feat: initial release v0.3.0
This commit is contained in:
saturn
185
tests/unit/helpers/json-repair.test.ts
Normal file
185
tests/unit/helpers/json-repair.test.ts
Normal file
@@ -0,0 +1,185 @@
|
||||
import { describe, expect, it } from 'vitest'
|
||||
import { safeParseJson, safeParseJsonObject, safeParseJsonArray } from '@/lib/json-repair'
|
||||
|
||||
// ─── safeParseJson ───────────────────────────────────────────────────
|
||||
|
||||
describe('safeParseJson', () => {
|
||||
it('正常 JSON 字符串 -> 直接解析成功', () => {
|
||||
const result = safeParseJson('{"name":"孙悟空","age":500}')
|
||||
expect(result).toEqual({ name: '孙悟空', age: 500 })
|
||||
})
|
||||
|
||||
it('包含 markdown 代码块 -> 剥离后解析成功', () => {
|
||||
const input = '```json\n{"key":"value"}\n```'
|
||||
const result = safeParseJson(input)
|
||||
expect(result).toEqual({ key: 'value' })
|
||||
})
|
||||
|
||||
it('包含大写 JSON 标记的 markdown 代码块 -> 剥离后解析成功', () => {
|
||||
const input = '```JSON\n{"key":"value"}\n```'
|
||||
const result = safeParseJson(input)
|
||||
expect(result).toEqual({ key: 'value' })
|
||||
})
|
||||
|
||||
it('尾部逗号 -> jsonrepair 修复后解析成功', () => {
|
||||
const input = '{"a":1,"b":2,}'
|
||||
const result = safeParseJson(input)
|
||||
expect(result).toEqual({ a: 1, b: 2 })
|
||||
})
|
||||
|
||||
it('单引号包裹字符串 -> jsonrepair 修复后解析成功', () => {
|
||||
const input = "{'name':'张三','age':25}"
|
||||
const result = safeParseJson(input)
|
||||
expect(result).toEqual({ name: '张三', age: 25 })
|
||||
})
|
||||
|
||||
it('JSON 前后有多余文字 -> jsonrepair 修复后解析成功', () => {
|
||||
const input = '以下是分析结果：\n{"result":"success"}\n以上是所有内容。'
|
||||
const result = safeParseJson(input)
|
||||
expect(result).toEqual({ result: 'success' })
|
||||
})
|
||||
|
||||
it('完全无效内容（无任何 JSON 结构字符）-> jsonrepair 将其视为字符串', () => {
|
||||
// jsonrepair 会把纯文本修复为 JSON 字符串
|
||||
const result = safeParseJson('这不是JSON')
|
||||
expect(result).toBe('这不是JSON')
|
||||
})
|
||||
})
|
||||
|
||||
// ─── safeParseJsonObject ─────────────────────────────────────────────
|
||||
|
||||
describe('safeParseJsonObject', () => {
|
||||
it('正常 JSON 对象 -> 返回对象', () => {
|
||||
const result = safeParseJsonObject('{"characters":[],"locations":[]}')
|
||||
expect(result).toEqual({ characters: [], locations: [] })
|
||||
})
|
||||
|
||||
it('markdown 包裹的 JSON 对象 -> 剥离后返回对象', () => {
|
||||
const input = '```json\n{"episodes":[{"number":1}]}\n```'
|
||||
const result = safeParseJsonObject(input)
|
||||
expect(result).toHaveProperty('episodes')
|
||||
expect((result.episodes as unknown[])[0]).toEqual({ number: 1 })
|
||||
})
|
||||
|
||||
it('包含中文角引号「」的内容 -> 正常解析保留', () => {
|
||||
const input = '{"lines":"孙悟空怒道，「一个冒牌货，也敢拦你孙爷爷的路！」"}'
|
||||
const result = safeParseJsonObject(input)
|
||||
expect(result.lines).toBe('孙悟空怒道，「一个冒牌货，也敢拦你孙爷爷的路！」')
|
||||
})
|
||||
|
||||
it('LLM 输出数组而非对象 -> 抛出 Expected JSON object 错误', () => {
|
||||
expect(() => safeParseJsonObject('[1,2,3]')).toThrow('Expected JSON object')
|
||||
})
|
||||
|
||||
it('尾部逗号 + markdown 包裹 -> 修复后返回正确对象', () => {
|
||||
const input = '```json\n{"a":1,"b":"hello",}\n```'
|
||||
const result = safeParseJsonObject(input)
|
||||
expect(result).toEqual({ a: 1, b: 'hello' })
|
||||
})
|
||||
})
|
||||
|
||||
// ─── safeParseJsonArray ──────────────────────────────────────────────
|
||||
|
||||
describe('safeParseJsonArray', () => {
|
||||
it('正常 JSON 数组 -> 返回对象数组', () => {
|
||||
const input = '[{"id":1,"name":"角色A"},{"id":2,"name":"角色B"}]'
|
||||
const result = safeParseJsonArray(input)
|
||||
expect(result).toHaveLength(2)
|
||||
expect(result[0]).toEqual({ id: 1, name: '角色A' })
|
||||
expect(result[1]).toEqual({ id: 2, name: '角色B' })
