feedkeeper 0.26.5 → 0.26.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "feedkeeper",
-  "version": "0.26.5",
+  "version": "0.26.6",
   "description": "A lightweight, self-hosted RSS reader and remote MCP server for your news, blogs, and AI workflows.",
   "keywords": [
     "rss",
@@ -47,7 +47,7 @@
     "express-rate-limit": "^8.7.1",
     "fast-xml-parser": "^5.11.2",
     "helmet": "^8.3.0",
-    "ipaddr.js": "^1.9.1",
+    "ipaddr.js": "^2.5.0",
     "jsdom": "^30.1.2",
     "node-cron": "^4.6.0",
     "otpauth": "^9.5.2",

package/server/dist/feeds/contentNoise.js

@@ -45,6 +45,23 @@ const CONSENT_COMPANIONS = [
     /\bInhalte? (von \S+ )?erlauben\b/i, /-Inhalte erlauben\b/i, /^Ich bin damit einverstanden, dass mir .+ angezeigt werden/i,
     /Personenbezogene Daten können an Drittplattformen/i, /^Link zum .+-Inhalt$/i, /^Mehr dazu in unserer Datenschutzerklärung/i,
 ];
+const MONTHS = "Januar|Februar|März|Maerz|April|Mai|Juni|Juli|August|September|Oktober|November|Dezember|January|February|March|June|July|October|December|Jan|Feb|Mar|Apr|Jun|Jul|Aug|Sep|Sept|Oct|Nov|Dec|Jan\\.|Feb\\.";
+const TIME_OF_DAY = "(?:[,\\s•·|-]+\\d{1,2}[:.]\\d{2}(?::\\d{2})?(?:\\s?(?:Uhr|[ap]\\.?m\\.?))?)?";
+/** A whole line that is a date, with or without a time: `15. Oktober 2026, 10:51 Uhr`, `Oct 15, 2026`, `15.10.2026`. */
+const DATE_ONLY = new RegExp(`^(?:\\d{1,2}\\.\\s?\\d{1,2}\\.\\s?\\d{2,4}|\\d{1,2}\\.?\\s(?:${MONTHS})\\.?\\s\\d{4}|(?:${MONTHS})\\.?\\s\\d{1,2},?\\s\\d{4})${TIME_OF_DAY}$`, "iu");
+/** A whole line that is a reading time: `2 min`, `5 Min. Lesezeit`, `Lesezeit: 3 Minuten`, `4 min read`. */
+const READING_TIME_ONLY = /^(?:Lesezeit:?\s*)?\d{1,3}\s*(?:min\.?|minuten?|minutes?)(?:\s*(?:Lesezeit|read|lesen))?$/i;
+/** JSON-LD and similar: an object with a schema context or type, standing as text. */
+function isStructuredData(text) {
+    if (text.length < 20 || !text.startsWith("{") || !/"@(context|type)"/.test(text))
+        return false;
+    try {
+        return typeof JSON.parse(text) === "object";
+    }
+    catch {
+        return false;
+    }
+}
 function removeReaderNoise(root, title, stats) {
     let changed = false;
     const drop = (element, rule) => {
@@ -73,6 +90,16 @@ function removeReaderNoise(root, title, stats) {
         if (onlyLinks)
             drop(list, "breadcrumb");
     }
+    // Structured data is for machines. Frameworks render it as an element of any name (`<component type="application/ld+json">`),
+    // and a stray block may hold nothing else: neither is text for a reader.
+    for (const element of [...root.querySelectorAll("[type]")]) {
+        if (/^application\/([\w.+-]*\+)?json(;.*)?$/i.test(element.getAttribute("type") ?? ""))
+            drop(element, "structured-data");
+    }
+    for (const block of leafBlocks(root)) {
+        if (block.parentNode && isStructuredData(flat(block.textContent)))
+            drop(block, "structured-data");
+    }
     const headline = normalized(title ?? "");
