feat(metadata): overview falls back to Chinese variants only (MD-2)

- adapters: overviewLanguages() limits overview/tagline chain to Chinese
  variants (zh-CN→zh-SG→zh-TW→zh-HK); titles keep full chain; season
  supplement fetch skips languages useless for overview
- setting metadata.overviewNonChineseFallback (default off) + UI switch
- media detail shows "暂无简介" for empty entity/episode overview
- scripts/overview-recompute.ts: recompute stored non-Chinese overviews
  from kv_cache only (no TMDB requests), dry-run by default, --apply
- tests: chain unit tests, Heart Signal (103147) fixture, server
  recompute test, MP differential for episode overview
- runbook + CHANGELOG

Co-Authored-By: Claude Opus 5.5 <[email protected]>
This commit is contained in:
rootandClaude Opus 5.5 committed 2026-09-30 18:04:23 +08:00
1 parent 081db4651c
commit f9901b514b
22 files changed
+887 -38

No files matched your search

+6
View File
@@ -4,6 +4,12 @@
## [Unreleased] ## [Unreleased]
### 变更(MD-2 简介不回退英文)
- 作品、季、集的简介(以及作品标语)只回退到中文变体(缺省链 zh-CN → zh-SG → zh-TW → zh-HK),不再回退到 en-US;都没有就留空。标题、别名仍按完整回退链(可到 en-US)。排在第一个中文之前的语言照常用,链里没有中文时按原链。对照 MP:每集信息只按配置语言取,简介为空就为空。
- 设置 → 元数据 → TMDB 新增「简介回退到非中文」(`metadata.overviewNonChineseFallback`,缺省关),打开后恢复旧行为。
- 媒体详情:作品和每集没有简介时显示一行小字「暂无简介」。已有的英文简介在下次刷新时按新规则清掉。
- 存量重算脚本 `scripts/overview-recompute.ts`:找出语言不在简介链里的作品 / 季 / 集简介,不请求 TMDB,只用 `kv_cache` 里缓存的 TMDB 响应按新规则重算(缓存已淘汰的直接清空);缺省 dry-run,`--apply` 写库。见 `docs/runbook/metadata.md`「存量简介重算(MD-2)」。
## [0.3.2] - 2026-09-30 ## [0.3.2] - 2026-09-30
### 新增(BR-4 浏览器 UA 与指纹统一) ### 新增(BR-4 浏览器 UA 与指纹统一)
@@ -23,7 +23,7 @@ import {
createEntityFromTmdbCandidate, createEntityFromTmdbCandidate,
enqueueNewEntityRefresh, enqueueNewEntityRefresh,
} from "../sources-identity/service/entity-from-candidate.ts"; } from "../sources-identity/service/entity-from-candidate.ts";
import { metadataLanguages } from "../sources-identity/service/metadata-rich.ts"; import { metadataLanguages, metadataTextOptions } from "../sources-identity/service/metadata-rich.ts";
import { identifyParsed, type NameIdentifyResult } from "../sources-identity/service/resolve.ts"; import { identifyParsed, type NameIdentifyResult } from "../sources-identity/service/resolve.ts";
import type { ServerContext } from "../types.ts"; import type { ServerContext } from "../types.ts";
@@ -133,7 +133,7 @@ export function createPorts(
() => rt().tmdb(), () => rt().tmdb(),
{ kind: ref.type, externalId: ref.id }, { kind: ref.type, externalId: ref.id },
{ type, id: actor.id }, { type, id: actor.id },
{ languages: await metadataLanguages(ctx) }, { languages: await metadataLanguages(ctx), ...(await metadataTextOptions(ctx)) },
); );
if (r.created) await enqueueNewEntityRefresh(ctx, r.mediaEntityId); if (r.created) await enqueueNewEntityRefresh(ctx, r.mediaEntityId);
return r.mediaEntityId; return r.mediaEntityId;
@@ -5,7 +5,7 @@ import {
createEntityFromTmdbCandidate, createEntityFromTmdbCandidate,
enqueueNewEntityRefresh, enqueueNewEntityRefresh,
} from "../sources-identity/service/entity-from-candidate.ts"; } from "../sources-identity/service/entity-from-candidate.ts";
import { metadataLanguages } from "../sources-identity/service/metadata-rich.ts"; import { metadataLanguages, metadataTextOptions } from "../sources-identity/service/metadata-rich.ts";
import { actorOfP, FilesError, type FmDeps } from "./common.ts"; import { actorOfP, FilesError, type FmDeps } from "./common.ts";
import { audit } from "./ledger.ts"; import { audit } from "./ledger.ts";
@@ -24,7 +24,10 @@ export async function entityFromCandidate(
): Promise<FilesEntityFromCandidateResult> { ): Promise<FilesEntityFromCandidateResult> {
try { try {
const languages = await metadataLanguages(d.ctx); const languages = await metadataLanguages(d.ctx);
const r = await createEntityFromTmdbCandidate(d.ctx.db, d.tmdb, ref, actorOf(principal), { languages }); const r = await createEntityFromTmdbCandidate(d.ctx.db, d.tmdb, ref, actorOf(principal), {
languages,
...(await metadataTextOptions(d.ctx)),
});
if (r.created) { if (r.created) {
await audit(d, principal, "entity_from_candidate", { ...ref, mediaEntityId: r.mediaEntityId }); await audit(d, principal, "entity_from_candidate", { ...ref, mediaEntityId: r.mediaEntityId });
// 订阅建档 / 搜索建档 / 整理向导都走这里:建档即入队 full 刷新(每集信息、富字段) // 订阅建档 / 搜索建档 / 整理向导都走这里:建档即入队 full 刷新(每集信息、富字段)
+3 -3
View File
@@ -28,7 +28,7 @@ import type { z } from "zod";
import { type AuthEnv, requireScope } from "../../http/auth.ts"; import { type AuthEnv, requireScope } from "../../http/auth.ts";
import { ApiException } from "../../http/errors.ts"; import { ApiException } from "../../http/errors.ts";
import type { SourcesRuntime } from "../sources-identity/runtime.ts"; import type { SourcesRuntime } from "../sources-identity/runtime.ts";
import { metadataLanguages } from "../sources-identity/service/metadata-rich.ts"; import { metadataLanguages, metadataTextOptions } from "../sources-identity/service/metadata-rich.ts";
import { upsertSiteRelease } from "../sources-identity/service/snapshot.ts"; import { upsertSiteRelease } from "../sources-identity/service/snapshot.ts";
import type { ModuleApp, ServerContext } from "../types.ts"; import type { ModuleApp, ServerContext } from "../types.ts";
import { matchActions } from "./actions.ts"; import { matchActions } from "./actions.ts";
@@ -169,7 +169,7 @@ export function registerSearchRoutes(app: ModuleApp, ctx: ServerContext, d: Sear
try { try {
const langs = await metadataLanguages(ctx); const langs = await metadataLanguages(ctx);
return await raceAbort( return await raceAbort(
fetchRichDetail(client, type, id, langs), fetchRichDetail(client, type, id, langs, await metadataTextOptions(ctx)),
AbortSignal.timeout(d.siteTimeoutMs), AbortSignal.timeout(d.siteTimeoutMs),
"TMDB 查询超时", "TMDB 查询超时",
); );
@@ -413,7 +413,7 @@ export function registerSearchRoutes(app: ModuleApp, ctx: ServerContext, d: Sear
try { try {
const langs = await metadataLanguages(ctx); const langs = await metadataLanguages(ctx);
const sd = await raceAbort( const sd = await raceAbort(
fetchRichSeason(client, p.id, p.season, langs), fetchRichSeason(client, p.id, p.season, langs, await metadataTextOptions(ctx)),
AbortSignal.timeout(d.siteTimeoutMs * 2), AbortSignal.timeout(d.siteTimeoutMs * 2),
"TMDB 查询超时", "TMDB 查询超时",
); );
@@ -1,4 +1,10 @@
import { fetchRichDetail, type MetadataCandidate, type TmdbClient, TmdbError } from "@umt/adapters"; import {
fetchRichDetail,
type MetadataCandidate,
type RichTextOptions,
type TmdbClient,
TmdbError,
} from "@umt/adapters";
import type { MediaKind } from "@umt/contracts"; import type { MediaKind } from "@umt/contracts";
import type { DB, Kysely } from "@umt/db"; import type { DB, Kysely } from "@umt/db";
import type { ParsedRelease } from "@umt/domain"; import type { ParsedRelease } from "@umt/domain";
@@ -26,11 +32,12 @@ export async function withLanguageChain(
client: TmdbClient, client: TmdbClient,
detail: MetadataCandidate, detail: MetadataCandidate,
languages: readonly string[] | undefined, languages: readonly string[] | undefined,
textOpts: RichTextOptions = {},
): Promise<MetadataCandidate> { ): Promise<MetadataCandidate> {
if (!languages?.length || typeof client.getJson !== "function") return detail; if (!languages?.length || typeof client.getJson !== "function") return detail;
if (detail.entityType !== "movie" && detail.entityType !== "tv") return detail; if (detail.entityType !== "movie" && detail.entityType !== "tv") return detail;
try { try {
const rich = await fetchRichDetail(client, detail.entityType, detail.externalId, languages); const rich = await fetchRichDetail(client, detail.entityType, detail.externalId, languages, textOpts);
return { ...detail, title: rich.title || detail.title, overview: rich.overview ?? detail.overview }; return { ...detail, title: rich.title || detail.title, overview: rich.overview ?? detail.overview };
} catch (e) { } catch (e) {
if (e instanceof TmdbError) return detail; if (e instanceof TmdbError) return detail;
@@ -74,7 +81,7 @@ export async function createEntityFromTmdbCandidate(
tmdb: () => Promise<TmdbClient | null>, tmdb: () => Promise<TmdbClient | null>,
ref: { externalId: string; kind: "movie" | "tv" }, ref: { externalId: string; kind: "movie" | "tv" },
