Peach: Local-first video metadata scraper and media manager.
fix(feeds): 官方还没出图的未发售新作先不列,代理不再转发 NOW PRINTING 占位图
longmeidao1058ee4b9c67
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 9b0d472de..a8d5ee915 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -21,6 +21,10 @@ Removed / Fixed / Security。## [未发布]+### 修复++- **关注**:离发售还远、DMM 还没出封面的新作,不再在首页、女优页的新作行和 JAV 新作页上显示「NOW PRINTING」占位图,而是先不列出;后台照常定时重取,官方出图后自动出现。已发售的旧作只拿到占位图时,封面格显示「封面未取得」,不再显示那张占位图;取图原因里也不再一律说成「多半已下架」。+## [0.55.1] - 2026-11-05### 修复diff --git a/src/peach/feed_followup.py b/src/peach/feed_followup.pyindex 8bbe2f292..30d0357d0 100644--- a/src/peach/feed_followup.py+++ b/src/peach/feed_followup.py@@ -308,6 +308,20 @@ def cover_attempt_path(cover_root: Path, code: str) -> Path:return Path(cover_root).parent / 'feed-cover-attempts' / (normalise_code_key(code) + '.json')+def awaiting_official_cover(contract, code: str, release_date: str | None) -> bool:+ """还没发售、本机没有封面、上次取图官方只回「准备中」占位图:DMM 提前一个多月上架的新作常这样。+ 新作列表先不放它,`/cover-image` 也不代理那张占位图;`cover_backlog` 照旧按 `RETRY_AFTER`+ 重取,官方出图、本机有了封面就回到列表里。已发售的旧作不算:官方那几个地址里常有一张+ 偏窄的真封面,取图记录照样写成占位图。"""+ from .jav_cover_fetch import PLACEHOLDER_MARK+ if not release_date or str(release_date)[:10] <= datetime.now().date().isoformat():+ return False+ if contract.has_cover(code):+ return False+ attempt = cover_attempt(contract.cover_root, code)+ return bool(attempt and not attempt.get("ok") and PLACEHOLDER_MARK in str(attempt.get("reason") or ""))++def cover_states(contract, rows) -> dict[str, tuple[str, str]]:"""本机还没有封面的新作,封面那一格说什么:番号键 → (状态, 原因)。diff --git a/src/peach/jav_cover_fetch.py b/src/peach/jav_cover_fetch.pyindex 7e0eb1061..ca9083a98 100644--- a/src/peach/jav_cover_fetch.py+++ b/src/peach/jav_cover_fetch.py@@ -97,10 +97,12 @@ SMALL_MIN_WIDTH = 240PLACEHOLDER = re.compile(r"now_?printing", re.I)#: 官方候选一张都量不出可用尺寸时的两种说法,下标是「有没有量到过偏小的图」。NO_USABLE_OFFICIAL = ("官方封面地址都没有取到图片", "官方封面只有缩略图或占位图")-#: 占位图是来源自己说「没有图」:DMM 下架的作品两个版本都只剩这一张。-PLACEHOLDER_REASON = "来源给的是「准备中」占位图"+#: 两句占位图原因共有的那几个字,取图记录按它认出「官方只回占位图」。+PLACEHOLDER_MARK = "「准备中」占位图"+#: 占位图是来源自己说「没有图」:DMM 发售前还没出图、下架的作品两个版本都只剩这一张。+PLACEHOLDER_REASON = f"来源给的是{PLACEHOLDER_MARK}"#: 官方候选里取到的只有占位图、没有偏小的真图时的说法。-OFFICIAL_PLACEHOLDER_ONLY = "官方封面地址只回「准备中」占位图,作品多半已从官方下架"+OFFICIAL_PLACEHOLDER_ONLY = f"官方封面地址只回{PLACEHOLDER_MARK}:未发售的新作是官方还没出图,旧作多半已从官方下架"#: 韩国 MIB 的编号不在 JAV 目录站上,封面来源一律不问;记进日志算确认落空,不重探。MIB_NOT_JAV = "韩国 MIB 不适用 JAV 封面来源"#: 量尺寸只需要 JPEG 头部,别把整张 1 MB 的图拉下来。diff --git a/src/peach/routes_media.py b/src/peach/routes_media.pyindex d66f24f54..2876422d6 100644--- a/src/peach/routes_media.py+++ b/src/peach/routes_media.py@@ -1236,8 +1236,9 @@ def cover_image(request: Request, code: str, args: dict[str, str] = Depends(requ"""库外作品的官方封面:按番号在服务端查地址,经 Peach 的出网设置取图并落盘缓存。取不到一律回 404 且不缓存:卡片按 `data-drop` 换成占位,网络恢复后下一次加载重取。+ DMM 缺图时 302 到同一主机族上的「NOW PRINTING」占位图,跟到那一跳也算取不到。"""