geo/apps/desktop-client/src/main/adapters/yuanbao.test.ts

import { describe, expect, it } from 'vitest'

import { __yuanbaoTestUtils } from './yuanbao'

const {
  appendUniqueSourceItems,
  buildOrderedDomSourceItems,
  buildSourceItem,
  extractCitationMarkerIndexes,
  extractQuestionText,
  findUnresolvedCitationIndexes,
  mergeText,
  normalizeCitationMarkers,
  normalizeSourceUrl,
  parseYuanbaoCaptures,
  resolveCitationsFromMarkers,
} = __yuanbaoTestUtils

describe('yuanbao adapter helpers', () => {
  it('unwraps redirect URLs and reads Yuanbao document URL fields', () => {
    expect(
      normalizeSourceUrl(
        'https://yuanbao.tencent.com/link?target=https%3A%2F%2Fexample.com%2Fa%23ref',
      ),
    ).toBe('https://example.com/a')

    expect(
      buildSourceItem({
        docUrl: 'https://example.com/doc?from=yuanbao#source',
        title: '引用文章',
        siteName: 'Example',
      }),
    ).toMatchObject({
      url: 'https://example.com/doc?from=yuanbao',
      normalized_url: 'https://example.com/doc?from=yuanbao',
      title: '引用文章',
      site_name: 'Example',
      host: 'example.com',
    })
  })

  it('rejects browser and build diagnostics as monitoring questions', () => {
    expect(extractQuestionText({ question_text: '合肥全屋定制推荐' })).toBe('合肥全屋定制推荐')
    expect(() => extractQuestionText({ title: '监控任务 · 混元 / 元宝' })).toThrow(
      'yuanbao_question_text_missing',
    )
    expect(() =>
      extractQuestionText({
        question_text:
          "The resource http://localhost:8000/static/js/runtime.c632b2fe.js was preloaded using link preload but not used within a few seconds from the window's load event.",
      }),
    ).toThrow('yuanbao_question_text_runtime_diagnostic')
    expect(() =>
      extractQuestionText({
        question_text:
          'failed to solve: process "/bin/sh -c go build" did not complete successfully: exit code: 1',
      }),
    ).toThrow('yuanbao_question_text_runtime_diagnostic')
  })

  it('merges streaming answer fragments without duplicating a repeated prefix', () => {
    const first =
      '通过三轮搜索，我已经掌握了合肥全屋定制市场的高性价比品牌。一定要合同中的定标准确和责任。'
    const repeated =
      '通过三轮搜索，我已经掌握了合肥全屋定制市场的高性价比品牌。结合你的预算，我建议优先看本土老牌。'

    expect(mergeText(first, repeated)).toBe(repeated)
  })

  it('resolves citation markers against the best available source pool', () => {
    const sources = [
      { url: 'https://example.com/1', normalized_url: 'https://example.com/1' },
      { url: 'https://example.com/2', normalized_url: 'https://example.com/2' },
      { url: 'https://example.com/3', normalized_url: 'https://example.com/3' },
    ]

    expect(extractCitationMarkerIndexes('正文 [citation:1] 和 [citation:3]')).toEqual([1, 3])
    expect(
      resolveCitationsFromMarkers('正文 [citation:1] 和 [citation:3]', [[], sources]).map(
        (item) => item.url,
      ),
    ).toEqual(['https://example.com/1', 'https://example.com/3'])
  })

  it('preserves indexed DOM citation slots even when URLs repeat', () => {
    const domSources = buildOrderedDomSourceItems([
      {
        url: 'https://example.com/repeated',
        title: '来源 1',
        text: null,
        siteName: 'Example',
        sourceIndex: 1,
      },
      {
        url: 'https://example.com/repeated',
        title: '来源 2',
        text: null,
        siteName: 'Example',
        sourceIndex: 2,
      },
      {
        url: 'https://example.com/third',
        title: '来源 3',
        text: null,
        siteName: 'Example',
        sourceIndex: 3,
      },
    ])

    expect(domSources).toHaveLength(3)
    expect(domSources.map((item) => item.title)).toEqual(['来源 1', '来源 2', '来源 3'])
    expect(findUnresolvedCitationIndexes('正文 [citation:2] [citation:3]', domSources)).toEqual([])
  })

  it('keeps primary DOM citation order when appending parsed extras', () => {
    const primary = [
      { url: 'https://example.com/1', normalized_url: 'https://example.com/1' },
      { url: 'https://example.com/2', normalized_url: 'https://example.com/2' },
    ]
    const appended = appendUniqueSourceItems(primary, [
      { url: 'https://example.com/2', normalized_url: 'https://example.com/2', title: 'duplicate' },
      { url: 'https://example.com/3', normalized_url: 'https://example.com/3' },
    ])

    expect(appended.map((item) => item.url)).toEqual([
      'https://example.com/1',
      'https://example.com/2',
      'https://example.com/3',
    ])
  })

  it('does not report incomplete citations when the answer has no markers', () => {
    expect(findUnresolvedCitationIndexes('正文没有引用小图标', [])).toEqual([])
  })

  it("recognises both '[N]' and '[citation:N]' inline markers", () => {
    // The new yuanbao model lowers <div data-idx-list="1,2,10"> placeholders to "[1][2][10]";
    // older SSE frames may still emit "[citation:N]". Both must extract the same indexes.
    expect(extractCitationMarkerIndexes('合肥老牌选 [1][2] 性价比 [10]。')).toEqual([1, 2, 10])
    expect(extractCitationMarkerIndexes('混合两种格式 [citation:3] 还有 [4]')).toEqual([3, 4])

