Skip to content

armgit / peach

Public Imported from GitHub
0

Peach: Local-first video metadata scraper and media manager.

fix(metadata): 按原文快照复核 r18 标题候选

longmeidao6a32255f30e0
diff --git a/docs/REUSE.md b/docs/REUSE.md
index a42e2ddc3..2370b223b 100644
--- a/docs/REUSE.md
+++ b/docs/REUSE.md
@@ -1,5 +1,9 @@
# 复用清单
+## 元数据标题原文
+
+持久元数据候选复用 `metadata.extract_peach_fields` 的原文取值和 `metadata_auto_apply` 的统一写入入口。r18 的纯拉丁字母标题需由同一作品快照的原文侧精确印证;缺少快照或只有英文译文时不进入自动结算,人工批准入口同样拒收。来源原文中的拉丁字母片名保留,不按字符集自动翻译。无新增依赖。
+
## 浏览器错误的构建映射
前端复用官方 [`@sentry/vite-plugin` 11.8.0](https://github.com/getsentry/sentry-javascript-bundler-plugins)(MIT,维护中,Node ≥23.2,与项目 Node 24 和 Vite 8 兼容)生成确定性 Debug ID 与 hidden source maps;[`@sentry/cli` 3.8.0](https://github.com/getsentry/sentry-cli)(FSL-1.1-MIT)仅用于构建期本地定位和显式上传。两者精确固定在前端开发依赖,纳入现有 Dependabot;不进入 Python 运行时或独立包依赖。
diff --git a/src/peach/metadata.py b/src/peach/metadata.py
index c7991c9f1..8b8ed9eee 100644
--- a/src/peach/metadata.py
+++ b/src/peach/metadata.py
@@ -443,7 +443,7 @@ def extract_catalog_evidence(payload: dict) -> dict[str, dict]:
out: dict[str, dict] = {}
text_fields = {
"title": original_wording(payload, "title"),
- "original_title": payload.get("original_title"),
+ "original_title": original_wording(payload, "original_title"),
"director": japanese.get("director") or payload.get("director"),
"label": japanese.get("label") or payload.get("label"),
}
diff --git a/src/peach/metadata_auto_apply.py b/src/peach/metadata_auto_apply.py
index 6388e7d2c..e24edcff7 100644
--- a/src/peach/metadata_auto_apply.py
+++ b/src/peach/metadata_auto_apply.py
@@ -724,6 +724,31 @@ def _auto_apply_rule(candidate: dict, agreed: int) -> str:
return f"adr-0018-empty-field-single-{kind}-source"
+def _original_title_candidate(field: str, candidate: dict, code: str = '') -> bool:
+ """r18 拉丁字母标题须能在快照的原文侧精确复核,适用于持久候选重放。"""
+ if field not in {'title', 'original_title'} or candidate.get('source') != 'r18dev':
+ return True
+ value = ' '.join(str(candidate.get('value') or '').split())
+ if not re.search(r'[A-Za-z]', value) or re.search(r'[\u3040-\u30ff\u3400-\u9fff]', value):
+ return True
+ snapshot = candidate.get('raw_snapshot')
+ if not snapshot:
+ return False
+ try:
+ payload = json.loads(Path(snapshot).read_text(encoding='utf-8'))
+ if isinstance(payload, dict) and isinstance(payload.get('result'), dict):
+ payload = payload['result']
+ if not isinstance(payload, dict):
+ return False
+ from .metadata import extract_peach_fields, identifies_code
+ if code and not identifies_code(code, payload):
+ return False
+ original = extract_peach_fields(payload).get(field, {}).get('value')
+ return bool(original) and ' '.join(str(original).split()) == value
+ except (OSError, ValueError, TypeError):
+ return False
+
+
def _evidence_candidates(row: dict, field: str = "") -> list[dict]:
"""能当补空证据的候选:已登记来源与本地 NFO,去掉这个字段上被拉黑的来源。
@@ -736,7 +761,8 @@ def _evidence_candidates(row: dict, field: str = "") -> list[dict]:
return [c for c in row.get("candidates") or []
if (str(c.get("source") or "").strip() in SOURCE_SPECS
or str(c.get("source") or "").strip() == LOCAL_NFO_SOURCE)
- and not blacklisted(field, str(c.get("source") or "").strip())]
+ and not blacklisted(field, str(c.get("source") or "").strip())
+ and _original_title_candidate(field, c, str(row.get('code') or ''))]
def _unmasked_titles(field: str, candidates: list[dict], current: str = "") -> list[dict]:
@@ -1263,6 +1289,8 @@ def _apply_metadata_candidate(
marks = ",".join("?" * len(asset_ids))
if field in {"title", "original_title"}:
+ if not _original_title_candidate(field, candidate, code):
+ raise ValueError('r18 标题原文证据未取得,请重新采集资料')
raw_value = str(candidate.get("value") or "")
value = " ".join(raw_value.split())
if (not value or len(value) > 1000
diff --git a/tests/test_metadata_r18dev.py b/tests/test_metadata_r18dev.py
index 1213979c5..798afdefc 100644
--- a/tests/test_metadata_r18dev.py
+++ b/tests/test_metadata_r18dev.py
@@ -1,5 +1,6 @@
"""r18.dev 套进站点解析器契约:作品 JSON 的解析、combined 页的日文补全、payload 逐键钉住,以及失败分档。"""
import json
+import tempfile
import unittest
from unittest.mock import Mock, patch
@@ -206,6 +207,28 @@ class OriginalWordingTests(unittest.TestCase):
