Peach: Local-first video metadata scraper and media manager.
"""用作品番号回查 javbus 的製作商,把罗马音厂牌名换回日文原名。
用户 2026-09-02 指出 `Celeb no Tomo` 其实是 `セレブの友`。这类罗马音是番号站的转写,
不是厂牌自己的写法,账本不该留着它。判据也由用户当场定下:
- 製作商含汉字或平假名 → 采用日文原名(`セレブの友`、`痴女ヘブン`、`アロマ企画`)。
- 製作商是纯片假名 → 默认当英文品牌的外来语写法,保留账本里的英文原名。
`ムーディーズ` 不该顶掉 `MOODYZ`,`プレステージ` 不该顶掉 `Prestige`。这只是默认值:
拉丁名若只是这串片假名的罗马音(`Akinori` 对 `アキノリ`),厂牌自己并不这么写,
复核时按官网或官方 X 的自称改判(用户 2026-09-23:`PREMIUM` 用英文,`Celeb no Tomo` 用日文)。
- 账本现名本来就含日文 → 不动。
**日文名是查出来的,不是转写出来的。** 罗马音回日文没有唯一解(`Hon Naka` 可以是
`本中` 也可以是`ほんなか`),所以这里不做音译:拿厂牌自己作品的番号去 javbus 读
`製作商` 字段。一个厂牌尽量取两个不同前缀的番号,两页说法一致才敢提改名,
不一致就标出来交给人看——同一个账本厂牌名底下混了两家片商,本身就是要人处理的问题。
javbus 有年龄门,带 `age=verified` cookie 才给内容页,否则只回一张 21 KB 确认页。
番号页走代理。
只产出复核 CSV,不写账本:改厂牌名要同时动 `entity.canonical_name` 与 `asset.studio`,
那是另一次授权的事。
"""
from __future__ import annotations
import argparse
import re
import sqlite3
import sys
from collections import Counter
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "src"))
from peach.config import STATE_DIR # noqa: E402
from peach.jobs import job_main # noqa: E402
from peach.page_cache import HttpStatusError, Site # noqa: E402
from peach.review_csv import write_rows # noqa: E402
FIELDS = ("entity_id", "studio", "assets", "codes", "source_url", "maker",
"shape", "verdict", "proposed", "evidence")
JAVBUS = "https://www.javbus.com/{}"
JAVBUS_COOKIES = {"age": "verified", "dv": "1", "existmag": "all"}
#: 番号页上「製作商」标签之后的第一个链接文本就是片商名。
MAKER = re.compile(r"製作商.{0,200}?<a[^>]*>([^<]{1,60})</a>", re.S)
JAV_CODE = re.compile(r"^([A-Za-z]{2,6})-(\d{2,5})$")
KANA_KANJI = re.compile(r"[-ゟ一-鿿]") # 平假名或汉字
KATAKANA = re.compile(r"[゠-ヿ]")
RENAME, KEEP_KATAKANA, KEEP_LATIN, KEEP_JP, SPLIT, UNKNOWN, SKIP = (
"改名", "保留(片假名外来语)", "保留(来源也是拉丁)", "保留(已是日文)",
"不一致", "未取得", "不适用(非番号体系)")
# ---------------------------------------------------------------- 判定
def shape(name: str) -> str:
"""按用户的判据给名字分类:只有「汉字假名」这一类才值得顶掉英文写法。"""
if KANA_KANJI.search(name):
return "汉字假名"
if KATAKANA.search(name):
return "片假名"
return "拉丁"
def decide(current: str, makers: list[str]) -> tuple[str, str, str]:
"""返回 (判词, 建议名, 证据说明)。makers 是各番号页读到的製作商,按出现顺序。"""
if shape(current) != "拉丁":
return KEEP_JP, "", "账本现名已含日文"
found = [m for m in dict.fromkeys(makers) if m]
if not found:
return UNKNOWN, "", "番号页没读到製作商"
if len(found) > 1:
return SPLIT, "", "多个番号给出不同製作商:" + "/".join(found)
maker = found[0]
corroboration = f"{len(makers)} 个番号一致" if len(makers) > 1 else "只有 1 个番号可查"
if shape(maker) == "片假名":
return KEEP_KATAKANA, "", f"製作商 {maker} 是纯片假名,按规则保留英文原名({corroboration})"
if maker == current:
return KEEP_JP, "", f"製作商与现名相同({corroboration})"
if shape(maker) == "拉丁":
# 来源自己也写拉丁,两边差的只是空格或符号,那不是「去罗马音」,别拿它改名。
return KEEP_LATIN, "", f"製作商 {maker} 也是拉丁写法,与现名只差写法({corroboration})"
return RENAME, maker, f"製作商 {maker}({corroboration})"
# ---------------------------------------------------------------- 取数
def studios(connection: sqlite3.Connection, minimum: int) -> list[sqlite3.Row]:
return connection.execute(
"""
SELECT * FROM (
SELECT e.id AS id, e.canonical_name AS canonical_name,
(SELECT COUNT(*) FROM asset a WHERE a.studio = e.canonical_name) AS assets
FROM entity e WHERE e.kind = 'studio'
) WHERE assets >= ? ORDER BY assets DESC, canonical_name
""",
(minimum,),
).fetchall()
def code_groups(connection: sqlite3.Connection, studio: str, wanted: int,
depth: int = 3) -> list[list[str]]:
"""给一个厂牌挑几组番号:一个前缀一组,最多 `wanted` 组,组内最多 `depth` 个。
只有跨前缀才算互相印证——同前缀的两页必然出自同一家,证明不了什么。组内的第二、
第三个番号不参与印证,只用来顶替 404:番号页不存在是那一页的事,不代表这家查不到。
"""
rows = connection.execute(
"SELECT code, COUNT(*) n FROM asset WHERE studio = ? AND code IS NOT NULL AND code <> ''"
" GROUP BY code ORDER BY n DESC, code",
(studio,),
).fetchall()