- from .jav_cover_fetch import COVER_PROXY_HOSTS, candidate_for+ from .jav_cover_fetch import COVER_PROXY_HOSTS, PLACEHOLDER, candidate_forfrom .web_jav_detail import detail_code, stored_cover_urlstate = request.app.statetry:@@ -1250,8 +1251,13 @@ def cover_image(request: Request, code: str, args: dict[str, str] = Depends(requif hit is not None:return hittarget = ResolvedFollowMedia(url, candidate_for(url).referer, allowed_hosts=COVER_PROXY_HOSTS)- return cache.stream(key, request, lambda: open_upstream(- state.http_transport.client, request.method, target, incoming=request.headers))+ def fetch():+ upstream = open_upstream(state.http_transport.client, request.method, target, incoming=request.headers)+ if PLACEHOLDER.search(str(upstream.url)):+ upstream.close()+ raise FollowProxyError("上游给的是「准备中」占位图")+ return upstream+ return cache.stream(key, request, fetch)except (ValueError, FollowProxyError, OSError, httpx.HTTPError):return Response(status_code=404, headers={"cache-control": "no-store"})diff --git a/src/peach/web_feeds.py b/src/peach/web_feeds.pyindex d4737618c..92a798507 100644--- a/src/peach/web_feeds.py+++ b/src/peach/web_feeds.py@@ -216,9 +216,8 @@ def _backfill_solo_works(contract, transport, source, report: dict) -> None:entity_id = source["entity_id"]if source["kind"] != feeds.KIND_JAVDB_ACTOR or not entity_id:return- hidden = web_settings.hidden_compilations(contract.database)with contract.database.read_connection() as connection:- feeds.register_functions(connection, hidden)+ register_listed(contract, connection)listed = connection.execute("SELECT count(*) FROM feed_discovery d WHERE d.ignored_at IS NULL"f" AND {' AND '.join(LISTED)} AND EXISTS (SELECT 1 FROM feed_discovery_entity de"@@ -435,9 +434,8 @@ def w_feed_schedule_run(contract, _body) -> dict:def q_feeds(contract, args) -> dict:"""设置页「订阅源」小节的首屏。"""- hidden = web_settings.hidden_compilations(contract.database)with contract.database.read_connection() as connection:- feeds.register_functions(connection, hidden)+ register_listed(contract, connection)rows = feeds.sources(connection)# 表里名字前面那个圆框:有图走 `/entity-image`,判据与取景都取自身份引用那一份。for row in rows:@@ -686,8 +684,9 @@ def _follow_by_name(contract, body) -> dict:DISCOVERY_LIMIT = 24#: 哪些壳算「新作」。已经入库的不算——那条新作的使命已经完成了;设置里收起的那几类-#: 合集不算(`feeds.compilation_kind`)。两条都是读的时候现算。-#: 用到它的连接要先按设置 `feeds.register_functions`。+#: 合集不算(`feeds.compilation_kind`);还没发售、官方只回占位图的不算+#: (`feed_followup.awaiting_official_cover`)。三条都是读的时候现算。+#: 用到它的连接要先 `register_listed`。#: 已入库那一条写成不相关子查询:库里的番号键只算一遍、建成临时索引,每条壳再去查它。#: 相关子查询的写法要给每条壳把整张 `asset` 的番号重算一遍(60 条壳 × 2881 部是 486ms),#: 资料页的形状名单和新作列表都等在它上面。判空与 `=` 一致:键算不出来的壳照常列出。@@ -697,9 +696,17 @@ LISTED = (" (SELECT key FROM (SELECT normalise_code_key(a.code) AS key FROM asset a"" WHERE a.code IS NOT NULL AND COALESCE(a.disposal,'')<>'vanished') WHERE key IS NOT NULL))","NOT is_feed_hidden(d.title,d.performers,d.studio)",+ "NOT is_cover_pending(d.code,d.release_date)",)+def register_listed(contract, connection) -> None:+ """挂上 `LISTED` 用到的两个函数:合集收起按设置,「官方还没出图」按本机封面与取图记录。"""+ feeds.register_functions(connection, web_settings.hidden_compilations(contract.database))+ connection.create_function("is_cover_pending", 2, lambda code, release_date:+ feed_followup.awaiting_official_cover(contract, code, release_date))++def _discovery_performers(connection, contract, rows) -> dict[int, list[dict]]:"""新作出演名单按已有身份解析,头像契约与馆藏卡片共用。"""from .web_jav_discovery import _cached_movie@@ -765,9 +772,8 @@ def q_feed_discoveries(contract, args) -> dict:" WHERE de.discovery_id=d.id AND de.entity_id=?)")with contract.database.read_connection() as connection:params.append(entities.resolve_entity_id(connection, int(entity_id)) or int(entity_id))- hidden = web_settings.hidden_compilations(contract.database)with contract.database.read_connection() as connection:- feeds.register_functions(connection, hidden)+ register_listed(contract, connection)rows = [dict(row) for row in connection.execute("SELECT d.