     const seen = new Set();
     let leading = 0;
@@ -98,6 +125,17 @@ function removeReaderNoise(root, title, stats) {
                 continue;
             }
         }
+        // Near the start, a line that is only a date or a reading time repeats what the reader shows in its own head.
+        if (text.length <= 60 && charsBefore(root, block) <= 800) {
+            if (DATE_ONLY.test(text)) {
+                drop(block, "dateline");
+                continue;
+            }
+            if (READING_TIME_ONLY.test(text)) {
+                drop(block, "reading-time");
+                continue;
+            }
+        }
         if (headline && normalized(text) === headline && (position < 4 || (heading && position < 14))) {
             drop(block, "title-repeat");
             continue;
@@ -245,9 +283,10 @@ export function cleanReaderHtml(html, title, stats = {}) {
         return html;
     const root = JSDOM.fragment(html);
     const images = resolveCustomImages(root, stats);
+    const variants = collapseImageVariants(root, stats);
     const noise = removeReaderNoise(root, title, stats);
     const typography = removeTypography(root, stats);
-    return images || noise || typography ? serialize(root) : html;
+    return images || variants || noise || typography ? serialize(root) : html;
 }
 /**
  * Some publishers wrap a picture in their own element and let a script fill the image: `<a-img src="/imgs/…">`
@@ -276,6 +315,31 @@ function resolveCustomImages(root, stats) {
         stats["custom-image"] = (stats["custom-image"] ?? 0) + 1;
     return changed;
 }
+/**
+ * Publishers put one badge in two colour schemes into a link (`<a><img alt="X"><img alt="X"></a>`) and let their
+ * stylesheet show one of them. A reader has no such stylesheet, so both stand there at full size. Images that carry
+ * the same label inside one link and no other text are variants of one thing: the link keeps its label as text, or,
+ * without a label, its first image. The call to action itself stays.
+ */
+function collapseImageVariants(root, stats) {
+    let changed = false;
+    for (const anchor of [...root.querySelectorAll("a")]) {
+        const images = [...anchor.querySelectorAll("img")];
+        if (images.length < 2 || flat(anchor.textContent) !== "")
+            continue;
+        const labels = new Set(images.map(image => flat(image.getAttribute("alt"))));
+        if (labels.size !== 1)
+            continue;
+        const [label] = labels;
+        if (label)
+            anchor.replaceChildren(anchor.ownerDocument.createTextNode(label));
+        else
+            images.slice(1).forEach(image => image.remove());
+        stats["image-variants"] = (stats["image-variants"] ?? 0) + 1;
+        changed = true;
+    }
+    return changed;
+}
 const CREDIT_PATTERNS = [
     /^(©|\(c\)\s|copyright\b)/i,
     /^[([]?(Bild|Foto|Fotos|Bildquelle|Bildrechte|Bildnachweis|Titelbild|Beitragsbild|Photo|Image|Credits?|Grafik|Illustration)[)\]]?\s*[:/|]/i,

