Peach: Local-first video metadata scraper and media manager.
fix(junk): 云下载包里的附带文件按来源事实达到自动隔离
longmeidaof99edfff2b0d
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 16289df85..6e4049870 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -38,6 +38,7 @@ Removed / Fixed / Security。- **采集**:一次勾选上万部视频补资料或探测时长,不再因数据库参数超限报错。- **配置**:卸载被在跑的任务拦下时,提示写出是哪个任务,可以去任务中心等它完成或取消。- **馆藏**:影片目录改名又换了编码后,扫描会沿用原来那条记录,标签、封面和观看记录都保留,不再多出一条没有番号的重复条目;同番号的旧记录有多条时仍交给资源同步处理。+- **馆藏**:云下载留在落地目录里的推广短视频、网址文本和二维码图片会自动移入回收站(可以还原),不再混进馆藏;带本片番号的封面、截图和分卷照常保留,附带的压缩包进入垃圾复核。逐字加空格的推广名(如 `x u u 6 2 . c o m`)和同一份附件被塞进多个发行包的情况也能认出。## [0.53.0] - 2026-11-02diff --git a/docs/OPERATIONS.md b/docs/OPERATIONS.mdindex e14f8743c..dd8982531 100644--- a/docs/OPERATIONS.md+++ b/docs/OPERATIONS.md@@ -641,6 +641,7 @@ PikPak 官方核验通道使用「推送发现」中该媒体根对应的唯一- 点「不是垃圾」会记下这个文件的判断(可以撤销),它会出现在「已排除」视图里。这个判断只针对这一个文件,不会把整个域名或来源列入白名单。- 自动隔离要求达到自动线,并保护人工保留、喜欢、观看或播放列表等个人记录、已配对字幕、文件自身的作品番号和本次绑定的下载正片。写入前再次核对当前资产;已离开馆藏的条目不进入后续采集或整理。- 无关键词的极短附件可凭同目录长正片、数量及跨目录测量值进入复核;这类结构证据与「域名加序号」命名不能单独触发自动隔离。还原后确认要保留的文件,可同时标记「不是垃圾」。+- 云下载落地目录里,正片以外、名字不带本片番号的文件按下载包附带文件计分:网址文本、图片等可直接达到自动线;短视频须有测量值,且正片远长于它;压缩包只进复核。带本片番号的封面、截图、分卷和 `poster`、`fanart` 这类作品图不算附件。同名同体积的小文件出现在三个以上带番号的发行包目录时,也计为垃圾证据。维护脚本 `scripts/trash_junk.py` 默认生成候选清单;永久删除使用 `--purge --review-csv`,复核 CSV 每行包含 `id`、`path`、`size`、`decision`、`confidence`、`why`。diff --git a/src/peach/junk.py b/src/peach/junk.pyindex da5eb3f50..005135b3b 100644--- a/src/peach/junk.py+++ b/src/peach/junk.py@@ -7,7 +7,9 @@ import timefrom functools import cachefrom pathlib import PureWindowsPath-from .catalog_rules import is_jav_code, promo_free_key, release_code_from_filename+from .catalog_rules import (+ _jav_code_pattern, is_jav_code, promo_free_key, release_code_from_filename, without_store_prefix,+)from .media_sources import SCANNABLE_LOCATIONSfrom .settings_file import PROJECT_ROOTfrom .personal_records import record_holders@@ -17,7 +19,8 @@ from .repository import LedgerDatabase# 实测正片标题里就有(「还要微信跟老公汇报战果」是剧情,不是联系方式)。PROMO_PHRASE = re.compile(r"(扫码|扫一扫|掃一掃|扫描|掃描|QR ?CODE|二维码|二維碼|加微信|加微|威信\d|微信号|"- r"微信\s*[::]|免费看|免费玩|福利群|最新地址|"+ r"微信\s*[::]|免费看|免费玩|福利群|最新(?:地址|位址|網址|网址)|"+ r"(?:地址|位址|網址|网址)(?:發佈|发布|獲取|获取)|收藏不迷路|"r"永久(?:域名|地址|发布)|地址发布器|防屏蔽|资源获取|点击(?:观看|下载|进入)|下载APP|下载|签到|代币|领取|"r"强力推荐|国产大片|在线视频|線上影片每天火熱更新|手机看片|手機影城|"r"大饱眼福|房间火爆|澳门|赌场|博彩|棋牌|加我|"@@ -34,9 +37,10 @@ PROMO_FILLER = re.compile(r"(免费\d*禁|免费|免費|手游|手遊|成人遊?戲|成人游戏|下载安装|安装|安裝|访问|訪問)",re.I)# 结尾不能用 \b:`uuc82.com_2` 里 `m` 和 `_` 都是词字符,构不成边界,域名会漏掉。+#: `manko.fun`、`58m.co` 是云下载包里的推广附件(2026-11-03 只读盘点)。PROMO_DOMAIN = re.compile(r"(?:https?://)?(?:www\.)?[\w-]{2,}\.(?:com|net|me|la|xyz|cc|tv|top|vip|club|"- r"info|org|pw|cn|app|site|online|shop)(?![a-z0-9])", re.I)+ r"info|org|pw|cn|app|site|online|shop|fun|live|co)(?![a-z0-9])", re.I)# 真番号带厂牌前缀和连字号(ABW-153、259LUXU-1141)。