,OAAO,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,MAAM,MAAM,IAAI,CAAC,KAAK,OAAO,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;gBAC1E,MAAM,EAAE,CAAC;YACX,CAAC;QACH,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,MAAM,MAAM,IAAI,CAAC,KAAK,cAAe,GAAa,CAAC,OAAO,EAAE,CAAC,CAAC;YACzF,MAAM,EAAE,CAAC;QACX,CAAC;IACH,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,KAAK,MAAM,IAAI,MAAM,GAAG,MAAM,SAAS,CAAC,CAAC;IACrD,OAAO,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACnC,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE;IACjB,OAAO,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;IAC7B,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@devflow-tools/benchmark",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.13.1",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"main": "./dist/index.js",
|
|
6
6
|
"types": "./dist/index.d.ts",
|
|
@@ -18,8 +18,8 @@
|
|
|
18
18
|
"accuracy": "npx tsx src/run-accuracy-tests.ts"
|
|
19
19
|
},
|
|
20
20
|
"dependencies": {
|
|
21
|
-
"@devflow-tools/context-engine": "^0.
|
|
22
|
-
"@devflow-tools/sdk": "^0.
|
|
21
|
+
"@devflow-tools/context-engine": "^0.13.1",
|
|
22
|
+
"@devflow-tools/sdk": "^0.13.1"
|
|
23
23
|
},
|
|
24
24
|
"devDependencies": {
|
|
25
25
|
"typescript": "^5.5.0",
|
|
@@ -28,5 +28,5 @@
|
|
|
28
28
|
"files": [
|
|
29
29
|
"dist"
|
|
30
30
|
],
|
|
31
|
-
"gitHead": "
|
|
31
|
+
"gitHead": "28618c4d84bea33bdd4f433bfc740c1ce2591d51"
|
|
32
32
|
}
|