actor: Actor, actor: Actor,
opts: { languages?: readonly string[] } = {}, opts: { languages?: readonly string[] } & RichTextOptions = {},
): Promise<EntityFromCandidate> { ): Promise<EntityFromCandidate> {
const extRef = { provider: "tmdb", entityType: ref.kind, externalId: ref.externalId }; const extRef = { provider: "tmdb", entityType: ref.kind, externalId: ref.externalId };
const existing = await entityByExternal(db, extRef); const existing = await entityByExternal(db, extRef);
@@ -91,7 +98,9 @@ export async function createEntityFromTmdbCandidate(
if (e instanceof TmdbError) throw new CandidateError("upstream", `TMDB 查询失败:${e.kind}`); if (e instanceof TmdbError) throw new CandidateError("upstream", `TMDB 查询失败:${e.kind}`);
throw e; throw e;
} }
detail = await withLanguageChain(client, detail, opts.languages); detail = await withLanguageChain(client, detail, opts.languages, {
overviewNonChineseFallback: opts.overviewNonChineseFallback === true,
});
for (let attempt = 0; ; attempt++) { for (let attempt = 0; ; attempt++) {
try { try {
@@ -24,7 +24,7 @@ import {
} from "./entities.ts"; } from "./entities.ts";
import { enqueueNewEntityRefresh, withLanguageChain } from "./entity-from-candidate.ts"; import { enqueueNewEntityRefresh, withLanguageChain } from "./entity-from-candidate.ts";
import { detachFromIdentityDecisions } from "./identity-decisions.ts"; import { detachFromIdentityDecisions } from "./identity-decisions.ts";
import { metadataLanguages } from "./metadata-rich.ts"; import { metadataLanguages, metadataTextOptions } from "./metadata-rich.ts";
import { recordAttempt } from "./resolve.ts"; import { recordAttempt } from "./resolve.ts";
import { backfillRecognizeCache } from "./tmdb-recognize-cache.ts"; import { backfillRecognizeCache } from "./tmdb-recognize-cache.ts";
@@ -59,7 +59,7 @@ async function entityForExternal(
if (e instanceof TmdbError) throw new ApiException("upstream_unavailable", `TMDB 查询失败:${e.kind}`); if (e instanceof TmdbError) throw new ApiException("upstream_unavailable", `TMDB 查询失败:${e.kind}`);
throw e; throw e;
} }
detail = await withLanguageChain(tmdb, detail, await metadataLanguages(rt.ctx)); detail = await withLanguageChain(tmdb, detail, await metadataLanguages(rt.ctx), await metadataTextOptions(rt.ctx));
const r = await createEntityFromTmdb(exec, detail, { const r = await createEntityFromTmdb(exec, detail, {
level: "manual", level: "manual",
evidence: { by: actorName(actor) }, evidence: { by: actorName(actor) },
@@ -22,7 +22,7 @@ import { type Actor, audit, type Exec } from "./common.ts";
import { ensureSeason, promoteUnits } from "./metadata-totals.ts"; import { ensureSeason, promoteUnits } from "./metadata-totals.ts";
// MD-1 语言回退链 + 每集元数据(REQ §7.3 字段级): // MD-1 语言回退链 + 每集元数据(REQ §7.3 字段级):
// - 作品:TMDB 富详情(fetchRichDetail)按设置 metadata.languages 取标题 / 简介 / 标语,另存中文名 titleZh、类型、评分、 // - 作品:TMDB 富详情(fetchRichDetail)按设置 metadata.languages 取标题 / 简介 / 标语(MD-2:简介 / 标语缺省只回退到中文变体),另存中文名 titleZh、类型、评分、
// 状态、时长、季数、主创、外部 ID;来源值写 metadata_field(language = 实际取到的语言),同字段换语言时删旧行; // 状态、时长、季数、主创、外部 ID;来源值写 metadata_field(language = 实际取到的语言),同字段换语言时删旧行;
// - 别名:translations / alternative_titles 写 title_alias(来源 tmdb,标语言);已有同文本别名只补空的语言; // - 别名:translations / alternative_titles 写 title_alias(来源 tmdb,标语言);已有同文本别名只补空的语言;
// - 季 / 集:fetchRichSeason 的每集 name / overview / airDate / runtime / stillPath / vote 写 metadata_field(content_unit_id), // - 季 / 集:fetchRichSeason 的每集 name / overview / airDate / runtime / stillPath / vote 写 metadata_field(content_unit_id),
@@ -40,6 +40,15 @@ export async function metadataLanguages(ctx: ServerContext): Promise<string[]> {
return langs?.length ? langs : [...DEFAULT_METADATA_LANGUAGES]; return langs?.length ? langs : [...DEFAULT_METADATA_LANGUAGES];
} }
/** 设置 metadata.overviewNonChineseFallback(MD-2):简介 / 标语是否允许回退到非中文;缺省 false */
export async function metadataTextOptions(ctx: ServerContext): Promise<{ overviewNonChineseFallback: boolean }> {
const m = await ctx.settings.get("metadata").catch(() => null);
return {
overviewNonChineseFallback:
(m as { overviewNonChineseFallback?: boolean } | null)?.overviewNonChineseFallback === true,
};
}
/** 设置 metadata.includeSpecials:full 刷新 / 回刷是否包含季 0(特别篇);缺省 true */ /** 设置 metadata.includeSpecials:full 刷新 / 回刷是否包含季 0(特别篇);缺省 true */
export async function metadataIncludeSpecials(ctx: ServerContext): Promise<boolean> { export async function metadataIncludeSpecials(ctx: ServerContext): Promise<boolean> {
const m = await ctx.settings.get("metadata").catch(() => null); const m = await ctx.settings.get("metadata").catch(() => null);
@@ -82,7 +91,7 @@ const same = (a: unknown, b: unknown) => JSON.stringify(a ?? null) === JSON.stri
* 写一组 tmdb 来源字段(作品或内容单元):值和语言都没变的跳过;语言变了删掉该字段其它语言的 tmdb 行再写。 * 写一组 tmdb 来源字段(作品或内容单元):值和语言都没变的跳过;语言变了删掉该字段其它语言的 tmdb 行再写。
* 返回写入(新增 / 变化)的字段名。 * 返回写入(新增 / 变化)的字段名。
*/ */
async function writeSourceFields( export async function writeSourceFields(
exec: Exec, exec: Exec,
target: { mediaEntityId: string } | { contentUnitId: string }, target: { mediaEntityId: string } | { contentUnitId: string },
fields: Record<string, FieldValue>, fields: Record<string, FieldValue>,
@@ -25,6 +25,7 @@ import {
applySeasonMetadata, applySeasonMetadata,
metadataIncludeSpecials, metadataIncludeSpecials,
metadataLanguages, metadataLanguages,
metadataTextOptions,
type SeasonMetadataResult, type SeasonMetadataResult,
seasonsInLibrary, seasonsInLibrary,
seasonsToRefresh, seasonsToRefresh,
@@ -189,13 +190,14 @@ export async function refreshMetadata(
// MD-1 富详情(语言回退链、类型、评分、主创…)与每集信息;注入的简化客户端没有 getJson 时走旧路径 // MD-1 富详情(语言回退链、类型、评分、主创…)与每集信息;注入的简化客户端没有 getJson 时走旧路径
const rich = typeof tmdb.getJson === "function"; const rich = typeof tmdb.getJson === "function";
const langs = rich ? await metadataLanguages(ctx) : []; const langs = rich ? await metadataLanguages(ctx) : [];
const textOpts = await metadataTextOptions(ctx);
let richDetail: TmdbRichDetail | null = null; let richDetail: TmdbRichDetail | null = null;
const richSeasons: TmdbRichSeasonDetail[] = []; const richSeasons: TmdbRichSeasonDetail[] = [];
// 可以新建集单元的季:full 刷新的全部季、追踪中订阅的季;其余季只更新已有集单元 // 可以新建集单元的季:full 刷新的全部季、追踪中订阅的季;其余季只更新已有集单元
let creatable = new Set<number>(); let creatable = new Set<number>();
if (rich) { if (rich) {
try { try {
richDetail = await fetchRichDetail(tmdb, ident.entity_type, ident.external_id, langs); richDetail = await fetchRichDetail(tmdb, ident.entity_type, ident.external_id, langs, textOpts);
} catch (e) { } catch (e) {
if (!(e instanceof TmdbError)) throw e; if (!(e instanceof TmdbError)) throw e;
ctx.logger.debug({ entityId: id, err: { message: e.message } }, "TMDB 富详情获取失败,按基础详情刷新"); ctx.logger.debug({ entityId: id, err: { message: e.message } }, "TMDB 富详情获取失败,按基础详情刷新");
@@ -213,7 +215,7 @@ export async function refreshMetadata(
creatable = new Set(opts.full ? want : tracked); creatable = new Set(opts.full ? want : tracked);
for (const n of want) { for (const n of want) {
try { try {
richSeasons.push(await fetchRichSeason(tmdb, ident.external_id, n, langs, { now: rt.now() })); richSeasons.push(await fetchRichSeason(tmdb, ident.external_id, n, langs, { now: rt.now(), ...textOpts }));
} catch (e) { } catch (e) {
ctx.logger.debug({ entityId: id, season: n, err: { message: (e as Error).message } }, "季详情获取失败"); ctx.logger.debug({ entityId: id, season: n, err: { message: (e as Error).message } }, "季详情获取失败");
} }
@@ -0,0 +1,289 @@
import {
fetchRichDetail,
fetchRichSeason,
overviewLanguages,
type TmdbClient,
TmdbError,
tmdbRequestKey,
} from "@umt/adapters";
import { sql } from "@umt/db";
import { CACHE_REGIONS } from "../../../shared-cache.ts";
import { type Actor, audit, type Exec } from "./common.ts";
import { project } from "./metadata.ts";
import { projectUnit, writeSourceFields } from "./metadata-rich.ts";
// MD-2 存量简介重算:MD-1 的回退链会把 en-US 等非中文简介填进作品 / 季 / 集的简介;新规则下简介只回退到中文变体
// (见 adapters overviewLanguages)。这里找出来源 tmdb、语言不在简介链里的简介,**不请求 TMDB**,只读 kv_cache
// (region tmdb.request)里当初取值用过的响应,按新规则重算:
// - 缓存里有响应:用同一套 fetchRichDetail / fetchRichSeason 重算(缓存读取不续期、不改最近使用时间,过期未清理的行也用);
// - 缓存里没有(已淘汰):当初能落到非中文,说明当时链里的中文变体都没有简介,按新规则直接清空(basis = no_cache);
// - 作品另按缓存重算标语(标语不记语言,只能在缓存命中时重算)。
// 写入走字段来源表 + 投影(与刷新同一套 writeSourceFields / project / projectUnit),用户锁定 / 覆盖的字段最终值不变。