RAIKUN325 这类没有连字号的# 是被误填进 code 的创作者账号名,不能拿来做「同番号有完整版」的比较,# 判据与 `is_jav_code` 同源:分隔符正是番号与账号名的唯一线索。@@ -46,7 +50,7 @@ PART_MARK = re.compile(r"(CD\d|part\d|分卷|-\d{1,2}$|\(\d+\)$)", re.I)# 创作者位是旧导入器的目录名投影,`bbsxv.xyz-DOCP-324` 这类广告包会直接落在那里;# 裸域名目录(98T.la@账号、huachishe.com@系列)是转载水印,不是广告,不能进判据。AD_DOMAIN = re.compile(- r"\b[0-9a-z][-0-9a-z]{1,20}\.(?:cc|xyz|xxx|com|net|la|me|top|vip|club|app|cn|pw|tv|gg)\b", re.I)+ r"\b[0-9a-z][-0-9a-z]{1,20}\.(?:cc|xyz|xxx|com|net|la|me|top|vip|club|app|cn|pw|tv|gg|fun|live)\b", re.I)AD_DIRPACK = re.compile(r"[0-9a-z][-0-9a-z]{1,20}\.[a-z]{2,10}[ \-_]+\[?[A-Za-z]{2,6}-?\d{2,5}", re.I)#: 同一目录里有这么多资产,它就是成套下载的资源包,不是塞进别人目录的插页。@@ -131,6 +135,30 @@ NUMBERED_SOURCE_VIDEO = re.compile(r"^(?:\[[^\]]+\]|【[^】]+】)[\s._-]*\d+(?:PAGE_COMPONENT_SUFFIXES = frozenset({".js", ".css", ".aspx"})PAGE_CHROME_IMAGE = re.compile(r"^(?:banner(?:_[bs])?|1000x90(?:yunding)?|count\d*|logo|lan\d*|you|zuo|zuoxia)$", re.I)PAGE_PROMO_CLAIM = re.compile(r"(?:记住|記住).{0,12}(?:网址|網址)|(?:网址|網址).{0,8}(?:发布|發布)")+#: 云下载包里的附带视频按这个时长以内、且正片至少长这么多倍认定。+#:+#: 2026-11-03 只读盘点:364 个已入库下载任务留下 142 个附属文件,推广视频全在 27~157 秒+#: (`人间尤物.mp4` 61 秒、`manko.fun.mp4` 157 秒),正片都是一两个小时。+DOWNLOAD_ATTACHMENT_SECONDS = 240+DOWNLOAD_MAIN_RATIO = 20+#: 媒体中心约定的作品图名,下载包自带时属于本片资料,不当附件。+ARTWORK_NAME = re.compile(+ r"^(?:poster|fanart|thumb|cover|folder|landscape|backdrop|clearlogo|clearart)(?:[-_ ]?\d+)?$", re.I)+#: 同名同体积的文件散布在这么多个目录,就是被成批塞进不同发行包的同一份附件。+REPEATED_COPY_FOLDERS = 3+REPEATED_COPY_MAX_BYTES = 16 * 1024**2+++def despaced(name: str) -> str:+ """把逐字加空格的推广名(`x u u 6 2 . c o m`、`最 新 位 址 獲 取`)合回去。++ 这种写法专门用来躲关键词与域名匹配。七成以上的分段只有一两个字符才合并,+ 正常标题里的空格分隔的是词,不会这样碎。+ """+ tokens = str(name or "").split()+ if len(tokens) >= 5 and sum(len(token) <= 2 for token in tokens) >= len(tokens) * 0.7:+ return "".join(tokens)+ return str(name or "")def promo_residue(name: str) -> int:@@ -383,6 +411,124 @@ def _short_attachment_reasons(connection, rows, *, roots=None) -> dict[int, str]return reasons+def names_release(stem: str, code: str | None) -> bool:+ """名字里带着这部作品的番号:`hjd2048.com-0907mide580-h264`、`tkmj001jp`、`kpkp3.com_483SGKX-016B`。"""+ for variant in dict.fromkeys((code, without_store_prefix(code))):+ pattern = _jav_code_pattern(variant)+ if pattern and re.search(rf"(?<![A-Z]){pattern}(?!\d)", str(stem or ""), re.I):+ return True+ return False+++def _download_folders(connection) -> dict[str, tuple[str, int, float]]:+ """已入库且绑定了正片的云下载目录 → (番号,正片 id,正片时长)。++ 目录至少在下载根之下一层:这个目录是这一个种子落地的地方,不是共用的下载根。