*, s.name AS source_name, s.kind AS source_kind"" FROM feed_discovery d LEFT JOIN feed_source s ON s.id=d.source_id"@@ -812,8 +818,8 @@ def q_feed_discoveries(contract, args) -> dict:def home_has_feed(contract, connection) -> bool:- """首页新作行是否有内容,与新作列表共用未入库、未忽略和合集筛选条件。"""- feeds.register_functions(connection, web_settings.hidden_compilations(contract.database))+ """首页新作行是否有内容,与新作列表共用 `LISTED` 与未忽略。"""+ register_listed(contract, connection)return bool(connection.execute("SELECT EXISTS(SELECT 1 FROM feed_discovery d"f" WHERE d.ignored_at IS NULL AND {' AND '.join(LISTED)})").fetchone()[0])@@ -821,7 +827,7 @@ def home_has_feed(contract, connection) -> bool:def feed_row_entity_ids(contract, connection) -> set[int]:"""资料页上会出现新作那一行的实体:判据和按人取新作的那一份同一套(`LISTED`、未忽略)。"""- feeds.register_functions(connection, web_settings.hidden_compilations(contract.database))+ register_listed(contract, connection)return {int(row[0]) for row in connection.execute("SELECT DISTINCT de.entity_id FROM feed_discovery_entity de"" JOIN feed_discovery d ON d.id=de.discovery_id"diff --git a/src/peach/web_jav_detail.py b/src/peach/web_jav_detail.pyindex 13cb341c4..f120a28bc 100644--- a/src/peach/web_jav_detail.py+++ b/src/peach/web_jav_detail.py@@ -37,11 +37,15 @@ def discovery_cover_url(item, source) -> str:def stored_cover_url(contract, code) -> str:- """`/cover-image` 按番号取的上游地址;只读账本与已缓存的来源资料,不联网。"""+ """`/cover-image` 按番号取的上游地址;只读账本与已缓存的来源资料,不联网。+ 官方只回占位图的那几部没有地址:代理缓存里存着的那张占位图也就不再给出去。"""+ from .feed_followup import awaiting_official_coverfrom .web_jav_discovery import _cached_moviewith contract.database.read_connection() as connection:- row = connection.execute("SELECT cover_url FROM feed_discovery WHERE code=? COLLATE NOCASE LIMIT 1",+ row = connection.execute("SELECT cover_url,release_date FROM feed_discovery WHERE code=? COLLATE NOCASE LIMIT 1",(code,)).fetchone()+ if row and awaiting_official_cover(contract, code, row["release_date"]):+ return ''return discovery_cover_url(dict(row) if row else {}, _cached_movie(code, contract))diff --git a/src/peach/web_jav_discovery.py b/src/peach/web_jav_discovery.pyindex ad4ac7bf6..d81fa4c69 100644--- a/src/peach/web_jav_discovery.py+++ b/src/peach/web_jav_discovery.py@@ -4,7 +4,7 @@ from __future__ import annotationsimport hashlibimport re-from . import entities, feed_followup, feeds, web_catalog, web_feeds, web_settings, wants+from . import entities, feed_followup, web_catalog, web_feeds, wantsfrom .related import rank_relatedfrom .taste_preferences import preference_profile, preference_scorefrom .genre_decisions import load_genre_decisions@@ -30,7 +30,7 @@ def _cached_movie(code, contract):def _external(connection, contract):- feeds.register_functions(connection, web_settings.hidden_compilations(contract.database))+ web_feeds.register_listed(contract, connection)rows = [dict(row) for row in connection.execute("SELECT d.* FROM feed_discovery d WHERE d.ignored_at IS NULL AND "+ " AND ".join(web_feeds.LISTED) + " ORDER BY d.id")]diff --git a/tests/test_fastapi_api.py b/tests/test_fastapi_api.pyindex a537672d5..a171116a9 100644--- a/tests/test_fastapi_api.py+++ b/tests/test_fastapi_api.py@@ -3019,6 +3019,34 @@ class FastApiContractTests(unittest.IsolatedAsyncioTestCase):self.app.state.http_transport.client = originalfake.close()+ async def test_cover_image_refuses_the_official_now_printing_placeholder(self):+ """DMM 缺图时 302 到同一主机族上的「NOW PRINTING」:回 404 不缓存,官方出图后下一次加载取到真图。"""