// 入口:scripts/overview-recompute.ts(缺省 dry-run,--apply 写库)。
export interface OverviewValue {
value: string | null;
language: string | null;
}
export interface OverviewChange {
target: "entity" | "season" | "episode";
mediaEntityId: string;
entityTitle: string;
contentUnitId: string | null;
season: number | null;
episode: number | null;
field: "overview" | "tagline";
from: OverviewValue;
to: OverviewValue;
/** cache = 按缓存的 TMDB 响应重算;no_cache = 缓存里没有,按"当时中文为空"清空 */
basis: "cache" | "no_cache";
/** tmdb 来源修订(写 metadata_field.source_revision) */
revision: string;
}
export interface OverviewRecomputePlan {
languages: string[];
overviewLanguages: string[];
/** 语言不在简介链里的简介行数(作品 / 季与集) */
scanned: { entities: number; units: number };
changes: OverviewChange[];
/** 读缓存次数 / 未命中次数 */
cache: { reads: number; misses: number };
}
type CacheClient = Pick<TmdbClient, "getJson"> & { reads: number; misses: number };
/** 只读 kv_cache 的 TMDB 客户端:命中返回缓存值,未命中抛 notfound(fetchRichSeason 的补拉遇错即停) */
export function cacheOnlyTmdb(exec: Exec): CacheClient {
const c: CacheClient = {
reads: 0,
misses: 0,
async getJson<T>(path: string, params: Record<string, string | undefined> = {}): Promise<T> {
const key = tmdbRequestKey(path, params);
c.reads++;
const r = await sql<{ value: unknown }>`select value from kv_cache
where region = ${CACHE_REGIONS.tmdbRequest} and key = ${key}`.execute(exec);
const v = r.rows[0]?.value;
if (v === undefined || v === null) {
c.misses++;
throw new TmdbError("notfound", `未缓存 ${key}`, { path });
}
return v as T;
},
};
return c;
}
const same = (a: OverviewValue, b: OverviewValue) =>
(a.value ?? null) === (b.value ?? null) && a.language === b.language;
const str = (v: unknown): string | null => (typeof v === "string" ? v : null);
/** 生成重算清单(只读;不请求 TMDB) */
export async function planOverviewRecompute(
exec: Exec,
opts: { languages: readonly string[]; overviewNonChineseFallback?: boolean; now?: Date; entityId?: string | null },
): Promise<OverviewRecomputePlan> {
const languages = [...opts.languages];
const textLangs = overviewLanguages(languages, opts.overviewNonChineseFallback === true);
const only = opts.entityId ?? null;
const tmdb = cacheOnlyTmdb(exec);
const changes: OverviewChange[] = [];
// ── 作品 ──
const ents = await sql<{
id: string;
title: string;
value: unknown;
language: string;
entity_type: "movie" | "tv";
external_id: string;
}>`
select distinct on (f.media_entity_id) f.media_entity_id as id, e.title, f.value, f.language, x.entity_type, x.external_id
from metadata_field f
join media_entity e on e.id = f.media_entity_id
join external_identity x on x.media_entity_id = e.id and x.provider = 'tmdb' and x.status = 'accepted'
and x.entity_type in ('movie', 'tv')
where f.content_unit_id is null and f.source = 'tmdb' and f.field = 'overview'
and f.language is not null and not (f.language = any(${textLangs}::text[]))
and (${only}::uuid is null or f.media_entity_id = ${only}::uuid)
order by f.media_entity_id, x.created_at`.execute(exec);
for (const e of ents.rows) {
const revision = `tmdb:${e.entity_type}/${e.external_id}`;
const base = {
target: "entity" as const,
mediaEntityId: e.id,
entityTitle: e.title,
contentUnitId: null,
season: null,
episode: null,
revision,
};
const from = { value: str(e.value), language: e.language };
try {
const rich = await fetchRichDetail(tmdb, e.entity_type, e.external_id, languages, {
overviewNonChineseFallback: opts.overviewNonChineseFallback === true,
});
const to = { value: rich.overview, language: rich.overviewLanguage };
if (!same(from, to)) changes.push({ ...base, field: "overview", from, to, basis: "cache" });
const tag = await exec
.selectFrom("metadata_field")
.select(["value", "language"])
.where("media_entity_id", "=", e.id)
.where("content_unit_id", "is", null)
.where("source", "=", "tmdb")
.where("field", "=", "tagline")
.executeTakeFirst();
const tagFrom = { value: str(tag?.value), language: tag?.language ?? null };
const tagTo = { value: rich.tagline, language: null };
if (tag && !same(tagFrom, tagTo))
changes.push({ ...base, field: "tagline", from: tagFrom, to: tagTo, basis: "cache" });
} catch (err) {
if (!(err instanceof TmdbError)) throw err;
changes.push({ ...base, field: "overview", from, to: { value: null, language: null }, basis: "no_cache" });
}
}
// ── 季 / 集 ──
const units = await sql<{
unit_id: string;
kind: string;
season_number: number | null;
episode_number: number | null;
media_entity_id: string;
title: string;
value: unknown;
language: string;
external_id: string;
}>`
select distinct on (f.content_unit_id) f.content_unit_id as unit_id, u.kind, u.season_number, u.episode_number,
u.media_entity_id, e.title, f.value, f.language, x.external_id
from metadata_field f
join content_unit u on u.id = f.content_unit_id
join media_entity e on e.id = u.media_entity_id
join external_identity x on x.media_entity_id = e.id and x.provider = 'tmdb' and x.status = 'accepted'
and x.entity_type = 'tv'
where f.media_entity_id is null and f.source = 'tmdb' and f.field = 'overview'
and f.language is not null and not (f.language = any(${textLangs}::text[]))
and u.status <> 'retired' and u.kind in ('season', 'episode') and u.season_number is not null
and (${only}::uuid is null or u.media_entity_id = ${only}::uuid)
order by f.content_unit_id, x.created_at`.execute(exec);
const groups = new Map<string, typeof units.rows>();
for (const u of units.rows) {
const k = `${u.media_entity_id}|${u.external_id}|${u.season_number}`;
groups.set(k, [...(groups.get(k) ?? []), u]);
}
for (const rows of groups.values()) {
const first = rows[0]!;
const season = first.season_number!;
let sd: Awaited<ReturnType<typeof fetchRichSeason>> | null = null;
try {
sd = await fetchRichSeason(tmdb, first.external_id, season, languages, {
now: opts.now ?? new Date(),
overviewNonChineseFallback: opts.overviewNonChineseFallback === true,
});
} catch (err) {
if (!(err instanceof TmdbError)) throw err;
}
const sorted = [...rows].sort((a, b) => (a.episode_number ?? -1) - (b.episode_number ?? -1));
for (const u of sorted) {
const from = { value: str(u.value), language: u.language };
let to: OverviewValue = { value: null, language: null };
let basis: OverviewChange["basis"] = "no_cache";
if (sd && u.kind === "season") {
to = { value: sd.overview, language: sd.overviewLanguage };
basis = "cache";
} else if (sd) {
const ep = sd.episodes.find((x) => x.episodeNumber === u.episode_number);
if (ep) {
to = { value: ep.overview, language: ep.overviewLanguage };
basis = "cache";
}
}
if (same(from, to)) continue;
changes.push({
target: u.kind === "season" ? "season" : "episode",
mediaEntityId: u.media_entity_id,
entityTitle: u.title,
contentUnitId: u.unit_id,
season,
episode: u.kind === "season" ? null : u.episode_number,
field: "overview",
from,
to,
basis,
revision: `tmdb:tv/${u.external_id}`,
});
}
}
return {
languages,
overviewLanguages: textLangs,
scanned: { entities: ents.rows.length, units: units.rows.length },
changes,
cache: { reads: tmdb.reads, misses: tmdb.misses },
};
}
/**
* 按清单写库:同刷新一样写字段来源(语言变了删旧语言行)并投影;锁定 / 覆盖的字段最终值不变。
* 调用方负责事务(脚本在一个事务里执行)。返回投影后最终值有变化的作品 / 单元数。
*/
export async function applyOverviewRecompute(
exec: Exec,
plan: OverviewRecomputePlan,
opts: { now: Date; actor?: Actor },
): Promise<{ entities: number; units: number }> {
const byTarget = new Map<string, OverviewChange[]>();
for (const c of plan.changes) {
const k = c.contentUnitId ? `u:${c.contentUnitId}` : `e:${c.mediaEntityId}`;
byTarget.set(k, [...(byTarget.get(k) ?? []), c]);
}
let entities = 0;
let units = 0;
for (const list of byTarget.values()) {
const c0 = list[0]!;
const fields = Object.fromEntries(list.map((c) => [c.field, c.to]));
const names = list.map((c) => c.field);
if (c0.contentUnitId) {
const existing = await exec
.selectFrom("metadata_field")
.select(["field", "value", "language"])
.where("content_unit_id", "=", c0.contentUnitId)
.where("media_entity_id", "is", null)
.where("source", "=", "tmdb")
.where("field", "in", names)
.execute();
await writeSourceFields(exec, { contentUnitId: c0.contentUnitId }, fields, existing, c0.revision, opts.now);
const r = await projectUnit(exec, c0.contentUnitId, opts.now);
if (r.changedFields.length) units++;
} else {
const existing = await exec
.selectFrom("metadata_field")
.select(["field", "value", "language"])