+ """+ folders = {}+ for path, code, main, duration in connection.execute(+ "SELECT t.ledger_path,t.code,t.asset_id,a.duration FROM download_task t "+ "LEFT JOIN asset a ON a.id=t.asset_id WHERE t.state='ingested' "+ "AND t.asset_id IS NOT NULL AND t.code IS NOT NULL AND t.ledger_path IS NOT NULL"):+ folder = PureWindowsPath(path)+ if len(folder.parts) >= 3:+ folders[str(folder).casefold()] = (code, main, duration or 0)+ return folders+++def _download_attachment_reasons(connection, rows) -> dict[int, tuple[str, bool]]:+ """云下载包里正片以外、名字不带本片番号的附带文件 → (理由,是否够自动隔离)。++ 这是来源事实而不是名字猜测:下载任务记着这包是为哪部作品下的、哪一条是正片,+ 整理把正片移走后留在落地目录里的推广视频、网址文本、二维码图都与这部作品无关。+ 带本片番号的封面、截图和分卷,以及媒体中心作品图名不算附件。压缩包可能装着+ 正片的资料,只进复核;视频要有测量值、足够短,且正片远长于它。+ """+ folders = _download_folders(connection)+ reasons = {}+ if not folders:+ return reasons+ for row in rows:+ path = PureWindowsPath(row["path"] or "")+ task = next((folders[key] for key in (str(parent).casefold() for parent in path.parents)+ if key in folders), None)+ if task is None or row["id"] == task[1]:+ continue+ code, _main, main_duration = task+ if names_release(path.stem, code) or ARTWORK_NAME.fullmatch(path.stem):+ continue+ if row["medium"] == "video":+ duration = row["duration"] or 0+ if 0 < duration < DOWNLOAD_ATTACHMENT_SECONDS and main_duration >= duration * DOWNLOAD_MAIN_RATIO:+ reasons[row["id"]] = (f"{code} 下载包里的附带短视频,正片长 {main_duration / 60:.0f} 分", True)+ elif row["medium"] == "archive":+ reasons[row["id"]] = (f"{code} 下载包里的附带压缩包,须核验内容", False)+ else:+ reasons[row["id"]] = (f"{code} 下载包里的附带文件,名字不带本片番号", True)+ return reasons+++def _repeated_copies(connection, *, roots=None) -> dict[tuple[str, int], int]:+ """同名同体积的小文件各出现在多少个发行包目录;只统计到达门槛的组合。++ 推广附件是同一份文件被成批塞进不同发行包:`manko.fun.mp4`(11 MB、157 秒)出现在+ 10 个下载目录,`最 新 位 址 獲 取.txt`(136 字节)出现在 7 个。名字和体积都一致是+ 文件层面的事实,和只比测量值的极短附件扩散(`_short_attachment_reasons`)不同。+ 只数目录名自己带番号的发行包:创作者目录里每期都带的同一段片头不是被塞进来的。+ """+ marks = ",".join("?" for _ in SCANNABLE_LOCATIONS)+ scope, parameters = _scope_clause(roots=roots)+ folders: dict[tuple[str, int], set[str]] = {}+ releases: dict[str, bool] = {}+ for path, name, size in connection.execute(+ f"SELECT path,name,size FROM asset WHERE disposal IS NULL AND location IN ({marks}) "+ "AND size>0 AND size<? AND path IS NOT NULL "+ "AND (COALESCE(medium,'other')<>'video' OR duration IS NULL OR duration<=1200)" + scope,+ (*SCANNABLE_LOCATIONS, JUNK_VIDEO_MAX_BYTES, *parameters)):+ folder = str(path).rpartition("\\")[0]+ if folder not in releases:+ releases[folder] = bool(release_code_from_filename(folder.rpartition("\\")[2]))+ if not releases[folder]:+ continue+ key = (str(name or "").casefold(), int(size))+ folders.setdefault(key, set()).add(folder.casefold())+ return {key: len(found) for key, found in folders.items() if len(found) >= REPEATED_COPY_FOLDERS}+++def _repeated_copy_reason(row, copies: dict[tuple[str, int], int]) -> str:+ """散布在多个目录的同一份小附件;带作品番号的文件与作品图名不算。"""