+ connection = sqlite3.connect(self.db)+ connection.execute("INSERT INTO feed_discovery(code,cover_url,discovered_at) VALUES(?,?,'2026-10-22')",+ ('ABC-007', 'https://pics.dmm.co.jp/digital/video/abc00007/abc00007pl.jpg'))+ connection.commit()+ connection.close()+ published = {'cover': False}+ def upstream(request):+ if request.url.path.endswith('abc00007pl.jpg') and not published['cover']:+ return httpx.Response(302, request=request, headers={+ 'location': 'https://imgsrc.dmm.com/pics/mono/movie/n/now_printing/now_printing.jpg?w=800&h=800'})+ return httpx.Response(200, headers={'content-type': 'image/jpeg', 'content-length': '5'},+ stream=httpx.ByteStream(b'cover'), request=request)+ original = self.app.state.http_transport.client+ fake = httpx.Client(transport=httpx.MockTransport(upstream))+ self.app.state.http_transport.client = fake+ try:+ with patch('peach.web_jav_discovery._cached_movie', return_value={}):+ pending = await self.client.get('/cover-image?code=ABC-007&t=secret')+ self.assertEqual((pending.status_code, pending.headers['cache-control']), (404, 'no-store'))+ published['cover'] = True+ cover = await self.client.get('/cover-image?code=ABC-007&t=secret')+ self.assertEqual((cover.status_code, cover.content), (200, b'cover'))+ finally:+ self.app.state.http_transport.client = original+ fake.close()+async def test_jav_image_serves_source_bytes_with_etag_and_drops_failures(self):"""库外作品的样张与来源头像以图片本身下发:带 ETag 与私有缓存时效,取不到回 404 不缓存。"""from peach.javdb_public import PublicSourceUnavailablediff --git a/tests/test_feeds_web.py b/tests/test_feeds_web.pyindex 2e6e881e1..c358caec9 100644--- a/tests/test_feeds_web.py+++ b/tests/test_feeds_web.py@@ -482,6 +482,23 @@ class FeedWebTest(FeedWebFixture):for item in dispatch_api_get(self.contract, path, {})["items"]}self.assertEqual({code: items[code] for code in expected}, expected, path)+ def test_unreleased_works_wait_off_the_list_while_the_official_cover_is_a_placeholder(self):+ """未发售、官方只回「准备中」占位图的先不列,后台照旧重取;已发售的旧作与没取过图的照常列出。"""+ from peach.jav_cover_fetch import NotFound, OFFICIAL_PLACEHOLDER_ONLY+ future = (datetime.now() + timedelta(days=30)).date().isoformat()+ rows = {"VRKM-971": future, "HIGM-183": "2024-03-09", "VRKM-972": future}+ with self.contract.database.write_transaction() as connection:+ for code, released in rows.items():+ connection.execute("INSERT INTO feed_discovery(code,source_id,title,link,release_date,discovered_at)"+ " VALUES(?,NULL,'',NULL,?,?)", (code, released, feeds.stamp()))+ for code in ("VRKM-971", "HIGM-183"):+ feed_followup.record_cover_attempt(self.contract, code, ok=False, error=NotFound(OFFICIAL_PLACEHOLDER_ONLY))+ with patch("peach.jav_discovery_metadata.DiscoveryMetadata.ensure", return_value=None):+ for path in ("/api/feeds/discoveries", "/api/jav/releases"):+ codes = {item["code"] for item in dispatch_api_get(self.contract, path, {})["items"]}+ self.assertEqual(codes & set(rows), {"HIGM-183", "VRKM-972"}, path)+ self.assertIn("VRKM-971", self._cover_backlog())+def test_a_failing_source_reports_on_its_own_row_and_stays_scheduled(self):self._add()self.transport.responses[self.url] = HttpResponse(503, {}, b"", self.url)