.where("media_entity_id", "=", c0.mediaEntityId)
.where("content_unit_id", "is", null)
.where("source", "=", "tmdb")
.where("field", "in", names)
.execute();
await writeSourceFields(exec, { mediaEntityId: c0.mediaEntityId }, fields, existing, c0.revision, opts.now);
const r = await project(exec, c0.mediaEntityId, opts.now);
if (r.changedFields.length) entities++;
}
}
if (plan.changes.length) {
await audit(
exec,
opts.actor ?? { type: "system", id: "overview-recompute" },
"metadata.overview_recompute",
{ type: "metadata", id: null },
{
overviewLanguages: plan.overviewLanguages,
changes: plan.changes.length,
noCache: plan.changes.filter((c) => c.basis === "no_cache").length,
},
);
}
return { entities, units };
}
@@ -0,0 +1,148 @@
import { createTmdbClient } from "@umt/adapters";
import { PgCache } from "@umt/db";
import { createMockTmdb, HEART_SIGNAL_FIXTURE, type MockTmdb, TMDB_FIXTURES } from "@umt/testkit";
import { afterAll, beforeAll, describe, expect, it } from "vitest";
import { createEntityFromTmdbCandidate } from "../../src/modules/sources-identity/service/entity-from-candidate.ts";
import { refreshMetadata } from "../../src/modules/sources-identity/service/metadata.ts";
import {
applyOverviewRecompute,
planOverviewRecompute,
} from "../../src/modules/sources-identity/service/overview-recompute.ts";
import { CACHE_REGIONS } from "../../src/shared-cache.ts";
import { createHarness, type Harness, pgAvailable } from "./harness.ts";
const up = await pgAvailable();
type Final = { value: unknown; source: string; language: string | null };
const CHAIN = ["zh-CN", "zh-SG", "zh-TW", "zh-HK", "en-US"];
describe.skipIf(!up)("MD-2 简介只回退到中文变体 + 存量重算", () => {
let h: Harness;
let tmdb: MockTmdb;
let id: string;
const actor = { type: "user" as const, id: "tester" };
const now = () => new Date("2026-01-01T00:00:00Z");
beforeAll(async () => {
tmdb = await createMockTmdb({ fixtures: [...TMDB_FIXTURES, HEART_SIGNAL_FIXTURE] });
// 走 PG 请求缓存(kv_cache tmdb.request),与生产一致:存量重算只读这里
const cache = {
get: (k: string) => new PgCache(h.db).get(CACHE_REGIONS.tmdbRequest, k),
set: (k: string, v: unknown, ttlMs: number) =>
new PgCache(h.db).set(CACHE_REGIONS.tmdbRequest, k, v, { ttl: ttlMs / 1000 }),
};
h = await createHarness("md2", {
now,
tmdb: () =>
createTmdbClient({ apiKey: "test-tmdb-key", baseUrl: `${tmdb.url}/3`, retries: 0, sharedCache: cache }),
});
});
afterAll(async () => {
await h?.close();
await tmdb?.close();
});
const setMeta = async (v: Record<string, unknown> | null) => {
await h.db.deleteFrom("settings").where("key", "=", "metadata").execute();
if (v)
await h.db
.insertInto("settings")
.values({ key: "metadata", value: JSON.stringify(v) } as never)
.execute();
};
const entityMeta = async () =>
(await h.db.selectFrom("media_entity").select("metadata").where("id", "=", id).executeTakeFirstOrThrow())
.metadata as Record<string, Final>;
const eps = async () =>
(
await h.db
.selectFrom("content_unit")
.select(["episode_number", "title", "metadata"])
.where("media_entity_id", "=", id)
.where("kind", "=", "episode")
.orderBy("episode_number")
.execute()
).map(
(u) =>
({ ...(u.metadata as Record<string, Final | undefined>), ep: u.episode_number, title: u.title }) as {
ep: number | null;
title: string | null;
overview?: Final;
name?: Final;
},
);
it("允许回退到非中文时(旧行为):英文简介进库", async () => {
await setMeta({ languages: CHAIN, overviewNonChineseFallback: true });
const r = await createEntityFromTmdbCandidate(h.db, () => h.rt.tmdb(), { externalId: "103147", kind: "tv" }, actor);
id = r.mediaEntityId;
expect((await refreshMetadata(h.rt, id, { full: true })).status).toBe("refreshed");
expect((await entityMeta()).overview).toMatchObject({ value: "A Chinese dating reality show.", language: "en-US" });
const [e1, e2] = await eps();
expect(e1!.overview).toMatchObject({
value: "If Your Heart Skips a Beat, Play Your Cards Openly!",
language: "en-US",
});
expect(e2!.overview).toMatchObject({ value: "Who Is the One?", language: "en-US" });
});
it("存量重算:dry-run 只列清单,不请求 TMDB、不改库;--apply 按缓存清空,标题不动", async () => {
await setMeta({ languages: CHAIN });
const calls = tmdb.calls.length;
const plan = await planOverviewRecompute(h.db, { languages: CHAIN, now: now() });
expect(tmdb.calls.length).toBe(calls);
expect(plan.overviewLanguages).toEqual(["zh-CN", "zh-SG", "zh-TW", "zh-HK"]);
expect(plan.scanned).toEqual({ entities: 1, units: 2 });
expect(plan.cache.misses).toBe(0);
const brief = plan.changes.map((c) => [c.target, c.episode, c.field, c.from.language, c.to.value, c.basis]);
expect(brief).toEqual([
["entity", null, "overview", "en-US", null, "cache"],
["episode", 1, "overview", "en-US", null, "cache"],
["episode", 2, "overview", "en-US", null, "cache"],
]);
// dry-run 不改库
expect((await entityMeta()).overview?.language).toBe("en-US");
const r = await h.db.transaction().execute((trx) => applyOverviewRecompute(trx, plan, { now: now() }));
expect(r).toEqual({ entities: 1, units: 2 });
expect(tmdb.calls.length).toBe(calls);
expect((await entityMeta()).overview).toMatchObject({ value: null, language: null, source: "tmdb" });
const [e1, e2, e3] = await eps();
expect(e1!.overview?.value).toBeNull();
expect(e1!.title).toBe("心动不如行动,大胆亮出底牌!");
expect(e2!.overview?.value).toBeNull();
// 标题仍可回退到英文
expect(e2!.name).toMatchObject({ value: "Who Is the One?", language: "en-US" });
expect(e3!.overview).toMatchObject({ value: "約會之夜。", language: "zh-TW" });
// 旧语言行删掉,只留一行 tmdb 来源
const rows = await h.db
.selectFrom("metadata_field")
.select(["language", "value"])
.where("media_entity_id", "=", id)
.where("field", "=", "overview")
.execute();
expect(rows).toEqual([{ language: null, value: null }]);
// 再跑一次没有要改的
expect((await planOverviewRecompute(h.db, { languages: CHAIN, now: now() })).changes).toEqual([]);
});
it("缓存被淘汰时不请求 TMDB,按「当时中文为空」清空;刷新按新规则也不再写英文", async () => {
await setMeta({ languages: CHAIN, overviewNonChineseFallback: true });
await refreshMetadata(h.rt, id, { full: true });
expect((await eps())[0]!.overview?.language).toBe("en-US");
await setMeta({ languages: CHAIN });
await h.db.deleteFrom("kv_cache" as never).execute();
const calls = tmdb.calls.length;
const plan = await planOverviewRecompute(h.db, { languages: CHAIN, now: now() });
expect(tmdb.calls.length).toBe(calls);
expect(plan.changes.every((c) => c.basis === "no_cache" && c.to.value === null)).toBe(true);
expect(plan.changes).toHaveLength(3);
// 不跑重算,直接刷新:按新规则清掉英文简介
await refreshMetadata(h.rt, id, { full: true });
expect((await entityMeta()).overview?.value).toBeNull();
const [e1, e2] = await eps();
expect(e1!.overview?.value).toBeNull();
expect(e2!.overview?.value).toBeNull();
expect(e2!.name?.value).toBe("Who Is the One?");
});
});
@@ -183,7 +183,7 @@ export function MediaHero({ facts, badges }: { facts: MediaFacts; badges?: React
className="mt-2 max-w-3xl text-caption leading-relaxed text-on-media" className="mt-2 max-w-3xl text-caption leading-relaxed text-on-media"
/> />
) : ( ) : (
<p className="mt-3 text-caption text-on-media-tertiary">没有简介。</p> <p className="mt-2 text-meta text-on-media-tertiary">暂无简介</p>
)} )}
{dates.length || facts.externalLinks.length ? ( {dates.length || facts.externalLinks.length ? (
<div className="mt-3 flex flex-wrap items-center gap-x-3 gap-y-1 text-meta text-on-media-tertiary"> <div className="mt-3 flex flex-wrap items-center gap-x-3 gap-y-1 text-meta text-on-media-tertiary">
@@ -332,7 +332,11 @@ export function EpisodeRow({ e }: { e: EpisodeRowData }) {
</div> </div>
</div> </div>
<p className="mt-0.5 text-meta text-muted-foreground tabular-nums">{meta.join(" · ")}</p> <p className="mt-0.5 text-meta text-muted-foreground tabular-nums">{meta.join(" · ")}</p>
{e.overview ? <ClampText text={e.overview} lines={2} className="mt-1 text-caption text-foreground/85" /> : null} {e.overview ? (
<ClampText text={e.overview} lines={2} className="mt-1 text-caption text-foreground/85" />
) : (
<p className="mt-0.5 text-meta text-muted-foreground">暂无简介</p>
)}
{e.extra} {e.extra}
</div> </div>
</li> </li>
+10 -1
View File
@@ -51,7 +51,14 @@ const TMDB_CARD: CardSpec = {
label: "语言回退链", label: "语言回退链",
kind: "text", kind: "text",
placeholder: DEFAULT_LANGUAGES.join(", "), placeholder: DEFAULT_LANGUAGES.join(", "),
