diff --git a/public-api.js b/public-api.js index 1286b64..8585441 100644 --- a/public-api.js +++ b/public-api.js @@ -660,7 +660,9 @@ const attrsOf = (s) => { const a = {}; String(s || '').replace(/([\w:-]+)\s*(?:= function parseHtml(html) { const root = { tag: '#root', attrs: {}, kids: [] }; const stack = [root]; - const re = /|<(script|style)\b[^>]*>[\s\S]*?<\/\1>|<\/([a-zA-Z][\w-]*)\s*>|<([a-zA-Z][\w-]*)([^>]*?)(\/?)>|([^<]+)/g; + // 标签名允许带冒号:从 Word 粘贴到公众号的文章里有 、 这类标签,以前没认出来,"o:p>" 会漏进正文 + // 、 直接跳过;落单的 "<"(比如正文里的 "电压<10V")按普通文字保留 + const re = /|<(script|style)\b[^>]*>[\s\S]*?<\/\1>|<[!?][^>]*>|<\/([a-zA-Z][\w:.-]*)\s*>|<([a-zA-Z][\w:.-]*)([^>]*?)(\/?)>|([^<]+)| { const m = html.match(new RegExp(`var\\s+${name}\\s*=\\s*(?:htmlDecode\\()?["']([^"']*)["']`)); return m ? decodeEnt(m[1]) : ''; }; const metaOf = (html, prop) => { const m = html.match(new RegExp(`]+(?:property|name)=["']${prop}["'][^>]+content=["']([^"']*)["']`, 'i')) || html.match(new RegExp(`]+content=["']([^"']*)["'][^>]+(?:property|name)=["']${prop}["']`, 'i')); return m ? decodeEnt(m[1]) : ''; }; +// Office 标签残留(o:p> /o:p> 等):解析器已修好;这里再兜底清理一次,旧数据发布时也会被清掉 +const OFFICE_JUNK = /\n]{0,80}?>/g; +const cleanText = (s) => String(s || '').replace(OFFICE_JUNK, '').replace(/[ \t]{2,}/g, ' ').trim(); // 正文树 → 块。文字按块级元素分段;图片单独成块;连续图片合并成两图 / 图集 const NOISE = /^(点击|长按|扫码|识别二维码|关注我们|欢迎关注|点个在看|阅读原文|▼|▲|↓|END$|end$)|^[—\-·•\s]+$|^(公众号|微信号|往期推荐|往期回顾|推荐阅读|分享|收藏|点赞|在看)[::]?$|关注我们$|二维码/; function toBlocks(root) { const out = []; let buf = '', bold = 0, all = 0, inQuote = false, center = false; const flush = (forceTag) => { - const text = buf.replace(/[ \t\r\f\v]+/g, ' ').replace(/\s*\n\s*/g, '\n').trim(); + const text = buf.replace(OFFICE_JUNK, '').replace(/[ \t\r\f\v]+/g, ' ').replace(/\s*\n\s*/g, '\n').trim(); if (text && !NOISE.test(text)) { const isHead = forceTag === 'h' || (all > 0 && bold / all > 0.85 && text.length <= 40 && !/[。!?.!?]$/.test(text)); out.push({ t: forceTag === 'quote' || inQuote ? 'quote' : isHead ? 'h' : 'p', zh: text, ...(center && text.length <= 50 ? { center: true } : {}) }); @@ -821,7 +827,7 @@ async function translateAny(ai, strings) { } const NEWS_CATS = ['projects', 'products', 'events', 'company']; const newsSlugify = (s) => String(s || '').toLowerCase().normalize('NFKD').replace(/[^\w\s-]/g, '').trim().replace(/[\s_]+/g, '-').replace(/-+/g, '-').slice(0, 70).replace(/-$/, ''); -const srcText = (o) => String((o && (o.zh || o.en)) || '').trim(); +const srcText = (o) => cleanText((o && (o.zh || o.en)) || ''); const articleTexts = (n) => [srcText(n.title), srcText(n.excerpt), srcText(n.place), ...(Array.isArray(n.blocks) ? n.blocks : []).flatMap((b) => [srcText(b), srcText(b.cap), ...(Array.isArray(b.caps) ? b.caps.map(srcText) : [])])]; const articleFiles = (n) => [n.cover, ...(Array.isArray(n.blocks) ? n.blocks : []).flatMap((b) => (b.file ? [b.file] : Array.isArray(b.files) ? b.files : []))].filter(Boolean).map(String); @@ -953,7 +959,7 @@ module.exports = function mountPublicApi(app, { dataFile, uploadsDir, requireAut const at = await translateAny(ai, arts.flatMap(articleTexts)); const L = (o) => { const src = srcText(o); const r = {}; - for (const lang of Object.keys(ZH_TARGETS)) r[lang] = String((o && lang !== 'zh' && o[lang]) || '').trim() || (src && at[lang][src]) || (hasZh(src) ? '' : src); + for (const lang of Object.keys(ZH_TARGETS)) r[lang] = cleanText((o && lang !== 'zh' && o[lang]) || '') || (src && at[lang][src]) || (hasZh(src) ? '' : src); return r; }; const usedSlugs = new Set();