Create-knowledge-find-decode/tests/test_classification_coarse.py

from __future__ import annotations

import acquisition.classification.coarse as coarse
import acquisition.classify as legacy_classify
from core.config import PgConfig, Settings
from core.text_limits import CLASSIFY_BODY_MAX_CHARS


def _settings() -> Settings:
    return Settings(
        pg=PgConfig(host="h", port=5432, user="u", password="p", database="d"),
        aiddit_crawler_base_url="http://crawler.test",
        crawler_timeout=30,
        openrouter_timeout_seconds=90,
        openrouter_model="m",
        openrouter_base_url="http://openrouter.test",
        openrouter_api_key="k",
        llm_model="m",
        max_cards=12,
        frames_dir="f",
        douyin_ratio="540p",
        data_dir="data",
    )


def test_formal_classify_imgtext_passes_http_image_urls(monkeypatch):
    captured = {}

    def fake_judge(messages, settings, timeout):
        captured["messages"] = messages
        return 1, "ok", "先定受众", ""

    monkeypatch.setattr(coarse, "_judge", fake_judge)

    result = coarse.classify_imgtext(
        {
            "platform": "xiaohongshu",
            "title": "脚本创作",
            "body_text": "讲怎么设计开头",
            "images": ["https://cdn.test/a.jpg"],
        },
        _settings(),
    )

    assert result[0] == 1
    user_content = captured["messages"][1]["content"]
    assert {"type": "image_url", "image_url": {"url": "https://cdn.test/a.jpg"}} in user_content


def test_formal_classify_imgtext_uses_wide_body_limit(monkeypatch):
    captured = {}

    def fake_judge(messages, settings, timeout):
        captured["messages"] = messages
        return 1, "ok", "长正文方法", ""

    monkeypatch.setattr(coarse, "_judge", fake_judge)
    long_body = "甲" * (CLASSIFY_BODY_MAX_CHARS + 7)

    coarse.classify_imgtext(
        {
            "platform": "xiaohongshu",
            "title": "长文",
            "body_text": long_body,
            "images": [],
        },
        _settings(),
    )

    user_text = captured["messages"][1]["content"][0]["text"]
    assert user_text.count("甲") == CLASSIFY_BODY_MAX_CHARS


def test_legacy_classify_imgtext_also_passes_http_image_urls(monkeypatch):
    captured = {}

    def fake_judge(messages, settings, timeout):
        captured["messages"] = messages
        return 1, "ok", "先定受众", ""

    monkeypatch.setattr(legacy_classify, "_judge", fake_judge)

    result = legacy_classify.classify_imgtext(
        {
            "platform": "weixin",
            "title": "公众号选题",
            "body_text": "讲怎么做标题",
            "images": ["https://cdn.test/w.jpg"],
        },
        _settings(),
    )

    assert result[0] == 1
    user_content = captured["messages"][1]["content"]
    assert {"type": "image_url", "image_url": {"url": "https://cdn.test/w.jpg"}} in user_content


def test_legacy_classify_video_includes_title_and_body_text(monkeypatch):
    captured = {}

    def fake_judge(messages, settings, timeout):
        captured["messages"] = messages
        return 1, "ok", "视频方法", ""

    monkeypatch.setattr(legacy_classify, "_judge", fake_judge)

    legacy_classify.classify_video(
        {
            "platform": "douyin",
            "title": "视频标题",
            "body_text": "视频文案",
            "video": "https://cdn.test/video.mp4",
        },
        _settings(),
    )

    user_text = captured["messages"][1]["content"][0]["text"]
    assert "平台：douyin" in user_text
    assert "标题：视频标题" in user_text
    assert "正文/文案：视频文案" in user_text


def test_coarse_classify_item_maps_judge_result(monkeypatch):
    monkeypatch.setattr(
        coarse,
        "classify_imgtext",
        lambda payload, settings: (0, "不是创作知识", "", ""),
    )

    result = coarse.coarse_classify_item(
        platform="weixin",
        title="普通知识",
        body_text="只是作品介绍",
        image_urls=["https://cdn.test/a.jpg"],
        settings=_settings(),
    )

    assert result.is_creation_knowledge is False
    assert result.label == "not_creation"
    assert result.status == "classified"


def test_coarse_classify_item_routes_douyin_image_post_to_imgtext(monkeypatch):
    called = {}

    def fake_imgtext(payload, settings):
        called["imgtext"] = payload
        return 1, "是创作知识", "图片方法", ""

    def fake_video(payload, settings):
        called["video"] = payload
        return 1, "不应调用", "", ""

    monkeypatch.setattr(coarse, "classify_imgtext", fake_imgtext)
    monkeypatch.setattr(coarse, "classify_video", fake_video)

    result = coarse.coarse_classify_item(
        platform="douyin",
        content_mode="image_post",
        title="图片帖",
        body_text="图片正文",
        image_urls=["https://cdn.test/dy.jpg"],
        settings=_settings(),
    )

    assert result.is_creation_knowledge is True
    assert "imgtext" in called
    assert "video" not in called


def test_coarse_classify_item_skips_unsupported_and_missing_video():
    unsupported = coarse.coarse_classify_item(
        platform="douyin",
        content_mode="unsupported",
        settings=_settings(),
    )
    missing_video = coarse.coarse_classify_item(
        platform="douyin",
        content_mode="video_post",
        settings=_settings(),
    )

    assert unsupported.status == "skipped"
    assert unsupported.label == "unsupported_content_mode"
    assert missing_video.status == "skipped"
    assert missing_video.label == "video_missing"