hint: "标题、简介和每集信息按顺序取第一个有值的语言;中文名取链中第一个中文值。逗号分隔,如 zh-CN, zh-SG", hint: "标题和每集信息按顺序取第一个有值的语言,简介默认只取链里的中文;中文名取链中第一个中文值。逗号分隔,如 zh-CN, zh-SG",
wide: true,
},
{
name: "metadata.overviewNonChineseFallback",
label: "简介回退到非中文",
kind: "switch",
hint: "没有中文简介时,按回退链取英文等语言。关闭时留空,标题不受影响",
wide: true, wide: true,
}, },
{ {
@@ -68,6 +75,7 @@ const TMDB_CARD: CardSpec = {
"metadata.tmdbBaseUrl": s.metadata.tmdbBaseUrl ?? "", "metadata.tmdbBaseUrl": s.metadata.tmdbBaseUrl ?? "",
"metadata.languages": (s.metadata.languages ?? DEFAULT_LANGUAGES).join(", "), "metadata.languages": (s.metadata.languages ?? DEFAULT_LANGUAGES).join(", "),
"metadata.includeSpecials": s.metadata.includeSpecials !== false, "metadata.includeSpecials": s.metadata.includeSpecials !== false,
"metadata.overviewNonChineseFallback": s.metadata.overviewNonChineseFallback === true,
}), }),
toPatch: (v) => { toPatch: (v) => {
const key = secretPatch(v["metadata.tmdbApiKey"]); const key = secretPatch(v["metadata.tmdbApiKey"]);
@@ -76,6 +84,7 @@ const TMDB_CARD: CardSpec = {
tmdbBaseUrl: str(v["metadata.tmdbBaseUrl"]) || null, tmdbBaseUrl: str(v["metadata.tmdbBaseUrl"]) || null,
languages: parseLanguages(str(v["metadata.languages"])), languages: parseLanguages(str(v["metadata.languages"])),
includeSpecials: v["metadata.includeSpecials"] === true, includeSpecials: v["metadata.includeSpecials"] === true,
overviewNonChineseFallback: v["metadata.overviewNonChineseFallback"] === true,
...(key !== undefined ? { tmdbApiKey: key } : {}), ...(key !== undefined ? { tmdbApiKey: key } : {}),
}, },
}; };
+1 -1
View File
@@ -204,5 +204,5 @@ NFO / 图片投影、身份巡检与挂载根因聚合、HTTP 失败自动切浏
- **文件预览不加 ffmpeg(用户 2026-09-30)**:视频缩略图 / 截图条不做,镜像不引入 ffmpeg(+141MB);`docs/design/file-preview-research.md` 对应项标为不做。 - **文件预览不加 ffmpeg(用户 2026-09-30)**:视频缩略图 / 截图条不做,镜像不引入 ffmpeg(+141MB);`docs/design/file-preview-research.md` 对应项标为不做。
- 用户 2026-09-30:BR-4 直接做完上线;《征途》重复单元不再处理(刷新时自动合并即可)。 - 用户 2026-09-30:BR-4 直接做完上线;《征途》重复单元不再处理(刷新时自动合并即可)。
- **用户 2026-09-30 决定**:短剧补全(META-1 后续)继续做;文件预览的音频 / 压缩包 / 原盘目录不做;修完零碎问题后接其他站点真机测试。 - **用户 2026-09-30 决定**:短剧补全(META-1 后续)继续做;文件预览的音频 / 压缩包 / 原盘目录不做;修完零碎问题后接其他站点真机测试。
- **每集简介英文(MD-2,用户以《心动的信号》为例)**:协调者核对——TMDB 匹配正确(103147),每期标题为 zh-CN;但 TMDB 上这些集没有中文简介,只有英文"简介"(实为标题英译),MD-1 的语言回退链把 en-US 填进了简介。改法:简介的回退链只到中文变体(zh-CN→zh-SG→zh-TW→zh-HK),不回退到英文,没有就留空并显示"暂无简介";标题仍可回退英文;存量数据按新规则重算(只读 dry-run 后应用)。 - **每集简介英文(MD-2,用户以《心动的信号》为例)**:协调者核对——TMDB 匹配正确(103147),每期标题为 zh-CN;但 TMDB 上这些集没有中文简介,只有英文"简介"(实为标题英译),MD-1 的语言回退链把 en-US 填进了简介。改法:简介的回退链只到中文变体(zh-CN→zh-SG→zh-TW→zh-HK),不回退到英文,没有就留空并显示"暂无简介";标题仍可回退英文;存量数据按新规则重算(只读 dry-run 后应用)。**状态:已实现(分支 md2,待合并)**——adapters `overviewLanguages`(简介链)+ 设置 `metadata.overviewNonChineseFallback`(缺省关)+ 详情页「暂无简介」+ 存量重算 `scripts/overview-recompute.ts`(只读 kv_cache,缺省 dry-run);见 runbook metadata.md「存量简介重算(MD-2)」。
- **零碎问题(2026-09-30 讨论)**:跨进程限流已在迁移 0041 做完,CHANGELOG「已知限制」过期条目删除;子会话思考英文 → 子会话系统提示补中文要求 + 测试;`agent.spawn` 重复调用 → 同目标去重(返回已有子会话);camofox 空闲 5 分钟回收 → 桌面接管 / 智能体会话打开期间由 app 定期保活,关闭后照常回收;憨憨 UA"待切换"→ 站点页「刷新登录」即可切换,不另做。 - **零碎问题(2026-09-30 讨论)**:跨进程限流已在迁移 0041 做完,CHANGELOG「已知限制」过期条目删除;子会话思考英文 → 子会话系统提示补中文要求 + 测试;`agent.spawn` 重复调用 → 同目标去重(返回已有子会话);camofox 空闲 5 分钟回收 → 桌面接管 / 智能体会话打开期间由 app 定期保活,关闭后照常回收;憨憨 UA"待切换"→ 站点页「刷新登录」即可切换,不另做。
+22 -2
View File
@@ -5,8 +5,8 @@
- 设置 → 元数据 → TMDB →「语言回退链」,即 `settings.metadata.languages`,缺省 `zh-CN, zh-SG, zh-TW, zh-HK, en-US`。 - 设置 → 元数据 → TMDB →「语言回退链」,即 `settings.metadata.languages`,缺省 `zh-CN, zh-SG, zh-TW, zh-HK, en-US`。
- 作品标题:按链逐个语言查 TMDB `translations`(语言 + 地区精确匹配);该语言译名为空时查同地区、文字相符的 `alternative_titles`(类型为空的优先);原始语言等于链中语言时原名也算。全部落空用请求语言下的 name / 原名。对照 MP `__update_tmdbinfo_cn_title`(中文名 → CN 别名 → SG 译名)。 - 作品标题:按链逐个语言查 TMDB `translations`(语言 + 地区精确匹配);该语言译名为空时查同地区、文字相符的 `alternative_titles`(类型为空的优先);原始语言等于链中语言时原名也算。全部落空用请求语言下的 name / 原名。对照 MP `__update_tmdbinfo_cn_title`(中文名 → CN 别名 → SG 译名)。
- 中文名 `titleZh`:只在链里的中文语言中按顺序取(zh-CN 优先),值必须含汉字。链首改成 en-US 时标题变英文,`titleZh` 不变。 - 中文名 `titleZh`:只在链里的中文语言中按顺序取(zh-CN 优先),值必须含汉字。链首改成 en-US 时标题变英文,`titleZh` 不变。
- 简介 / 标语:按链取 `translations` 的第一个非空值。 - 简介 / 标语(MD-2):只按「简介链」取 `translations` 的第一个非空值,取不到留空(界面显示「暂无简介」)。简介链 = 回退链里第一个中文之前的语言 + 链里全部中文,之后的非中文去掉;缺省链得到 `zh-CN, zh-SG, zh-TW, zh-HK`。链里没有中文时简介链就是原链。设置 → 元数据 → TMDB「简介回退到非中文」(`metadata.overviewNonChineseFallback`,缺省关)打开后简介链 = 原链(旧行为)。原因:TMDB 上很多剧集(如《心动的信号》103147)只有英文"简介",而且常是标题英译;MP 每集信息只按配置语言取,简介为空就为空。请求语言下的原值若等于链外某语言的译文(TMDB / 反代把缺失字段回退了),不当作链内的值。标题 / 别名不受影响。
- 季 / 集:先按链首请求 `/tv/{id}/season/{n}`;已播出的集缺标题(TMDB 占位名"第 N 集""Episode N")或缺简介时,按链逐个语言补拉同一端点,补齐即停;链走完仍没有标题的集保留占位名(语言记为空)。未播出的集不触发补拉。 - 季 / 集:先按链首请求 `/tv/{id}/season/{n}`;已播出的集缺标题(TMDB 占位名"第 N 集""Episode N")或缺简介时,按链逐个语言补拉同一端点,补齐即停(简介只从简介链里的语言取;只缺简介且剩下的语言都不在简介链里时不再补拉);链走完仍没有标题的集保留占位名(语言记为空)。未播出的集不触发补拉。
- 每个字段的来源值写 `metadata_field`,`language` 记实际取到的语言;同字段语言变了,旧语言的 tmdb 行删掉。 - 每个字段的来源值写 `metadata_field`,`language` 记实际取到的语言;同字段语言变了,旧语言的 tmdb 行删掉。
- `translations` 与 `alternative_titles` 写 `title_alias`(来源 tmdb,标语言;别名只在地区与文字相符时标,如 CN + 汉字 → zh-CN)。已有同文本别名(繁简归一化后同键)不重复登记,只给语言为空的 tmdb 行补语言。 - `translations` 与 `alternative_titles` 写 `title_alias`(来源 tmdb,标语言;别名只在地区与文字相符时标,如 CN + 汉字 → zh-CN)。已有同文本别名(繁简归一化后同键)不重复登记,只给语言为空的 tmdb 行补语言。
@@ -44,6 +44,25 @@ curl -s -H "Authorization: Bearer $KEY" "$BASE/api/v1/media-entities?limit=200"
while read -r id; do curl -s -X POST -H "Authorization: Bearer $KEY" "$BASE/api/v1/media-entities/$id/refresh?full=1"; done while read -r id; do curl -s -X POST -H "Authorization: Bearer $KEY" "$BASE/api/v1/media-entities/$id/refresh?full=1"; done
``` ```
## 存量简介重算(MD-2)
MD-1 的回退链把 en-US 等非中文简介写进过作品 / 季 / 集。新规则上线后,刷新会按新规则清掉;不想等刷新、或想先看范围时用脚本:
```sh
# 看清单(只读):按作品列出每项「旧语言 旧值 → 新语言 新值(依据)」和合计
DATABASE_URL=postgres://umt:***@<host>:5432/umt deno run -A scripts/overview-recompute.ts
DATABASE_URL=… deno run -A scripts/overview-recompute.ts --entity <作品 id> # 只看一部
DATABASE_URL=… deno run -A scripts/overview-recompute.ts --json # 清单 JSON
DATABASE_URL=… deno run -A scripts/overview-recompute.ts --apply # 写库(单个事务 + 审计 metadata.overview_recompute)
```
- 对象:来源 tmdb、`language` 不为空且不在简介链里的 `overview`(作品 / 季单元 / 集单元,未退役,有已接受的 TMDB 身份)。语言链与开关取库里的设置。
- **不请求 TMDB**:用只读 `kv_cache`(region `tmdb.request`)的客户端跑同一套 `fetchRichDetail` / `fetchRichSeason`,读缓存不续期、不改最近使用时间,过期但还没被清理的行也用。依据写在清单里:
- `缓存`:按缓存的响应重算,可能改成某个中文变体的简介,也可能为空;作品的标语一起按缓存重算;
- `无缓存→清空`:缓存已淘汰。当初能落到非中文,说明那时简介链里的语言都没有简介,按新规则直接清空。
- 写库与刷新同一套:写字段来源(语言变了删旧语言行)再投影;用户锁定 / 覆盖的字段最终值不变。重跑幂等(第二次清单为空)。
- 按 lesson 22:对真实库先跑 dry-run,由协调者过目清单再 `--apply`。代码:`apps/server/src/modules/sources-identity/service/overview-recompute.ts`。
## 集字段(字段级,可锁) ## 集字段(字段级,可锁)
- 集单元字段:`name`(投影到 `content_unit.title`)、`overview`、`airDate`(投影到 `air_date`)、`runtime`、`stillPath`、`voteAverage`、`voteCount`;季单元另有 `posterPath`。最终值在 `content_unit.metadata`,每个字段带来源 / 语言 / 时间 / 锁。 - 集单元字段:`name`(投影到 `content_unit.title`)、`overview`、`airDate`(投影到 `air_date`)、`runtime`、`stillPath`、`voteAverage`、`voteCount`;季单元另有 `posterPath`。最终值在 `content_unit.metadata`,每个字段带来源 / 语言 / 时间 / 锁。
@@ -65,6 +84,7 @@ curl -s -H "Authorization: Bearer $KEY" "$BASE/api/v1/media-entities?limit=200"
| 现象 | 看哪里 | | 现象 | 看哪里 |
|---|---| |---|---|
| 标题仍是英文原名 | 作品详情「字段来源与锁」里 `标题` 的来源与语言;是否被锁;设置里的语言链;对作品跑 `full=1` | | 标题仍是英文原名 | 作品详情「字段来源与锁」里 `标题` 的来源与语言;是否被锁;设置里的语言链;对作品跑 `full=1` |
| 简介显示「暂无简介」 | TMDB 该作品 / 集在简介链的语言里都没有简介,属正常(MD-2 不再用英文顶替);要英文时打开「简介回退到非中文」再刷新;也可在每集行手动填写并锁定 |
| 某集没有中文标题 | 该集在 TMDB 各语言都没有译名时保留占位名 / 英文名,属正常;可在每集行手动改并锁定 | | 某集没有中文标题 | 该集在 TMDB 各语言都没有译名时保留占位名 / 英文名,属正常;可在每集行手动改并锁定 |
| 季里没有集 | 默认刷新不为未追踪的季建集单元,点「更新每集信息」(full) | | 季里没有集 | 默认刷新不为未追踪的季建集单元,点「更新每集信息」(full) |
| 没有特别篇 | TMDB 该剧有没有季 0;设置 `metadata.includeSpecials` 是否关了;对作品跑 `full=1` | | 没有特别篇 | TMDB 该剧有没有季 0;设置 `metadata.includeSpecials` 是否关了;对作品跑 `full=1` |
+12 -2
View File
@@ -123,6 +123,17 @@ export interface TmdbSharedCache {
set(key: string, value: unknown, ttlMs: number): Promise<void>; set(key: string, value: unknown, ttlMs: number): Promise<void>;
} }
function requestQuery(params: Record<string, string | undefined>): URLSearchParams {
const qs = new URLSearchParams();
for (const k of Object.keys(params).sort()) if (params[k] !== undefined) qs.set(k, params[k]!);
return qs;
}
/** 请求缓存键 `path?query`(参数按名排序、不含凭证);只读缓存的离线重算(MD-2)据此找已缓存的响应 */
export function tmdbRequestKey(path: string, params: Record<string, string | undefined> = {}): string {
return `${path}?${requestQuery(params)}`;
}
