feat: offer a non-Mews web page as a simplified view or in the browser

app/lib/mews/parser.dart

@@ -26,7 +26,16 @@ MewsPage parseMewsPage({required String source, required String url}) {
   } on XmlException catch (e) {
     throw MewsParseException(e.toString());
   }
+  return parseMewsDocument(doc, url: url);
+}
 
+/// Builds the page from an already-parsed document. A tolerant HTML parse
+/// arrives here too (see html.dart), so the mapping below stays the one place
+/// that decides what the Mews subset shows, whatever parsed the markup.
+/// [specVersionOverride] stands in for the conformance marker a converted
+/// document carries none of.
+MewsPage parseMewsDocument(XmlDocument doc,
+    {required String url, String? specVersionOverride}) {
   final html = doc.rootElement;
   if (html.name.qualified != 'html') {
     throw const MewsParseException('The root element is not html.');
@@ -47,7 +56,7 @@ MewsPage parseMewsPage({required String source, required String url}) {
   final dir = _dirOf(html) ?? _dirOf(bodyElement);
   return MewsPage(
     url: url,
-    specVersion: headData.specVersion,
+    specVersion: specVersionOverride ?? headData.specVersion,
     title: headData.title,
     description: headData.description,
     author: headData.author,
@@ -139,7 +148,12 @@ List<MewsEntry> _entriesOf(XmlElement body, String pageUrl) {
 // Blocks --------------------------------------------------------------------
 
 /// Elements whose whole subtree is not content and must never reach the model.
-const _droppedElements = {'script', 'style', 'object', 'meta', 'link', 'head'};
+const _droppedElements = {
+  'script', 'style', 'object', 'meta', 'link', 'head',
+  // Not content on the web either: markup the renderer has nothing to draw
+  // for, and controls that are interface rather than text.
+  'svg', 'iframe', 'noscript', 'template', 'canvas', 'audio', 'video', 'button',
+};
 
 void _appendBlocks(XmlElement element, String pageUrl, List<MewsBlock> out) {
   switch (element.name.qualified) {
@@ -170,7 +184,9 @@ void _appendBlocks(XmlElement element, String pageUrl, List<MewsBlock> out) {
     case 'dl':
       out.add(_definitionBlock(element, pageUrl));
     case 'table':
-      out.add(_tableBlock(element, pageUrl));
+      final nested = <MewsBlock>[];
+      out.add(_tableBlock(element, pageUrl, nested));
+      out.addAll(nested);
     case 'img':
       out.add(_imageBlock(element, pageUrl));
     case 'form':
@@ -180,11 +196,11 @@ void _appendBlocks(XmlElement element, String pageUrl, List<MewsBlock> out) {
         _appendBlocks(child, pageUrl, out);
       }
     default:
-      // Unknown or dropped wrappers: keep the readable content, drop the rest.
+      // Unknown wrappers: keep the readable content, drop the rest. Mixed
+      // content has to be handled here, not just element children, because
+      // text sits directly inside a div on most of the web.
       if (!_droppedElements.contains(element.name.qualified)) {
-        for (final child in element.children.whereType<XmlElement>()) {
-          _appendBlocks(child, pageUrl, out);
-        }
+        out.addAll(_mixedBlocks(element, pageUrl));
       }
   }
 }
@@ -255,18 +271,31 @@ DefinitionListBlock _definitionBlock(XmlElement element, String pageUrl) {
   return DefinitionListBlock(items);
 }
 
-TableBlock _tableBlock(XmlElement element, String pageUrl) {
+/// The nearest table a node sits in, so a row can say which table owns it.
+XmlElement? _enclosingTable(XmlNode node) => node.ancestors
+    .whereType<XmlElement>()
+    .where((e) => e.name.qualified == 'table')
+    .firstOrNull;
+
+/// Builds the table, adding to [embedded] any table nested in one of its
+/// cells. Tables cannot nest in the model (spec 4.2), so a nested one — old
+/// web layout markup — becomes a table of its own rather than being flattened
+/// into the cell that held it.
+TableBlock _tableBlock(XmlElement element, String pageUrl,
+    [List<MewsBlock>? embedded]) {
   final rows = <MewsTableRow>[];
-  for (final row in element.descendants
-      .whereType<XmlElement>()
-      .where((e) => e.name.qualified == 'tr')) {
+  // Rows are found among the descendants because thead, tbody and tfoot sit
+  // between, and an HTML parse inserts a tbody even where the markup has
+  // none. A row owned by a nested table belongs to that table, not this one.
+  for (final row in element.descendants.whereType<XmlElement>().where(
+      (e) => e.name.qualified == 'tr' && _enclosingTable(e) == element)) {
     final cells = <MewsTableCell>[];
     for (final cell in row.children.whereType<XmlElement>()) {
       final isHeader = cell.name.qualified == 'th';
       if (!isHeader && cell.name.qualified != 'td') continue;
       cells.add(MewsTableCell(
         header: isHeader,
-        inline: _inline(cell, pageUrl),
+        inline: _inline(cell, pageUrl, embedded),
         colspan: int.tryParse(cell.getAttribute('colspan') ?? '') ?? 1,
         rowspan: int.tryParse(cell.getAttribute('rowspan') ?? '') ?? 1,
       ));
@@ -489,6 +518,11 @@ void _appendInline(XmlNode node, String pageUrl, List<MewsInline> out,
       // Every image is shown in its own block (spec 4.2), so an inline image
       // is lifted out of the paragraph that contains it.
       embedded?.add(_imageBlock(node, pageUrl));
+    case 'table':
+      // Likewise a table, which cannot sit inside another (spec 4.2). Without
+      // this it would fall to the default below and its cells would run
+      // together as text.
+      embedded?.add(_tableBlock(node, pageUrl, embedded));
     default:
       if (!_droppedElements.contains(node.name.qualified)) {
         for (final child in node.children) {
@@ -497,17 +531,3 @@ void _appendInline(XmlNode node, String pageUrl, List<MewsInline> out,
       }
   }
 }
-
-/// Recovers readable text from a page that failed XML parsing, for the "show
-/// its text" escape hatch (spec 8.1). Best effort by design.
-String mewsPlainText(String source) {
-  final withoutBlocks = source.replaceAll(RegExp(r'<(script|style)[\s\S]*?</\1\s*>', caseSensitive: false), '');
-  final stripped = withoutBlocks.replaceAll(RegExp(r'<[^>]+>'), ' ');
-  return stripped
-      .replaceAll('&amp;', '&')
-      .replaceAll('&lt;', '<')
-      .replaceAll('&gt;', '>')
-      .replaceAll('&quot;', '"')
-      .replaceAll(RegExp(r'\s+'), ' ')
-      .trim();
-}