groups: dict[str, list[str]] = {}
for row in rows:
match = JAV_CODE.match(str(row["code"]).strip())
if not match:
continue
prefix = match.group(1).upper()
if prefix not in groups and len(groups) >= wanted:
continue
bucket = groups.setdefault(prefix, [])
if len(bucket) < depth:
bucket.append(match.group(0).upper())
return list(groups.values())
def maker_of(page: str) -> str:
match = MAKER.search(page)
return match.group(1).strip() if match else ""
def inspect(site: Site, groups: list[list[str]]) -> tuple[list[str], list[str], list[str], list[str]]:
"""每组取到一个製作商就停。返回 (製作商, 证据页 URL, 实际打过的番号, 失败说明)。"""
makers, urls, tried, notes = [], [], [], []
for group in groups:
for code in group:
url = JAVBUS.format(code)
tried.append(code)
try:
page = site.get(url)
except (HttpStatusError, OSError, ValueError) as exc:
notes.append(f"{code}: {type(exc).__name__}: {exc}")
continue
maker = maker_of(page)
if not maker:
notes.append(f"{code}: 页面无製作商字段")
continue
urls.append(url)
makers.append(maker)
break
return makers, urls, tried, notes
# ---------------------------------------------------------------- 入口
def build_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
parser.add_argument("--database", type=Path, required=True)
parser.add_argument("--output", type=Path, required=True)
parser.add_argument("--min-assets", type=int, default=1)
parser.add_argument("--codes", type=int, default=2, help="每个厂牌最多查几个不同前缀的番号")
parser.add_argument("--depth", type=int, default=3, help="同前缀最多试几个番号(只为顶替 404)")
parser.add_argument("--limit", type=int, default=0)
parser.add_argument("--interval", type=float, default=1.5)
parser.add_argument("--timeout", type=float, default=25.0)
parser.add_argument("--refresh", action="store_true", help="忽略缓存重抓")
parser.add_argument("--cache-dir", type=Path, default=STATE_DIR / "studio-names")
parser.add_argument("--lock", type=Path, default=STATE_DIR / ".studio-names.lock")
return parser
def run(args: argparse.Namespace, site: Site | None = None) -> int:
connection = sqlite3.connect(f"file:{args.database}?mode=ro", uri=True)
connection.row_factory = sqlite3.Row
owned = site is None
if owned:
site = Site(args.cache_dir, args.interval, args.timeout,
refresh=args.refresh, via_proxy=True, cookies=JAVBUS_COOKIES)
rows, stats = [], Counter()
try:
targets = studios(connection, args.min_assets)
if args.limit:
targets = targets[:args.limit]
for entity in targets:
name = entity["canonical_name"]
if shape(name) != "拉丁":
stats[KEEP_JP] += 1
continue
groups = code_groups(connection, name, args.codes, args.depth)
if not groups:
# Blacked、Tushy、FC2-PPV 这些压根不走番号体系,回查无从谈起。
# 写「未取得」等于把「不适用」伪装成取证失败。
makers, urls, tried, notes = [], [], [], []
verdict, proposed, evidence = SKIP, "", "账本里没有 JAV 番号,不走番号体系"
else:
makers, urls, tried, notes = inspect(site, groups)
verdict, proposed, evidence = decide(name, makers)
stats[verdict] += 1
rows.append({
"entity_id": entity["id"], "studio": name, "assets": entity["assets"],
"codes": " ".join(tried), "source_url": " ".join(urls),
"maker": "/".join(dict.fromkeys(makers)), "shape": shape(name),
"verdict": verdict, "proposed": proposed,
"evidence": ";".join([evidence, *notes]),
})
finally:
if owned:
site.close()
connection.close()
order = {RENAME: 0, SPLIT: 1, UNKNOWN: 2, KEEP_KATAKANA: 3, KEEP_LATIN: 4,
KEEP_JP: 5, SKIP: 6}
rows.sort(key=lambda item: (order.get(item["verdict"], 9), -int(item["assets"]), item["studio"]))
write_rows(args.output, FIELDS, rows)
print({"复核行": len(rows), **stats, "取页": site.fetched, "命中缓存": site.cached,
"output": str(args.output)})
return 0
if __name__ == "__main__":
sys.stdout.reconfigure(encoding="utf-8", errors="replace")
raise SystemExit(job_main(build_parser, run))