/** 空结果(搜索 0 条)用短 TTL(对照 MP EMPTY_RESULT_CACHE_TTL) */ /** 空结果(搜索 0 条)用短 TTL(对照 MP EMPTY_RESULT_CACHE_TTL) */
const isEmptyResult = (v: unknown) => const isEmptyResult = (v: unknown) =>
Array.isArray((v as Json | null)?.results) && ((v as Json).results as unknown[]).length === 0; Array.isArray((v as Json | null)?.results) && ((v as Json).results as unknown[]).length === 0;
@@ -163,8 +174,7 @@ export class TmdbClient {
} }
private async get<T>(path: string, params: Record<string, string | undefined>): Promise<T> { private async get<T>(path: string, params: Record<string, string | undefined>): Promise<T> {
const qs = new URLSearchParams(); const qs = requestQuery(params);
for (const k of Object.keys(params).sort()) if (params[k] !== undefined) qs.set(k, params[k]!);
const key = `${path}?${qs}`; const key = `${path}?${qs}`;
const emptyTtl = this.cfg.cache?.emptyTtlMs ?? 30 * 60_000; const emptyTtl = this.cfg.cache?.emptyTtlMs ?? 30 * 60_000;
const baseTtl = this.cfg.cache?.ttlMs ?? 24 * 3600_000; const baseTtl = this.cfg.cache?.ttlMs ?? 24 * 3600_000;
+80 -2
View File
@@ -1,10 +1,11 @@
import { createMockTmdb, type MockTmdb } from "@umt/testkit"; import { createMockTmdb, HEART_SIGNAL_FIXTURE, type MockTmdb, TMDB_FIXTURES } from "@umt/testkit";
import { afterEach, describe, expect, it } from "vitest"; import { afterEach, describe, expect, it } from "vitest";
import { import {
createTmdbClient, createTmdbClient,
fetchRichDetail, fetchRichDetail,
fetchRichSeason, fetchRichSeason,
isPlaceholderEpisodeName, isPlaceholderEpisodeName,
overviewLanguages,
parseAltTitles, parseAltTitles,
parseTranslations, parseTranslations,
pickChineseTitle, pickChineseTitle,
@@ -19,7 +20,7 @@ afterEach(async () => {
}); });
async function setup() { async function setup() {
mock = await createMockTmdb(); mock = await createMockTmdb({ fixtures: [...TMDB_FIXTURES, HEART_SIGNAL_FIXTURE] });
const tmdb = createTmdbClient({ baseUrl: `${mock.url}/3`, apiKey: "test-tmdb-key", retries: 0, timeoutMs: 1000 }); const tmdb = createTmdbClient({ baseUrl: `${mock.url}/3`, apiKey: "test-tmdb-key", retries: 0, timeoutMs: 1000 });
return { mock, tmdb }; return { mock, tmdb };
} }
@@ -64,6 +65,36 @@ describe("语言回退链(纯函数)", () => {
expect(pickText(["ja-JP"], translations, "overview", { value: null, language: null })).toBeNull(); expect(pickText(["ja-JP"], translations, "overview", { value: null, language: null })).toBeNull();
}); });
it("MD-2 简介链:缺省只到中文变体;排在中文前面的语言保留;链里没有中文或允许回退时原链", () => {
expect(overviewLanguages(CHAIN)).toEqual(["zh-CN", "zh-SG", "zh-TW", "zh-HK"]);
expect(overviewLanguages(CHAIN, true)).toEqual(CHAIN);
expect(overviewLanguages(["en-US", "zh-CN", "ja-JP"])).toEqual(["en-US", "zh-CN"]);
expect(overviewLanguages(["ja-JP", "en-US"])).toEqual(["ja-JP", "en-US"]);
expect(overviewLanguages(["zh", "en-US"])).toEqual(["zh"]);
});
it("MD-2 简介只有英文:按中文链取不到就为空;标题仍回退到英文", () => {
const enOnly = parseTranslations(
{
translations: [
{ iso_639_1: "zh", iso_3166_1: "CN", data: { name: "", overview: "" } },
{ iso_639_1: "en", iso_3166_1: "US", data: { name: "Heart Signal", overview: "English only" } },
],
},
"name",
);
expect(pickText(overviewLanguages(CHAIN), enOnly, "overview")).toBeNull();
expect(pickText(CHAIN, enOnly, "overview")).toMatchObject({ value: "English only", language: "en-US" });
// 请求语言下的原值其实是英文译文(TMDB 回退):不当作 zh-CN
expect(
pickText(overviewLanguages(CHAIN), enOnly, "overview", { value: "English only", language: "zh-CN" }),
).toBeNull();
expect(pickTitle(CHAIN, { translations: enOnly, primary: null })).toMatchObject({
value: "Heart Signal",
language: "en-US",
});
});
it("占位集名识别", () => { it("占位集名识别", () => {
expect(isPlaceholderEpisodeName("第 3 集", 3)).toBe(true); expect(isPlaceholderEpisodeName("第 3 集", 3)).toBe(true);
expect(isPlaceholderEpisodeName("Episode 3", 3)).toBe(true); expect(isPlaceholderEpisodeName("Episode 3", 3)).toBe(true);
@@ -137,6 +168,53 @@ describe("fetchRichDetail / fetchRichSeason(mock TMDB)", () => {
expect(s2.fetchedLanguages).toEqual(["zh-CN"]); expect(s2.fetchedLanguages).toEqual(["zh-CN"]);
}); });
it("MD-2《心动的信号》:作品与每集简介只有英文 → 留空;标题照常回退;允许非中文时恢复英文", async () => {
const { tmdb, mock } = await setup();
const d = await fetchRichDetail(tmdb, "tv", 103147, CHAIN);
expect(d.title).toBe("心动的信号");
expect(d.overview).toBeNull();
expect(d.overviewLanguage).toBeNull();
const s = await fetchRichSeason(tmdb, 103147, 1, CHAIN, { now: new Date("2026-01-01") });
const [e1, e2, e3] = s.episodes;
expect(e1).toMatchObject({ name: "心动不如行动,大胆亮出底牌!", nameLanguage: "zh-CN", overview: null });
expect(e1!.overviewLanguage).toBeNull();
// 标题:zh-CN 是占位名,回退到 en-US;简介仍不回退
expect(e2).toMatchObject({ name: "Who Is the One?", nameLanguage: "en-US", overview: null });
expect(e3).toMatchObject({ name: "约会之夜", overview: "約會之夜。", overviewLanguage: "zh-TW" });
// 补拉 en-US 只为了标题
expect(s.fetchedLanguages).toEqual(CHAIN);
const any = await fetchRichSeason(tmdb, 103147, 1, CHAIN, {
now: new Date("2026-01-01"),
overviewNonChineseFallback: true,
});
expect(any.episodes[0]).toMatchObject({
overview: "If Your Heart Skips a Beat, Play Your Cards Openly!",
overviewLanguage: "en-US",
});
const anyD = await fetchRichDetail(tmdb, "tv", 103147, CHAIN, { overviewNonChineseFallback: true });
expect(anyD).toMatchObject({ overview: "A Chinese dating reality show.", overviewLanguage: "en-US" });
expect(mock.calls.length).toBeGreaterThan(0);
});
it("MD-2 标题都齐、只缺简介时,中文变体走完就停,不为简介补拉英文", async () => {
const fx = {
...HEART_SIGNAL_FIXTURE,
seasons: [
{
...HEART_SIGNAL_FIXTURE.seasons![0]!,
overviews: { "zh-CN": "第一季。" },
episodes: [HEART_SIGNAL_FIXTURE.seasons![0]!.episodes[0]!],
},
],
};
mock = await createMockTmdb({ fixtures: [fx] });
const tmdb = createTmdbClient({ baseUrl: `${mock.url}/3`, apiKey: "test-tmdb-key", retries: 0, timeoutMs: 1000 });
const s = await fetchRichSeason(tmdb, 103147, 1, CHAIN, { now: new Date("2026-01-01") });
expect(s.episodes[0]!.overview).toBeNull();
expect(s.fetchedLanguages).toEqual(["zh-CN", "zh-SG", "zh-TW", "zh-HK"]);
});
it("补拉走请求缓存:同一季第二次不再请求", async () => { it("补拉走请求缓存:同一季第二次不再请求", async () => {
const { tmdb, mock } = await setup(); const { tmdb, mock } = await setup();
await fetchRichSeason(tmdb, 95396, 1, CHAIN, { now: new Date("2026-01-01") }); await fetchRichSeason(tmdb, 95396, 1, CHAIN, { now: new Date("2026-01-01") });
+27 -9
View File
@@ -5,13 +5,14 @@
* get_tv_season_detail(季名、简介、海报、episodes[]:name/overview/air_date/runtime/still_path/vote_average)→ fetchRichSeason; * get_tv_season_detail(季名、简介、海报、episodes[]:name/overview/air_date/runtime/still_path/vote_average)→ fetchRichSeason;
* 请求缓存语义同 client.ts(MP tmdbv3api request 的 @cached:成功才缓存、空结果短 TTL)。 * 请求缓存语义同 client.ts(MP tmdbv3api request 的 @cached:成功才缓存、空结果短 TTL)。
* 调整:标题 / 简介按语言回退链取(translations.ts);季集在首选语言缺标题(占位名)或缺简介时按链逐个语言补拉, * 调整:标题 / 简介按语言回退链取(translations.ts);季集在首选语言缺标题(占位名)或缺简介时按链逐个语言补拉,
* 链走完或全部补齐即停。主创只留前 10 位演员 + 导演 / 主创(created_by),头像只存路径(前端经图片代理)。 * 链走完或全部补齐即停。MD-2:简介 / 标语只按 overviewLanguages(缺省只到中文变体,不回退英文)取,取不到留空。主创只留前 10 位演员 + 导演 / 主创(created_by),头像只存路径(前端经图片代理)。
*/ */
import type { MediaEntityType, TmdbClient } from "./client.ts"; import type { MediaEntityType, TmdbClient } from "./client.ts";
import { import {
DEFAULT_METADATA_LANGUAGES, DEFAULT_METADATA_LANGUAGES,
isPlaceholderEpisodeName, isPlaceholderEpisodeName,
isPlaceholderSeasonName, isPlaceholderSeasonName,
overviewLanguages,
parseAltTitles, parseAltTitles,
parseTranslations, parseTranslations,
pickChineseTitle, pickChineseTitle,
@@ -116,6 +117,12 @@ export interface TmdbRichSeasonDetail {
fetchedLanguages: string[]; fetchedLanguages: string[];
} }
/** MD-2 取值选项 */
export interface RichTextOptions {
/** 简介 / 标语是否允许回退到非中文(设置 metadata.overviewNonChineseFallback,缺省否) */
overviewNonChineseFallback?: boolean;
}
const chainOf = (languages?: readonly string[]) => const chainOf = (languages?: readonly string[]) =>
languages?.length ? [...languages] : [...DEFAULT_METADATA_LANGUAGES]; languages?.length ? [...languages] : [...DEFAULT_METADATA_LANGUAGES];
@@ -156,8 +163,10 @@ export async function fetchRichDetail(
type: MediaEntityType, type: MediaEntityType,
id: string | number, id: string | number,
languages?: readonly string[], languages?: readonly string[],
opts: RichTextOptions = {},
): Promise<TmdbRichDetail> { ): Promise<TmdbRichDetail> {
