Merge ticket/3-fun-section into feature/source-kinds (#3)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

lib/services/article_extractor.dart

@@ -49,9 +49,10 @@ class ArticleExtractor {
   /// Hard cap on what we send to the model per article (~10k tokens).
   static const maxChars = 40000;
 
-  /// Loads the article page and returns whichever is longer: the extracted
-  /// page or the full content some feeds ship. Never throws; on failure
-  /// (paywall, bot protection, CORS on web) it falls back to feed content.
+  /// Loads the article page and returns whichever has more text (on a tie,
+  /// more blocks): the extracted page or the full content some feeds ship.
+  /// Never throws; on failure (paywall, bot protection, CORS on web) it falls
+  /// back to feed content.
   ///
   /// With [acceptConsentWalls], a redirect to a consent page the
   /// [ConsentService] knows is accepted once and the article fetched again.
@@ -81,7 +82,12 @@ class ArticleExtractor {
       if (page.wallUrl != null) return ArticleContent(fromFeed.blocks, consentWall: true);
       if (page.status != 200) return fromFeed;
       final fromPage = parse(FeedService.decodeBody(page.bytes), base);
-      return fromPage.plainText.length > fromFeed.plainText.length ? fromPage : fromFeed;
+      final pageText = fromPage.plainText.length, feedText = fromFeed.plainText.length;
+      // Without text either way (a comic), the one with the picture wins.
+      return pageText > feedText ||
+              (pageText == feedText && fromPage.blocks.length > fromFeed.blocks.length)
+          ? fromPage
+          : fromFeed;
     } catch (_) {
       return fromFeed;
     }
@@ -119,7 +125,8 @@ class ArticleExtractor {
   /// Class/id fragments of page furniture that sits inside article bodies.
   static final _junkMarker = RegExp(
     r'comment|related|newsletter|share|social|promo|advert|paywall|subscribe|'
-    r'recommend|author-box|byline|breadcrumb|tags|toolbar|consent|cookie|card|native-ad',
+    r'recommend|author-box|byline|breadcrumb|tags|toolbar|consent|cookie|card|native-ad|'
+    r'license',
     caseSensitive: false,
   );
 
@@ -155,14 +162,15 @@ class ArticleExtractor {
     final blocks = <ReaderBlock>[];
     final seenImages = <String>{};
     var chars = 0;
-    void add(BlockType type, String text) {
-      if (chars > maxChars) return;
+    bool add(BlockType type, String text) {
+      if (chars > maxChars) return false;
       if (type == BlockType.image) {
-        if (!seenImages.add(text)) return;
+        if (!seenImages.add(text)) return false;
       } else {
         chars += text.length;
       }
       blocks.add(ReaderBlock(type, text));
+      return true;
     }
 
     void walk(dom.Element el) {
@@ -190,6 +198,8 @@ class ArticleExtractor {
         case 'blockquote':
           final t = _clean(el.text);
           if (t.isNotEmpty) add(BlockType.quote, t);
+          // Quoted screenshots, as in The Daily WTF's Error'd.
+          el.querySelectorAll('img').forEach((img) => _addImage(img, base, add));
           return;
         case 'pre':
           final t = el.text.trimRight();
@@ -221,7 +231,7 @@ class ArticleExtractor {
   static String _clean(String s) => s.replaceAll(RegExp(r'\s+'), ' ').trim();
 
   static void _addImage(
-      dom.Element img, Uri base, void Function(BlockType, String) add) {
+      dom.Element img, Uri base, bool Function(BlockType, String) add) {
     // Lazy-loading sites keep the real URL in data attributes.
     final raw = img.attributes['data-src'] ??
         img.attributes['data-lazy-src'] ??
@@ -232,19 +242,25 @@ class ArticleExtractor {
     }
     final width = int.tryParse(img.attributes['width'] ?? '');
     if (width != null && width < 120) return; // icons, avatars, tracking pixels
+    if (img.attributes.containsKey('usemap')) return; // clickable navigation
     // Images linking to another page are author avatars or related-story
-    // thumbnails; links to the full-size image file are fine.
+    // thumbnails; links to the full-size image file or within the page are fine.
     for (var p = img.parent; p != null; p = p.parent) {
       if (p.localName == 'a') {
         final href = p.attributes['href'] ?? '';
-        if (!RegExp(r'\.(jpe?g|png|webp|gif|avif)(\?|$)', caseSensitive: false)
-            .hasMatch(href)) {
+        if (!href.startsWith('#') &&
+            !RegExp(r'\.(jpe?g|png|webp|gif|avif)(\?|$)', caseSensitive: false).hasMatch(href)) {
           return;
         }
         break;
       }
     }
     final url = base.resolve(raw).toString();
-    if (url.startsWith('http')) add(BlockType.image, url);
+    if (!url.startsWith('http') || !add(BlockType.image, url)) return;
+    // Comics such as xkcd put a last joke in the title text.
+    final title = _clean(img.attributes['title'] ?? '');
+    if (title.length >= 20 && title != _clean(img.attributes['alt'] ?? '')) {
+      add(BlockType.caption, title);
+    }
   }
 }