package/server/dist/feeds/heroImage.js

@@ -0,0 +1,81 @@
+import { decodeEntities } from "./text.js";
+/** The widest picture a hero needs; a still larger variant is not worth the bytes. */
+const MAX_HERO_WIDTH = 2000;
+/** A larger variant must be clearly larger than the feed's picture, so a small difference does not change anything. */
+const MIN_GAIN = 1.5;
+/**
+ * Splits an image address into what identifies the picture and the width of this size. Publishers name sizes in the
+ * file name (`…__wide__148`, `-300x200.jpg`, `_300x200.jpg`) or in the query (`?w=300`); two addresses with the same
+ * key are the same picture in different sizes. Addresses without a recognised size have no variant.
+ */
+export function imageVariant(raw) {
+    let url;
+    try {
+        url = new URL(raw);
+    }
+    catch {
+        return null;
+    }
+    const path = url.pathname;
+    const named = /^(.*__[a-z]+__)(\d{2,4})$/i.exec(path) ?? /^(.*[-_])(\d{2,4})x\d{2,4}(\.[a-z0-9]{2,5})?$/i.exec(path);
+    if (named) {
+        // The query of such addresses is a cache version (`?v=…`), not part of the picture.
+        return { key: `${url.origin}${named[1]}#${named[3] ?? ""}`, width: Number(named[2]) };
+    }
+    const widthParam = ["w", "width"].find((name) => /^\d{2,4}$/.test(url.searchParams.get(name) ?? ""));
+    if (widthParam) {
+        const rest = new URLSearchParams(url.searchParams);
+        for (const name of ["w", "width", "h", "height", "q", "quality", "fit", "dpr"])
+            rest.delete(name);
+        return { key: `${url.origin}${path}?${rest.toString()}`, width: Number(url.searchParams.get(widthParam)) };
+    }
+    return null;
+}
+/** Every image address in the HTML: `src` of images and each entry of a `srcset`. */
+function imageAddresses(html, baseUrl) {
+    const found = new Set();
+    const add = (raw) => {
+        const value = decodeEntities(raw.trim());
+        if (!value || value.startsWith("data:"))
+            return;
+        try {
+            const url = new URL(value, baseUrl ?? undefined);
+            if (url.protocol === "https:" || url.protocol === "http:")
+                found.add(url.toString());
+        }
+        catch { /* A broken address is not a candidate. */ }
+    };
+    for (const [tag] of html.matchAll(/<(?:img|source)\b[^>]*>/gi)) {
+        const src = /\ssrc\s*=\s*(?:"([^"]*)"|'([^']*)')/i.exec(tag);
+        if (src)
+            add(src[1] ?? src[2] ?? "");
+        const srcset = /\ssrcset\s*=\s*(?:"([^"]*)"|'([^']*)')/i.exec(tag);
+        if (srcset)
+            for (const entry of (srcset[1] ?? srcset[2] ?? "").split(/,\s+/))
+                add(entry.trim().split(/\s+/)[0] ?? "");
+    }
+    return [...found];
+}
+/**
+ * A feed often offers a thumbnail (`…__wide__148`) of the picture that its article shows in full size. When the
+ * article carries the same picture in a clearly larger size, that size is the better hero; otherwise there is none.
+ */
+export function largerHeroImage(imageUrl, html, baseUrl) {
+    if (!imageUrl || !html || !html.includes("<"))
+        return null;
+    const current = imageVariant(imageUrl);
+    if (!current)
+        return null;
+    let best = null;
+    for (const address of imageAddresses(html, baseUrl)) {
+        const variant = imageVariant(address);
+        if (!variant || variant.key !== current.key)
+            continue;
+        if (variant.width < current.width * MIN_GAIN || variant.width > MAX_HERO_WIDTH)
+            continue;
+        if (!best || variant.width > best.width)
+            best = { url: address, width: variant.width };
+    }
+    return best?.url ?? null;
+}
+//# sourceMappingURL=heroImage.js.map

package/server/dist/feeds/repository.js

@@ -1,6 +1,7 @@
 import { db } from "../db/index.js";
 import { articleText } from "./articleContent.js";
 import { LEAD_VERSION } from "./contentNoise.js";
+import { largerHeroImage } from "./heroImage.js";
 import { refreshLead } from "./lead.js";
 import { decodeEntities } from "./text.js";
 export function findFeedByUrl(url) {
@@ -157,7 +158,7 @@ export function upsertItems(feedId, items) {
                 row.contentSnippet ?? null,
                 row.contentHtml ?? null,
                 row.publishedAt ?? null,
-                row.imageUrl ?? null,
+                largerHeroImage(row.imageUrl, row.contentHtml, row.link ?? null) ?? row.imageUrl ?? null,
             ];
             const result = insert.run(...values);
             if (result.changes > 0) {
@@ -425,6 +426,14 @@ export function findItemById(id) {
 export function updateItemFullContent(id, fullContentHtml) {
     db.prepare("UPDATE items SET full_content_html = ? WHERE id = ?").run(fullContentHtml, id);
     refreshLead(id);
+    upgradeHeroImage(id, fullContentHtml);
+}
+/** The feed's picture is often a thumbnail of the one the article shows: use the larger size once the text is known. */
+function upgradeHeroImage(id, html) {
+    const item = db.prepare("SELECT image_url, link FROM items WHERE id = ?").get(id);
+    const larger = item ? largerHeroImage(item.image_url, html, item.link) : null;
+    if (larger)
+        db.prepare("UPDATE items SET image_url = ? WHERE id = ?").run(larger, id);
 }
 /** Apply a per-item change to many items atomically; returns how many rows changed. */
 export function applyToItems(itemIds, change) {

package/server/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@feedkeeper/server",
-  "version": "0.26.5",
+  "version": "0.26.6",
   "private": true,
   "type": "module"
 }