+ item = PureWindowsPath(row["name"] or row["path"] or "")+ size = int(row["size"] or 0)+ count = copies.get((item.name.casefold(), size), 0)+ if (not count or (row["medium"] != "video" and size > REPEATED_COPY_MAX_BYTES)+ or ARTWORK_NAME.fullmatch(item.stem) or release_code_from_filename(item.stem)):+ return ""+ return f"同名同体积的副本散布在 {count} 个目录"+++def _attachment_evidence(row, suffix: str, name: str, download_attachments: dict,+ copies: dict) -> tuple[int, int, list[str]]:+ """附件证据:词库里的固定命名、下载包附带文件、同一份文件散布在多个发行包。++ 返回(其中只够复核的分数,分数,理由)。固定命名只够复核;下载包附带的视频与散布+ 副本只给 45 分,要再叠上时长与体积证据才到自动线;下载包里的压缩包只进复核。+ """+ score, review_only, why = 0, 0, []+ named = _promo_name_reason(suffix, row.get("path") or name, row.get("size") or 0)+ if named:+ score += JUNK_REVIEW_SCORE; review_only += JUNK_REVIEW_SCORE; why.append(named)+ reason, automatic = download_attachments.get(row["id"], ("", False))+ if reason:+ points = 45 if row.get("medium") == "video" else AUTO_TRASH_SCORE if automatic else JUNK_REVIEW_SCORE+ score += points; why.append(reason)+ if not automatic:+ review_only += points+ copied = _repeated_copy_reason(row, copies)+ if copied:+ score += 45; why.append(copied)+ return review_only, score, why++def _video_evidence(row, stem: str, longer: dict, promo_dir: bool,directory_points: int, attachment_reason: str = "") -> tuple[int, int, list[str]]:"""视频自己的证据:同番号完整版、时长、体积与推广目录。@@ -424,7 +570,7 @@ def _judgement_inputs(database, *, asset_ids, roots):with database.read_connection() as c:rows = _junk_rows(c, asset_ids=asset_ids, roots=roots)if not rows:- return [], {}, frozenset(), {}, {}, {}+ return [], {}, frozenset(), {}, {}, {}, {}, {}# 同番号是否存在明显更长的版本;只在 code 是真番号时才有意义。longer = {r[0]: r[1] for r in c.execute("SELECT code, max(duration) FROM asset WHERE medium='video' AND code IS NOT NULL "@@ -439,13 +585,16 @@ def _judgement_inputs(database, *, asset_ids, roots):folder_assets, folder_videos = _folder_index(c, roots=context)context_rows = _junk_rows(c, roots=context) if asset_ids is not None else rowsshort_attachments = _short_attachment_reasons(c, context_rows, roots=context)- return rows, longer, frozenset(dismissed_ids), folder_assets, folder_videos, short_attachments+ download_attachments = _download_attachment_reasons(c, rows)+ copies = _repeated_copies(c, roots=context)+ return (rows, longer, frozenset(dismissed_ids), folder_assets, folder_videos, short_attachments,+ download_attachments, copies)def scored_candidates(database, *, asset_ids=None, roots=None) -> tuple[list[dict], frozenset[int]]:"""候选按嫌疑分从高到低排序,连同用户已确认不是垃圾的 id。"""