    const sources = [
      { url: 'https://example.com/1', normalized_url: 'https://example.com/1' },
      { url: 'https://example.com/2', normalized_url: 'https://example.com/2' },
    ]
    expect(
      resolveCitationsFromMarkers('正文 [1] 然后 [2]', [[], sources]).map((item) => item.url),
    ).toEqual(['https://example.com/1', 'https://example.com/2'])
  })

  it('rewrites legacy [citation:N] markers to bare [N] for SaaS rendering', () => {
    expect(normalizeCitationMarkers('正文 [citation:1] 和 [citation:10]')).toBe('正文 [1] 和 [10]')
    // Already-normalised text passes through.
    expect(normalizeCitationMarkers('已经是 [1][2] 格式')).toBe('已经是 [1][2] 格式')
    // Null/empty pass through.
    expect(normalizeCitationMarkers(null)).toBeNull()
    expect(normalizeCitationMarkers('')).toBe('')
  })

  it('does not invent citations for fast-cache answers (no searchGuid frame, no [N] markers)', () => {
    // hunyuan_t1 sometimes returns a cached / non-search answer ("已深度思考(用时1秒)"
    // with no "找到了 N 篇相关资料" panel). The SSE then carries only `text`
    // frames and no `searchGuid`. The adapter must still surface that answer to
    // the SaaS pipeline — see the top-level query path: as long as `answer` is
    // non-empty and there are no [N] markers, neither the empty-response nor
    // the incomplete-citations branches fire, so we fall through to status=succeeded.
    const sseFrames = [
      JSON.stringify({ type: 'text', msg: '在合肥，' }),
      JSON.stringify({ type: 'text', msg: '全屋定制市场非常成熟，' }),
      JSON.stringify({ type: 'text', msg: '没有绝对"最好"的商家。' }),
    ]
      .map((line) => `data: ${line}`)
      .join('\n\n')

    const summary = parseYuanbaoCaptures([
      {
        url: 'https://yuanbao.tencent.com/api/chat/cache-hit',
        status: 200,
        contentType: 'text/event-stream',
        body: sseFrames,
      },
    ])

    expect(summary.answer).toBe('在合肥，全屋定制市场非常成熟，没有绝对"最好"的商家。')
    expect(summary.citations).toEqual([])
    expect(summary.searchResults).toEqual([])
    // No marker → no unresolved indexes, so the adapter does not gate on citations.
    expect(extractCitationMarkerIndexes(summary.answer)).toEqual([])
    expect(findUnresolvedCitationIndexes(summary.answer, summary.citations)).toEqual([])
  })

  it('aligns citations[index-1] with answer markers when SSE places refs in searchGuid.docs (new hunyuan_t1 protocol)', () => {
    // Captured against the live yuanbao endpoint: deep-search now ships every
    // reference inside `searchGuid.docs[]` with a 1-based `index`, and the
    // legacy `citations` field is always null in this frame. Whatever the
    // adapter normalises to must preserve idx alignment so that the
    // unresolved-citations check passes.
    const sseFrames = [
      JSON.stringify({
        type: 'searchGuid',
        title: '引用 3 篇资料作为参考',
        docs: [
          {
            index: 1,
            title: '志邦家居官网',
            url: 'https://www.zbom.com/about',
            web_url: 'https://www.zbom.com/about',
            web_site_name: '志邦家居',
          },
          {
            index: 2,
            title: '合肥本地装修评测',
            url: 'https://example.com/hefei-review',
            web_site_name: '今日头条',
          },
          {
            index: 3,
            title: '客来福官方介绍',
            url: 'https://example.com/kelaifu',
            web_site_name: '客来福',
          },
        ],
        citations: null,
      }),
      JSON.stringify({
        type: 'text',
        msg: '志邦家居 [1] 与客来福 [3] 都是合肥本土头部 [2]。',
      }),
    ]
      .map((line) => `data: ${line}`)
      .join('\n\n')

    const summary = parseYuanbaoCaptures([
      {
        url: 'https://yuanbao.tencent.com/api/chat/abc',
        status: 200,
        contentType: 'text/event-stream',
        body: sseFrames,
      },
    ])

    expect(summary.citations.map((item) => item.url)).toEqual([
      'https://www.zbom.com/about',
      'https://example.com/hefei-review',
      'https://example.com/kelaifu',
    ])
    expect(findUnresolvedCitationIndexes(summary.answer, summary.citations)).toEqual([])
  })

  it('collects source candidates from deep search thinking fields', () => {
    const summary = parseYuanbaoCaptures([
      {
        url: 'https://yuanbao.tencent.com/api/chat',
        status: 200,
        contentType: 'text/event-stream',
        body: [
          'data: {"type":"deepSearch","contents":[{"type":"step","webPages":[{"targetUrl":"https://example.com/a#ref","title":"来源 A","siteName":"Example"}],"references":[{"sourceUrl":"https://example.com/b","title":"来源 B"}]}]}',
          'data: {"type":"text","msg":"正文 [citation:1] [citation:2]"}',
        ].join('\n\n'),
      },
    ])

    expect(summary.searchResults.map((item) => item.url)).toContain('https://example.com/a')
    expect(summary.citations.map((item) => item.url)).toContain('https://example.com/b')
    expect(
      resolveCitationsFromMarkers(summary.answer, [
        summary.citations,
        summary.searchResults,
        [...summary.citations, ...summary.searchResults],
      ]).map((item) => item.url),
    ).toEqual(['https://example.com/b', 'https://example.com/a'])
  })
})