mixed = dict(self.ENGLISH, actresses=[{"japanese_name": "胡桃さくら"}, {"japanese_name": "Remu Suzumori"}])
self.assertEqual([row["name"] for row in extract_peach_fields(mixed)["performers"]["value"]], ["胡桃さくら"])
+ def test_an_english_original_title_in_an_r18_snapshot_is_source_only(self):
+ fields = extract_peach_fields(dict(self.ENGLISH, original_title='English alternative'))
+ self.assertNotIn('original_title', fields)
+
+ def test_latin_title_candidates_require_same_work_original_language_evidence(self):
+ from pathlib import Path
+ from peach.metadata_auto_apply import _original_title_candidate, _evidence_candidates
+ candidate = {'source':'r18dev', 'value':'BEAUTY VENUS 5', 'display_value':'BEAUTY VENUS 5'}
+ self.assertFalse(_original_title_candidate('title', candidate, 'IPX-219'))
+ self.assertEqual(_evidence_candidates({'code':'IPX-219','candidates':[candidate]}, 'title'), [])
+ with tempfile.TemporaryDirectory() as directory:
+ path = Path(directory) / 'source.json'
+ candidate['raw_snapshot'] = str(path)
+ payload = {'id':'IPX-219', 'source_url':'https://r18.dev/detail', 'title':'BEAUTY VENUS 5'}
+ path.write_text(json.dumps(payload),encoding='utf-8')
+ self.assertFalse(_original_title_candidate('title', candidate, 'IPX-219'))
+ payload['translations'] = [{'language':'ja','title':'BEAUTY VENUS 5'}]
+ path.write_text(json.dumps(payload),encoding='utf-8')
+ self.assertTrue(_original_title_candidate('title', candidate, 'IPX-219'))
+ self.assertFalse(_original_title_candidate('title', candidate, 'IPX-218'))
+ self.assertTrue(_original_title_candidate('title', dict(candidate,source='makers'), 'IPX-219'))
+
def test_a_top_level_value_written_in_japanese_script_stays(self):
payload = dict(self.ENGLISH, series="いつでも使えるオナホ後輩",
translations=[{"language": "ja", "series": ""}])
diff --git a/tests/test_web_review.py b/tests/test_web_review.py
index 47b71b22e..19f21d823 100644
--- a/tests/test_web_review.py
+++ b/tests/test_web_review.py
@@ -793,7 +793,7 @@ class ReviewQueueTests(unittest.TestCase):
{"item_key": "ABW", "field": "title", "current": "",
"candidates": ["涼森れむ流 HOW TO SEX!!"], "code": "ABW-358", "source": "local_nfo"},
{"item_key": "ABX", "field": "title", "current": "",
- "candidates": ["涼森れむ流", {"value": "Remu Style", "source": "r18dev"}],
+ "candidates": ["涼森れむ流", {"value": "涼森れむスタイル", "source": "r18dev"}],
"code": "ABW-359", "source": "local_nfo"},
])
self.assertEqual(self._auto()["applied"], 2)
@@ -814,7 +814,7 @@ class ReviewQueueTests(unittest.TestCase):
# 被压下的说法要留痕:事后得答得出当时还有哪个值、为什么没选它。
self.assertEqual(json.loads(notes["ABX"])["rule"], "adr-0038-chain-local_nfo")
self.assertEqual(json.loads(notes["ABX"])["overruled"],
- [{"source": "r18dev", "value": "Remu Style"}])
+ [{"source": "r18dev", "value": "涼森れむスタイル"}])
self.assertEqual(self.queue_keys("metadata_fields"), [])
def test_the_chain_settles_a_disagreement_between_two_official_sources(self):
@@ -2736,6 +2736,21 @@ class ReviewQueueTests(unittest.TestCase):
).fetchone()[0], "正式作品标题")
con.close()
+ def test_r18_english_title_without_original_evidence_is_not_written(self):
+ self._asset(890, 'PCH-890', 'PCH-890.mp4')
+ candidate = {'candidate_key':'PCH-890:title:r18dev:old', 'source':'r18dev',
+ 'source_url':'https://r18.dev/example','provider_id':'PCH-890',
+ 'confidence':0.9,'value':'English translated title','display_value':'English translated title'}
+ self.write_metadata_candidates([{'item_key':'PCH-890:title','code':'PCH-890','query':'PCH-890',
+ 'field':'title','field_label':'标题','current_value':'','candidates_json':json.dumps([candidate]),
+ 'source_count':'1','status':'candidate','size_gb':'1','videos':'1','fetched_at':'now'}])
+ self.assertEqual(self._auto()['applied'], 0)
+ with self.assertRaisesRegex(ValueError, '原文证据未取得'):
+ rm_review.w_review_decision(self.contract, {'category':'metadata_fields','item_key':'PCH-890:title',
+ 'candidate_key':candidate['candidate_key'],'status':'approved'})
+ with contextlib.closing(sqlite3.connect(self.db_path)) as connection:
+ self.assertFalse(connection.execute('SELECT catalog_title FROM asset WHERE id=890').fetchone()[0])
+
def test_metadata_approval_rejects_repeated_name_even_if_csv_is_tampered(self):
con = sqlite3.connect(self.db_path)
con.execute("UPDATE asset SET code='ABC-001' WHERE id=1")