const chain = chainOf(languages); const chain = chainOf(languages);
const textChain = overviewLanguages(chain, opts.overviewNonChineseFallback);
const j = await client.getJson<Json>(`/${type}/${id}`, { const j = await client.getJson<Json>(`/${type}/${id}`, {
language: chain[0], language: chain[0],
append_to_response: "alternative_titles,credits,external_ids,translations", append_to_response: "alternative_titles,credits,external_ids,translations",
@@ -171,8 +180,9 @@ export async function fetchRichDetail(
const titleSrc = { translations, altTitles, originalTitle, originalLanguage, primary }; const titleSrc = { translations, altTitles, originalTitle, originalLanguage, primary };
const title = pickTitle(chain, titleSrc); const title = pickTitle(chain, titleSrc);
const zh = pickChineseTitle(chain, titleSrc); const zh = pickChineseTitle(chain, titleSrc);
const overview = pickText(chain, translations, "overview", { value: s(j.overview), language: chain[0] ?? null }); // 请求语言 = 链首,链首总在 textChain 里(见 overviewLanguages),所以 primary 可以兜底
const tagline = pickText(chain, translations, "tagline", { value: s(j.tagline), language: chain[0] ?? null }); const overview = pickText(textChain, translations, "overview", { value: s(j.overview), language: chain[0] ?? null });
const tagline = pickText(textChain, translations, "tagline", { value: s(j.tagline), language: chain[0] ?? null });
const date = s(type === "movie" ? j.release_date : j.first_air_date); const date = s(type === "movie" ? j.release_date : j.first_air_date);
const ext = (j.external_ids ?? {}) as Json; const ext = (j.external_ids ?? {}) as Json;
const runtimes = Array.isArray(j.episode_run_time) ? (j.episode_run_time as unknown[]) : []; const runtimes = Array.isArray(j.episode_run_time) ? (j.episode_run_time as unknown[]) : [];
@@ -258,10 +268,12 @@ export async function fetchRichSeason(
tvId: string | number, tvId: string | number,
seasonNumber: number, seasonNumber: number,
languages?: readonly string[], languages?: readonly string[],
opts: { now?: Date } = {}, opts: { now?: Date } & RichTextOptions = {},
): Promise<TmdbRichSeasonDetail> { ): Promise<TmdbRichSeasonDetail> {
const chain = chainOf(languages); const chain = chainOf(languages);
const first = chain[0]!; const first = chain[0]!;
// MD-2:简介只从这些语言取(缺省只到中文变体);标题仍按完整链
const textChain = overviewLanguages(chain, opts.overviewNonChineseFallback);
const j = await client.getJson<Json>(`/tv/${tvId}/season/${seasonNumber}`, { language: first }); const j = await client.getJson<Json>(`/tv/${tvId}/season/${seasonNumber}`, { language: first });
const rawEps = (Array.isArray(j.episodes) ? j.episodes : []) as Json[]; const rawEps = (Array.isArray(j.episodes) ? j.episodes : []) as Json[];
const eps = rawEps.map((e) => episodeOf(e, seasonNumber, first)); const eps = rawEps.map((e) => episodeOf(e, seasonNumber, first));
@@ -274,9 +286,15 @@ export async function fetchRichSeason(
// 只为已播出的集补拉(未播出的集各语言通常都空,补拉只会白白耗请求);季名"第 N 季"各语言都是占位,不触发补拉 // 只为已播出的集补拉(未播出的集各语言通常都空,补拉只会白白耗请求);季名"第 N 季"各语言都是占位,不触发补拉
const today = (opts.now ?? new Date()).toISOString().slice(0, 10); const today = (opts.now ?? new Date()).toISOString().slice(0, 10);
const aired = (e: TmdbRichEpisode) => !!e.airDate && e.airDate <= today; const aired = (e: TmdbRichEpisode) => !!e.airDate && e.airDate <= today;
const missing = () => eps.some((e) => aired(e) && (!e.name || !e.overview)) || (!seasonOverview && eps.some(aired)); // 缺简介只在剩下的语言里还有可取简介的语言时才值得补拉
for (const lang of chain.slice(1)) { const missing = (rest: readonly string[]) => {
if (!eps.length || !missing()) break; const ov = rest.some((l) => textChain.includes(l));
return eps.some((e) => aired(e) && (!e.name || (ov && !e.overview))) || (ov && !seasonOverview && eps.some(aired));
};
for (const [i, lang] of chain.entries()) {
if (i === 0) continue;
if (!eps.length || !missing(chain.slice(i))) break;
const textOk = textChain.includes(lang);
let k: Json; let k: Json;
try { try {
k = await client.getJson<Json>(`/tv/${tvId}/season/${seasonNumber}`, { language: lang }); k = await client.getJson<Json>(`/tv/${tvId}/season/${seasonNumber}`, { language: lang });
@@ -296,12 +314,12 @@ export async function fetchRichSeason(
e.name = other.name; e.name = other.name;
e.nameLanguage = lang; e.nameLanguage = lang;
} }
if (!e.overview && other.overview) { if (textOk && !e.overview && other.overview) {
e.overview = other.overview; e.overview = other.overview;
e.overviewLanguage = lang; e.overviewLanguage = lang;
} }
} }
if (!seasonOverview && s(k.overview)) { if (textOk && !seasonOverview && s(k.overview)) {
seasonOverview = s(k.overview); seasonOverview = s(k.overview);
seasonOverviewLang = lang; seasonOverviewLang = lang;
} }
+23 -2
View File
@@ -147,7 +147,26 @@ export function pickChineseTitle(chain: readonly string[], src: TitleSources): P
return null; return null;
} }
/** 简介 / 标语:按链取 translations 的非空值;全部落空时用 primary(请求语言下的原值) */ /** 语言标签是否中文(zh / zh-CN / zh-TW …) */
export const isChineseTag = (tag: string): boolean => splitTag(tag).lang === "zh";
/**
* MD-2 简介 / 标语用的语言链:缺省只回退到中文变体,不回退到非中文(TMDB 上不少剧集没有中文简介,
* 英文"简介"常是标题英译,中文界面里显得不对)。规则:
* - allowNonChinese = true(设置 metadata.overviewNonChineseFallback):原链不变;
* - 链里没有中文:原链不变(用户本来就选了非中文);
* - 否则:第一个中文之前的语言(用户排在中文前面、是首选而不是回退)+ 链中的全部中文,之后的非中文去掉。
* 缺省链 zh-CN → zh-SG → zh-TW → zh-HK → en-US 得到 zh-CN → zh-SG → zh-TW → zh-HK。
* 标题 / 别名不受影响,仍按完整链回退。MP 对照:tmdbapi 每集信息只按配置语言取,简介为空就为空。
*/
export function overviewLanguages(chain: readonly string[], allowNonChinese = false): string[] {
if (allowNonChinese) return [...chain];
const firstZh = chain.findIndex(isChineseTag);
if (firstZh < 0) return [...chain];
return chain.filter((t, i) => i < firstZh || isChineseTag(t));
}
/** 简介 / 标语:按链取 translations 的非空值;全部落空时用 primary(请求语言下的原值,等于链外语言译文时不用) */
export function pickText( export function pickText(
chain: readonly string[], chain: readonly string[],
translations: readonly TmdbTranslation[], translations: readonly TmdbTranslation[],
@@ -158,7 +177,9 @@ export function pickText(
const tr = translations.find((t) => tagMatches(tag, t) && t[key]); const tr = translations.find((t) => tagMatches(tag, t) && t[key]);
if (tr?.[key]) return { value: tr[key]!, language: tag, via: "translation" }; if (tr?.[key]) return { value: tr[key]!, language: tag, via: "translation" };
} }
if (primary?.value) return { value: primary.value, language: primary.language, via: "primary" }; // 请求语言下的原值若正是链外某语言的译文(TMDB / 反代把缺译名的字段回退成了别的语言),不当作链内的值
const foreign = translations.some((t) => t[key] === primary?.value && !chain.some((tag) => tagMatches(tag, t)));
if (primary?.value && !foreign) return { value: primary.value, language: primary.language, via: "primary" };
return null; return null;
} }
@@ -0,0 +1,76 @@
/**
* 每集简介差分(MD-2):MP `TmdbApi.get_tv_season_detail`(e6820a8,app/modules/themoviedb/tmdbapi.py)只按配置语言
* (TMDB_LOCALE = zh)请求季详情,episodes[].overview 原样使用,为空就为空;本项目 `fetchRichSeason` 按语言链补拉。
* 响应按《心动的信号》(TMDB tv/103147)各语言季详情的真实形态裁剪(每期标题有 zh-CN,简介只有 en-US)。
* 期望:简介与 MP 一致(中文变体都没有 → 空);标题允许按链回退(登记偏离 intentional:MD-1 占位集名按链补标题)。
*/
import { describe, expect, it } from "vitest";
import { fetchRichSeason } from "../../src/tmdb/index.ts";
type Ep = { episode_number: number; air_date: string; name: string; overview: string };
const season = (episodes: Ep[]) => ({
season_number: 1,
name: "第 1 季",
overview: "",
air_date: "2018-08-01",
episodes,
});
/** 各语言的 /tv/103147/season/1 响应(裁剪) */
const RECORDED: Record<string, ReturnType<typeof season>> = {
"zh-CN": season([
{ episode_number: 1, air_date: "2018-08-01", name: "心动不如行动,大胆亮出底牌!", overview: "" },
{ episode_number: 2, air_date: "2018-08-08", name: "第 2 集", overview: "" },
]),
"zh-SG": season([
{ episode_number: 1, air_date: "2018-08-01", name: "第 1 集", overview: "" },
{ episode_number: 2, air_date: "2018-08-08", name: "第 2 集", overview: "" },
]),
"zh-TW": season([
{ episode_number: 1, air_date: "2018-08-01", name: "第 1 集", overview: "" },
{ episode_number: 2, air_date: "2018-08-08", name: "第 2 集", overview: "" },
]),
"zh-HK": season([
{ episode_number: 1, air_date: "2018-08-01", name: "第 1 集", overview: "" },
{ episode_number: 2, air_date: "2018-08-08", name: "第 2 集", overview: "" },
]),
"en-US": season([
{
episode_number: 1,
air_date: "2018-08-01",
