xiaoxia-saas/tests/unit/test_viral_parser_field_mapping.py

"""#2308 parser字段映射单测：
1. 新schema XML（<shot>+<visual>+<action_details>）字段正确映射
2. 旧schema XML（<visual>混写）能拆分出cam/visual/action
3. markdown中文分镜字段正确提取
"""

from __future__ import annotations

import sys
from unittest.mock import MagicMock

import pytest


@pytest.fixture
def mock_job():
    job = MagicMock()
    job.duration = 15
    job.video_ratio = "9:16"
    job.marketing_purpose = "品牌推广"
    job.image_analysis = {"images": []}
    job.images = []
    job.user_copy_text = ""
    job.fusion_level = "ai_polish"
    return job


def _import_wv(monkeypatch):
    for mod in ["worker_app.celery_app", "worker_app.db"]:
        monkeypatch.setitem(sys.modules, mod, MagicMock())
    import celery

    monkeypatch.setattr(celery, "shared_task", lambda **kw: (lambda f: f))
    from worker_app.tasks import viral_video as wv

    return wv


# === 1. 新schema（v3.1+）：独立 <shot>/<visual>/<action_details> ===
NEW_SCHEMA_XML = """<script>
  <copy_display_markdown><![CDATA[来御众堂放松一下]]></copy_display_markdown>
  <clips>
    <clip index="1">
      <time_range>0-4秒</time_range>
      <voiceover>生完娃，是不是连照镜子都没勇气了？</voiceover>
      <shot>中近景平视，50mm f/2.0，缓慢推镜</shot>
      <visual>暖光浴室镜前，妈妈略显疲惫地看向镜中自己，镜灯打在侧脸</visual>
      <action_details>妈妈手扶洗手台，微微低头再抬眼看镜，眼眶微红</action_details>
      <reference_image_index>0</reference_image_index>
    </clip>
    <clip index="2">
      <time_range>4-10秒</time_range>
      <voiceover>推开门，暖光和一杯热茶先接住了你。</voiceover>
      <shot>中景平视，35mm f/2.8，缓慢平移</shot>
      <visual>门店前台暖光木色装修，茶水冒热气，背景柔和虚化</visual>
      <action_details>店员微笑双手递上茶杯，妈妈伸手接过，指尖轻触杯沿</action_details>
      <reference_image_index>1</reference_image_index>
    </clip>
  </clips>
  <voiceover_script>生完娃，是不是连照镜子都没勇气了？推开门，暖光和一杯热茶先接住了你。</voiceover_script>
  <theme>产后妈妈走进御众堂</theme>
</script>"""


def test_new_schema_xml_field_mapping(monkeypatch, mock_job):
    wv = _import_wv(monkeypatch)
    r = wv._parse_script(NEW_SCHEMA_XML, mock_job)
    assert r is not None
    assert len(r["shots"]) == 2
    s0 = r["shots"][0]
    # shot_type 应是纯摄影参数
    assert "缓慢推镜" in s0["shot_type_angle_movement"]
    assert "妈妈" not in s0["shot_type_angle_movement"]
    # scene_and_dialogue 只放对白
    assert s0["scene_and_dialogue"] == "生完娃，是不是连照镜子都没勇气了？"
    # action_details 有内容
    assert "手扶洗手台" in s0["action_details"]
    # visual 字段有画面
    assert "暖光" in s0["visual"]
    assert "推镜" not in s0["visual"]


# === 2. 旧schema（无<shot>/<action_details>标签，<visual>混写）：兼容拆分 ===
OLD_SCHEMA_XML = """<script>
  <clips>
    <clip index="1">
      <time_range>0-4秒</time_range>
      <voiceover>生完娃，是不是连照镜子都没勇气了？</voiceover>
      <visual>中近景平视，暖光下妈妈略显疲惫地看向镜中，镜头缓缓推近，妈妈手扶洗手台微微低头</visual>
      <reference_image_index>0</reference_image_index>
    </clip>
  </clips>
  <voiceover_script>生完娃，是不是连照镜子都没勇气了？</voiceover_script>
  <theme>测试</theme>
</script>"""


def test_old_schema_xml_splits_visual(monkeypatch, mock_job):
    wv = _import_wv(monkeypatch)
    r = wv._parse_script(OLD_SCHEMA_XML, mock_job)
    assert r is not None
    s0 = r["shots"][0]
    # 摄影参数段应含景别/运镜关键词
    assert "中近景" in s0["shot_type_angle_movement"] or "推" in s0["shot_type_angle_movement"]
    # action_details 应含动作动词
    assert s0["action_details"], "action_details不应为空（旧schema兼容应从<visual>拆出动作）"
    assert "手扶" in s0["action_details"] or "低头" in s0["action_details"]
    # scene_and_dialogue 只放对白，不重复画面
    assert s0["scene_and_dialogue"] == "生完娃，是不是连照镜子都没勇气了？"


# === 3. markdown 中文分镜字段 ===
MARKDOWN_INPUT = """## 视频主题：御众堂探店

### 镜头1
- 时间范围：0-4秒
- 景别/角度与运镜：中近景平视，缓慢推镜
- 画面：暖光下妈妈看向镜中，镜灯打在侧脸
- 动作与真人细节：妈妈手扶洗手台，微微低头
- 口播："生完娃，是不是连照镜子都没勇气了？"

### 镜头2
- 时间范围：4-10秒
- 景别：中景平视
- 运镜：缓慢平移
- 画面：门店前台暖光木色装修
- 动作：店员递茶，妈妈伸手接过
- 口播："推开门，暖光和一杯热茶先接住了你"
"""


def test_markdown_field_extraction(monkeypatch, mock_job):
    wv = _import_wv(monkeypatch)
    r = wv._parse_script(MARKDOWN_INPUT, mock_job)
    assert r is not None
    assert len(r["shots"]) == 2
    s0 = r["shots"][0]
    assert "中近景" in s0["shot_type_angle_movement"]
    assert "妈妈手扶洗手台" in s0["action_details"]
    assert s0["scene_and_dialogue"] == "生完娃，是不是连照镜子都没勇气了？"


# === 4. _split_visual_field 单元测试 ===
def test_split_visual_field(monkeypatch):
    wv = _import_wv(monkeypatch)
    # 典型旧schema混写："中近景平视，暖光下一位妈妈略显疲惫地看向镜中，镜头缓缓推近，妈妈手扶洗手台微微低头"
    cam, vis, act = wv._split_visual_field(
        "中近景平视，暖光下妈妈略显疲惫地看向镜中，镜头缓缓推近，妈妈手扶洗手台微微低头"
    )
    assert cam and "中近景" in cam
    # 动作/表情段应被拆到 act（含"妈妈"主语+动词）
    assert act and ("手扶" in act or "疲惫" in act or "看向" in act)
    # 画面描述允许为空（当动作段本身承载了主要画面信息时），或含"暖光"
    assert (not vis) or ("暖光" in vis) or ("妈妈" in vis)


def test_split_visual_short(monkeypatch):
    wv = _import_wv(monkeypatch)
    cam, vis, act = wv._split_visual_field("特写固定镜头")
    # 切不开时整体归camera/visual，action为空
    assert "特写" in cam


if __name__ == "__main__":
    pytest.main([__file__, "-v"])