- rows, longer, dismissed_ids, folder_assets, folder_videos, short_attachments = _judgement_inputs(- database, asset_ids=asset_ids, roots=roots)+ (rows, longer, dismissed_ids, folder_assets, folder_videos, short_attachments,+ download_attachments, copies) = _judgement_inputs(database, asset_ids=asset_ids, roots=roots)# 每个目录里挂着推广名的候选有多少个;插页判据要看它有没有同伙。promo_neighbours = _promo_neighbour_counts(rows)out = []@@ -460,8 +609,9 @@ def scored_candidates(database, *, asset_ids=None, roots=None) -> tuple[list[dic"url" if suffix in INTERNET_SHORTCUT_SUFFIXESelse (d.get("medium") if d.get("medium") in JUNK_KINDS else "other"))- residue = promo_residue(nm)- promo = bool(PROMO_PHRASE.search(nm) or PROMO_DOMAIN.search(nm))+ spoken = despaced(nm)+ residue = promo_residue(spoken)+ promo = bool(PROMO_PHRASE.search(spoken) or PROMO_DOMAIN.search(spoken))page_archive = suffix in PAGE_ARCHIVE_SUFFIXESif suffix in INTERNET_SHORTCUT_SUFFIXES:s += 60; why.append("网址快捷方式")@@ -470,11 +620,10 @@ def scored_candidates(database, *, asset_ids=None, roots=None) -> tuple[list[dicattachment_reason = _attachment_junk_reason(suffix, d.get("path") or name, d.get("size") or 0)if attachment_reason:s += 60; why.append(attachment_reason)- # 只够进复核队列的分数:固定命名,和未探测时长视频的目录与体积证据。- review_only = 0- name_reason = _promo_name_reason(suffix, d.get("path") or name, d.get("size") or 0)- if name_reason:- s += JUNK_REVIEW_SCORE; review_only += JUNK_REVIEW_SCORE; why.append(name_reason)+ # 只够进复核队列的分数:固定命名、下载包里的压缩包,和未探测时长视频的目录与体积证据。+ review_only, attachment_points, attachment_reasons = _attachment_evidence(+ d, suffix, name, download_attachments, copies)+ s += attachment_points; why.extend(attachment_reasons)# 目录维度的证据:广告包的文件名往往干净(`极道世界.mp4`),唯一线索在旧导入器# 从目录名投影出来的创作者位或路径里。creator 位本身是推广站域名时,它就不再是# 「有归属所以是正片」的证据,下面两处对 creator 的信任都必须先排除这种情况。@@ -537,7 +686,7 @@ def scored_candidates(database, *, asset_ids=None, roots=None) -> tuple[list[dicd, nm, longer, promo_dir, directory_points, short_attachments.get(d["id"], ""))s += points; review_only += weak; why.extend(reasons)# 有真实创作者归属、且名字剥完仍有实质描述的,是被打了水印的正片,不是广告。- if real_owner and residue >= 14 and not (attachment_reason or name_reason):+ if real_owner and residue >= 14 and not (attachment_reason or attachment_reasons):s -= 45if review_only and s - review_only < AUTO_TRASH_SCORE:s = min(s, AUTO_TRASH_SCORE - 1)diff --git a/tests/test_ad_judgement.py b/tests/test_ad_judgement.pyindex a93900aca..4b400925a 100644--- a/tests/test_ad_judgement.py+++ b/tests/test_ad_judgement.py@@ -590,6 +590,87 @@ class ResourceJunkQueueTests(unittest.TestCase):self.assertEqual(junk._attachment_junk_reason('.gif', page + r'\1.gif', 6000), '')self.assertTrue(junk._attachment_junk_reason('.gif', page + r'\1000x90yunding.gif', 97843))+ def add_download(self, task_id, code, folder, main):+ with self.contract.database.write_transaction() as connection:+ connection.execute(+ "INSERT INTO download_task(id,provider,source_uri,target,code,state,ledger_path,asset_id,updated_at) "+ "VALUES(?,'pikpak','magnet:?xt=urn:btih:fixture',?,?,'ingested',?,?,'now')",+ (task_id, f'115-{task_id}', code, folder, main))++ def test_download_package_attachments_are_isolated_while_release_files_stay(self):+ # 样本取自 2026-11-03 的只读盘点:整理把正片移走后,推广附件留在落地目录。+ self.add(1, '115', r'B:\S1\SSIS-839\SSIS-839 标题.mp4', 'video', 5 * 1024**3, 7200)+ self.add_download(1, 'SSIS-839', r'B:\云下载\SSIS-839', 1)+ base = r'B:\云下载\SSIS-839' + '\\'+ self.add(2, '115', base + '人间尤物.mp4', 'video', 78893257, 61)+ self.add(3, '115', base + '好站全在这.txt', 'other', 470)+ self.add(4, '115', base + r'其他作品合集\kpkp3.com_259LUXU-1555.jpg', 'image', 258032)+ self.add(5, '115', base + 'hjd2048.com-0907ssis839pl.jpg', 'image', 176840)+ self.add(6, '115', base + 'poster.jpg', 'image', 180000)+ self.add(7, '115', base + '回家指南综合包.zip', 'archive', 3165927)+ self.add(8, '115', base + '特典映像.mp4', 'video', 90 * 1024**2, 600)+ self.add(9, '115', base + '未探测.mp4', 'video', 4650760)+ # 落地目录就是下载根时,不能把根下所有文件都当成这个种子的附件。+ self.add(10, '115', r'B:\云下载\notes.txt', 'other', 470)+ self.add_download(2, 'ABP-616', r'B:\云下载', 1)+ items = {item['id']: item for item in q_ads(self.contract, limit=200)['items']}+ self.assertEqual(set(items), {2, 3, 4, 7})+ self.assertIn('SSIS-839 下载包里的附带短视频,正片长 120 分', items[2]['why'])+ self.assertIn('附带压缩包', items[7]['why'])+ self.assertLess(items[7]['score'], web_batch.AUTO_TRASH_SCORE)+ checked = junk.isolate(self.contract.database, asset_ids=list(range(1, 11)), protected_ids=[1])+ self.assertEqual(sorted(checked['ids']), [2, 3, 4])++ def test_download_attachment_rules_need_a_bound_main_and_a_long_enough_release(self):+ self.add(1, '115', r'B:\云下载\IPX-867\IPX-867.mp4', 'video', 300 * 1024**2, 1200)+ self.add_download(1, 'IPX-867', r'B:\云下载\IPX-867', 1)+ self.add(2, '115', r'B:\云下载\IPX-867\片头.mp4', 'video', 20 * 1024**2, 90)+ self.add(3, '115', r'B:\云下载\SONE-136\好站全在这.txt', 'other', 470)+ with self.contract.database.write_transaction() as connection:+ connection.execute(+ "INSERT INTO download_task(id,provider,source_uri,target,code,state,ledger_path,updated_at) "+ "VALUES(2,'pikpak','magnet:?xt=urn:btih:unbound','115-2','SONE-136','ingested',?,'now')",+ (r'B:\云下载\SONE-136',))+ self.assertEqual(junk.isolate(self.contract.database, asset_ids=[1, 2, 3], protected_ids=[1])['ids'], [])++ def test_release_names_match_compact_padded_and_store_prefixed_spellings(self):+ for stem, code in (('hjd2048.com-0907mide580-h264', 'MIDE-580'), ('tkmj001jp', 'TKMJ-001'),+ ('hhd800.com@FC2-PPV-2753986_2', 