name: "Episode 1",
overview: "If Your Heart Skips a Beat, Play Your Cards Openly!",
},
{ episode_number: 2, air_date: "2018-08-08", name: "Who Is the One?", overview: "Who Is the One?" },
]),
};
const client = {
getJson: <T>(_path: string, params: Record<string, string | undefined> = {}) =>
Promise.resolve(RECORDED[params.language ?? "en-US"] as T),
};
/** MP 侧:只取配置语言(zh-CN)的季详情,集名 / 简介原样(空串 = 没有) */
function mpEpisodes(lang = "zh-CN") {
return RECORDED[lang]!.episodes.map((e) => ({ ep: e.episode_number, name: e.name, overview: e.overview || null }));
}
const CHAIN = ["zh-CN", "zh-SG", "zh-TW", "zh-HK", "en-US"];
const DEVIATIONS: Record<string, { kind: "intentional"; note: string }> = {
"E2.name": { kind: "intentional", note: "MD-1:zh-CN 是占位名「第 2 集」时按链补标题(可到 en-US)" },
};
describe("每集简介差分(MP get_tv_season_detail ↔ fetchRichSeason)", () => {
it("简介与 MP 一致:中文变体都没有时为空,不回退英文;标题偏离已登记", async () => {
const ours = await fetchRichSeason(client, 103147, 1, CHAIN, { now: new Date("2026-01-01") });
const mismatches: string[] = [];
for (const mp of mpEpisodes()) {
const e = ours.episodes.find((x) => x.episodeNumber === mp.ep)!;
if (e.overview !== mp.overview) mismatches.push(`E${mp.ep}.overview`);
if (e.name !== mp.name) mismatches.push(`E${mp.ep}.name`);
}
expect(mismatches.filter((m) => !DEVIATIONS[m])).toEqual([]);
expect(Object.keys(DEVIATIONS).every((k) => mismatches.includes(k))).toBe(true);
expect(ours.episodes.map((e) => e.overview)).toEqual([null, null]);
});
});
+7
View File
@@ -246,6 +246,11 @@ export const MetadataSettings = z.object({
.min(1) .min(1)
.max(8) .max(8)
.default(["zh-CN", "zh-SG", "zh-TW", "zh-HK", "en-US"]), .default(["zh-CN", "zh-SG", "zh-TW", "zh-HK", "en-US"]),
/**
* MD-2:简介 / 标语是否允许回退到非中文。缺省 false = 只按链里的中文变体(及排在第一个中文之前的语言)取,
* 取不到留空;标题不受影响。链里没有中文时不起作用
*/
overviewNonChineseFallback: z.boolean().default(false),
/** /**
* DH-1 季 0(特别篇):true = full 刷新与元数据回刷在 TMDB 有季 0 时一并拉取每集信息并建集单元(对照 MP); * DH-1 季 0(特别篇):true = full 刷新与元数据回刷在 TMDB 有季 0 时一并拉取每集信息并建集单元(对照 MP);
* 季 0 不计入正片季总集数与订阅完成判定 * 季 0 不计入正片季总集数与订阅完成判定
@@ -649,6 +654,7 @@ export const DEFAULT_SETTINGS: GlobalSettings = {
tmdbApiKeySecretRef: null, tmdbApiKeySecretRef: null,
tmdbBaseUrl: null, tmdbBaseUrl: null,
languages: ["zh-CN", "zh-SG", "zh-TW", "zh-HK", "en-US"], languages: ["zh-CN", "zh-SG", "zh-TW", "zh-HK", "en-US"],
overviewNonChineseFallback: false,
includeSpecials: true, includeSpecials: true,
}, },
sites: { sites: {
@@ -954,6 +960,7 @@ export const SETTINGS_SECTIONS: readonly SettingsSectionDef[] = [
"metadata.tmdbApiKeySecretRef", "metadata.tmdbApiKeySecretRef",
"metadata.tmdbBaseUrl", "metadata.tmdbBaseUrl",
"metadata.languages", "metadata.languages",
"metadata.overviewNonChineseFallback",
"metadata.includeSpecials", "metadata.includeSpecials",
"cache.tmdbMaxSize", "cache.tmdbMaxSize",
"identity.eagerTmdbLookup", "identity.eagerTmdbLookup",
+46
View File
@@ -328,3 +328,49 @@ export const TMDB_FIXTURES: TmdbFixture[] = [
], ],
}, },
]; ];
/**
* MD-2:《心动的信号》(TMDB tv/103147 的真实形态):作品与每期标题有 zh-CN,简介只有 en-US(实为标题英译)。
* 不在 TMDB_FIXTURES 里,用到的测试自行拼进 fixtures。
*/
export const HEART_SIGNAL_FIXTURE: TmdbFixture = {
type: "tv",
id: 103147,
original_title: "心动的信号",
original_language: "zh",
origin_country: ["CN"],
date: "2018-08-01",
titles: { "zh-CN": "心动的信号", "en-US": "Heart Signal" },
overviews: { "en-US": "A Chinese dating reality show." },
popularity: 6.2,
alternative_titles: [],
external_ids: { imdb_id: null, tvdb_id: null },
rich: { genres: ["真人秀"], status: "Returning Series" },
seasons: [
{
season_number: 1,
air_date: "2018-08-01",
names: { "zh-CN": "第 1 季", "en-US": "Season 1" },
episodes: [
{
episode_number: 1,
air_date: "2018-08-01",
names: { "zh-CN": "心动不如行动,大胆亮出底牌!", "en-US": "Episode 1" },
overviews: { "en-US": "If Your Heart Skips a Beat, Play Your Cards Openly!" },
},
{
episode_number: 2,
air_date: "2018-08-08",
names: { "zh-CN": "第 2 集", "en-US": "Who Is the One?" },
overviews: { "en-US": "Who Is the One?" },
},
{
episode_number: 3,
air_date: "2018-08-15",
names: { "zh-CN": "约会之夜", "en-US": "Date Night" },
overviews: { "zh-TW": "約會之夜。", "en-US": "Date Night." },
},
],
},
],
};
+94
View File
@@ -0,0 +1,94 @@
// 存量简介重算(MD-2):简介的语言回退只到中文变体后,把之前回退到英文等非中文的作品 / 季 / 集简介按新规则重算。
// 不请求 TMDB,只读 kv_cache(tmdb.request)里已缓存的 TMDB 响应;缓存里没有的按"当时中文为空"清空。
// 默认只读(dry-run),加 --apply 才改库(单个事务,写一条审计 metadata.overview_recompute)。
//
// 用法(DATABASE_URL 必须显式给出,脚本不带默认库):
// DATABASE_URL=postgres://umt:***@<host>:5432/umt deno run -A scripts/overview-recompute.ts # 看清单
// DATABASE_URL=… deno run -A scripts/overview-recompute.ts --apply # 写库
// DATABASE_URL=… deno run -A scripts/overview-recompute.ts --entity <作品 id> # 只看 / 只处理一部作品
// DATABASE_URL=… deno run -A scripts/overview-recompute.ts --json # 清单输出 JSON
//
// 语言链与"简介是否允许回退到非中文"取库里的设置 metadata.languages / metadata.overviewNonChineseFallback。
// 规则与实现见 apps/server/src/modules/sources-identity/service/overview-recompute.ts、docs/runbook/metadata.md。
import { DEFAULT_METADATA_LANGUAGES } from "@umt/adapters";
import type { DB } from "@umt/db";
import { Kysely, PostgresDialect, sql } from "kysely";
import pg from "pg";
import {
applyOverviewRecompute,
type OverviewChange,
planOverviewRecompute,
} from "../apps/server/src/modules/sources-identity/service/overview-recompute.ts";
const argv = [...Deno.args];
const APPLY = argv.includes("--apply");
const JSON_OUT = argv.includes("--json");
if (APPLY && argv.includes("--dry-run")) {
console.error("--apply 与 --dry-run 不能同时给出");
Deno.exit(2);
}
const entityArg = argv.indexOf("--entity");
const ONLY_ENTITY = entityArg >= 0 ? (argv[entityArg + 1] ?? null) : null;
const url = Deno.env.get("DATABASE_URL");
if (!url) {
console.error("需要 DATABASE_URL(脚本不带默认库)");
Deno.exit(2);
}
const db = new Kysely<DB>({ dialect: new PostgresDialect({ pool: new pg.Pool({ connectionString: url, max: 2 }) }) });
const meta = (
await sql<{ value: { languages?: string[]; overviewNonChineseFallback?: boolean } | null }>`
select value from settings where key = 'metadata'`.execute(db)
).rows[0]?.value;
const languages = meta?.languages?.length ? meta.languages : [...DEFAULT_METADATA_LANGUAGES];
const overviewNonChineseFallback = meta?.overviewNonChineseFallback === true;
const now = new Date();
const cut = (s: string | null, n = 40) => (s === null ? "(空)" : s.length > n ? `${s.slice(0, n)}…` : s);
const where = (c: OverviewChange) =>
c.target === "entity"
? "作品"
: c.target === "season"
? `S${String(c.season).padStart(2, "0")}`
: `S${String(c.season).padStart(2, "0")}E${String(c.episode).padStart(2, "0")}`;
try {
const plan = await planOverviewRecompute(db, { languages, overviewNonChineseFallback, now, entityId: ONLY_ENTITY });
if (JSON_OUT && !APPLY) {
console.log(JSON.stringify(plan, null, 2));
} else {
console.log(
`# overview-recompute (${APPLY ? "APPLY" : "DRY-RUN"})${ONLY_ENTITY ? ` --entity ${ONLY_ENTITY}` : ""}`,
);
console.log(`语言链 ${languages.join(" → ")};简介链 ${plan.overviewLanguages.join(" → ")}`);
console.log(
`非中文简介:作品 ${plan.scanned.entities},季 / 集 ${plan.scanned.units};读缓存 ${plan.cache.reads} 次,未命中 ${plan.cache.misses}\n`,
);
const byEntity = new Map<string, OverviewChange[]>();
for (const c of plan.changes) byEntity.set(c.mediaEntityId, [...(byEntity.get(c.mediaEntityId) ?? []), c]);
for (const [id, list] of byEntity) {
console.log(`## ${list[0]!.entityTitle} (${id}) ${list.length} 项`);
for (const c of list) {
const basis = c.basis === "cache" ? "缓存" : "无缓存→清空";
console.log(
` ${where(c).padEnd(8)} ${c.field.padEnd(8)} [${c.from.language ?? "-"}] ${cut(c.from.value)} → [${c.to.language ?? "-"}] ${cut(c.to.value)} (${basis})`,
);
}
}
const count = (t: OverviewChange["target"]) => plan.changes.filter((c) => c.target === t).length;
console.log(
`\n合计 ${plan.changes.length} 项:作品 ${count("entity")},季 ${count("season")},集 ${count("episode")};` +
`其中无缓存清空 ${plan.changes.filter((c) => c.basis === "no_cache").length},` +
`改成中文 ${plan.changes.filter((c) => c.to.value !== null).length}`,
);
}
if (APPLY && plan.changes.length) {
const r = await db.transaction().execute((trx) => applyOverviewRecompute(trx, plan, { now }));
console.log(`已写库:最终值有变化的作品 ${r.entities},季 / 集 ${r.units}(锁定 / 覆盖的字段不变)`);
} else if (!APPLY) {
console.log("dry-run:未改库。确认清单后加 --apply 执行");
}
} finally {
await db.destroy();
}