|
||||
})
|
||||
|
||||
it('对象包裹数组 + fallbackKey -> 提取内部数组', () => {
|
||||
const input = '{"clips":[{"id":1},{"id":2}]}'
|
||||
const result = safeParseJsonArray(input, 'clips')
|
||||
expect(result).toHaveLength(2)
|
||||
expect(result[0]).toEqual({ id: 1 })
|
||||
})
|
||||
|
||||
it('对象包裹数组 + 无 fallbackKey -> 自动发现第一个数组字段', () => {
|
||||
const input = '{"episodes":[{"number":1},{"number":2}]}'
|
||||
const result = safeParseJsonArray(input)
|
||||
expect(result).toHaveLength(2)
|
||||
expect(result[0]).toEqual({ number: 1 })
|
||||
})
|
||||
|
||||
it('markdown 包裹 + 尾部逗号 -> 修复后返回正确数组', () => {
|
||||
const input = '```json\n[{"a":1},{"b":2},]\n```'
|
||||
const result = safeParseJsonArray(input)
|
||||
expect(result).toHaveLength(2)
|
||||
expect(result[0]).toEqual({ a: 1 })
|
||||
expect(result[1]).toEqual({ b: 2 })
|
||||
})
|
||||
|
||||
it('过滤非对象元素（数字、字符串等）-> 只保留对象', () => {
|
||||
const input = '[{"valid":true}, 42, "string", null, {"also":true}]'
|
||||
const result = safeParseJsonArray(input)
|
||||
expect(result).toHaveLength(2)
|
||||
expect(result[0]).toEqual({ valid: true })
|
||||
expect(result[1]).toEqual({ also: true })
|
||||
})
|
||||
|
||||
it('空数组 -> 返回空数组', () => {
|
||||
const result = safeParseJsonArray('[]')
|
||||
expect(result).toHaveLength(0)
|
||||
})
|
||||
|
||||
it('非数组非对象 -> 抛出错误', () => {
|
||||
expect(() => safeParseJsonArray('"just a string"')).toThrow('Expected JSON array')
|
||||
})
|
||||
|
||||
it('对象不含数组字段 -> 抛出错误', () => {
|
||||
expect(() => safeParseJsonArray('{"key":"value"}')).toThrow('Expected JSON array')
|
||||
})
|
||||
})
|
||||
|
||||
// ─── 真实 LLM 畸形输出回归测试 ───────────────────────────────────────
|
||||
|
||||
describe('LLM 畸形 JSON 输出回归测试', () => {
|
||||
it('中文弯引号嵌套在 JSON 值中 -> jsonrepair 修复成功', () => {
|
||||
// 这是导致 "Invalid clip JSON format" 的典型场景
|
||||
const llmOutput = '```json\n[{"description":"孙悟空怒道，\\u201c一个冒牌货！\\u201d"}]\n```'
|
||||
const result = safeParseJsonArray(llmOutput)
|
||||
expect(result).toHaveLength(1)
|
||||
expect(result[0].description).toContain('孙悟空')
|
||||
})
|
||||
|
||||
it('LLM 输出前后带解释文字 -> 提取并解析 JSON', () => {
|
||||
const llmOutput = `好的，以下是分析结果：
|
||||
|
||||
{"locations":[{"name":"客厅_白天","summary":"主角居住的客厅"}]}
|
||||
|
||||
以上是所有场景分析。`
|
||||
const result = safeParseJsonObject(llmOutput)
|
||||
expect(result.locations).toBeDefined()
|
||||
const locations = result.locations as unknown[]
|
||||
expect(locations).toHaveLength(1)
|
||||
})
|
||||
|
||||
it('使用「」角引号的台词内容 -> 正确解析不破坏 JSON', () => {
|
||||
// 改造后的提示词要求 LLM 用「」替代引号
|
||||
const llmOutput = '[{"speaker":"孙悟空","content":"「你竟敢拦我的路！」","emotionStrength":0.4}]'
|
||||
const result = safeParseJsonArray(llmOutput)
|
||||
expect(result).toHaveLength(1)
|
||||
expect(result[0].speaker).toBe('孙悟空')
|
||||
expect(result[0].content).toBe('「你竟敢拦我的路！」')
|
||||
expect(result[0].emotionStrength).toBe(0.4)
|
||||
})
|
||||
|
||||
it('带控制字符的 JSON -> jsonrepair 修复成功', () => {
|
||||
// LLM 有时在字符串值中输出真实换行符
|
||||
const llmOutput = '{"text":"第一行\\n第二行","count":2}'
|
||||
const result = safeParseJsonObject(llmOutput)
|
||||
expect(result.text).toBe('第一行\n第二行')
|
||||
expect(result.count).toBe(2)
|
||||
})
|
||||
|
||||
it('clips 包裹在对象中 -> 正确提取', () => {
|
||||
// clips-build 中常见的 LLM 输出格式
|
||||
const llmOutput = '{"clips":[{"id":"clip_1","startText":"从前"},{"id":"clip_2","startText":"后来"}]}'
|
||||
const result = safeParseJsonArray(llmOutput, 'clips')
|
||||
expect(result).toHaveLength(2)
|
||||
expect(result[0].id).toBe('clip_1')
|
||||
expect(result[1].startText).toBe('后来')
|
||||
})
|
||||
})
|
||||
Reference in New Issue
Block a user