'FC2-PPV-2753986'), ('STZY-019', '107STZY-019')):+ with self.subTest(stem=stem):+ self.assertTrue(junk.names_release(stem, code))+ for stem, code in (('offkab@sukebei', 'FC2-PPV-4330669'), ('kpkp3.com_259LUXU-1555', 'SGKX-016'),+ ('MIDE5800', 'MIDE-580')):+ with self.subTest(stem=stem):+ self.assertFalse(junk.names_release(stem, code))++ def test_identical_small_copies_across_release_folders_are_junk_evidence(self):+ for offset, folder in enumerate(('MIDA-117-UC', 'MIDA-154-UC', 'MIDA-234-UC')):+ self.add(20 + offset, '115', rf'B:\云下载\{folder}\情報.mp4', 'video', 11057812, 157.28)+ self.add(30 + offset, '115', rf'B:\云下载\{folder}\说明.txt', 'other', 136)+ self.add(40 + offset, '115', rf'B:\云下载\{folder}\ABC-123.jpg', 'image', 52000)+ self.add(50, '115', r'B:\云下载\MIDA-117-UC\片段.mp4', 'video', 2001226, 11)+ self.add(51, '115', r'B:\云下载\MIDA-154-UC\片段.mp4', 'video', 2001226, 11)+ self.add(52, '115', r'B:\云下载\MIDA-234-UC\片段.mp4', 'video', 2001226, 3600)+ # 创作者目录里每期都带的同一段片头不算:目录名不是发行包。+ for offset, folder in enumerate(('第一期', '第二期', '第三期')):+ self.add(70 + offset, '115', rf'B:\创作者\{folder}\片头.mp4', 'video', 3000000, 8)+ items = {item['id']: item for item in q_ads(self.contract, limit=200)['items']}+ self.assertEqual(set(items), {20, 21, 22, 30, 31, 32})+ self.assertIn('同名同体积的副本散布在 3 个目录', items[20]['why'])+ self.assertGreaterEqual(items[20]['score'], web_batch.AUTO_TRASH_SCORE)+ self.assertLess(items[30]['score'], web_batch.AUTO_TRASH_SCORE)++ def test_letter_spaced_promo_names_are_read_as_one_word(self):+ self.assertEqual(junk.despaced('x u u 6 2 . c o m'), 'xuu62.com')+ self.assertEqual(junk.despaced('最 新 位 址 獲 取'), '最新位址獲取')+ self.assertEqual(junk.despaced('No.017 - 서서 펠라_뒤치기'), 'No.017 - 서서 펠라_뒤치기')+ self.add(60, '115', r'B:\资源\x u u 9 2 .c o m.mp4', 'video', 23141112, 88.7)+ self.add(61, '115', r'B:\资源\最 新 位 址 獲 取:489155.com 收藏不迷路.txt', 'other', 136)+ self.add(62, '115', r'B:\资源\manko.fun.mp4', 'video', 11057812, 157.28)+ items = {item['id']: item for item in q_ads(self.contract, limit=200)['items']}+ self.assertEqual(set(items), {60, 61, 62})+ self.assertTrue(all(item['score'] >= web_batch.AUTO_TRASH_SCORE for item in items.values()))+if __name__ == "__main__":unittest.main()diff --git a/tests/test_complexity_ratchet.py b/tests/test_complexity_ratchet.pyindex 37a0bb260..5ef1c8578 100644--- a/tests/test_complexity_ratchet.py+++ b/tests/test_complexity_ratchet.py@@ -36,7 +36,7 @@ LIMIT = 30#: 键是 `<仓库相对路径>:<限定名>`,值是当前的分支数,只能往下改。BASELINE: dict[str, int] = {- "src/peach/junk.py:scored_candidates": 57,+ "src/peach/junk.py:scored_candidates": 54,"scripts/localize_performer_names.py:collect": 63,"src/peach/web_resource_sync.py:_resource_orphan_plan": 50,"src/peach/web_review.py:_review_rows": 40,