Conversion pipeline: Disable HTML 5 parsing if it results in deeply nested trees. Fixes #908818 (RuntimeError: maximum recursion depth exceeded in cmp)

src/calibre/ebooks/mobi/reader.py

@@ -502,7 +502,7 @@ class MobiReader(object):
         self.processed_html = self.processed_html.replace('> <', '>\n<')
         self.processed_html = self.processed_html.replace('<mbp: ', '<mbp:')
         self.processed_html = re.sub(r'<\?xml[^>]*>', '', self.processed_html)
-        self.processed_html = re.sub(r'<(/?)o:p', r'<\1p', self.processed_html)
+        self.processed_html = re.sub(r'<\s*(/?)\s*o:p[^>]*>', r'', self.processed_html)
         # Swap inline and block level elements, and order block level elements according to priority
         # - lxml and beautifulsoup expect/assume a specific order based on xhtml spec
         self.processed_html = re.sub(r'(?i)(?P<styletags>(<(h\d+|i|b|u|em|small|big|strong|tt)>\s*){1,})(?P<para><p[^>]*>)', '\g<para>'+'\g<styletags>', self.processed_html)

src/calibre/ebooks/oeb/parse_utils.py

@@ -70,9 +70,27 @@ def clone_element(elem, nsmap={}, in_context=True):
     nelem.extend(elem)
     return nelem
 
-def html5_parse(data):
+def node_depth(node):
+    ans = 0
+    p = node.getparent()
+    while p is not None:
+        ans += 1
+        p = p.getparent()
+    return ans
+
+def html5_parse(data, max_nesting_depth=500):
     import html5lib
     data = html5lib.parse(data, treebuilder='lxml').getroot()
+
+    # Check that the asinine HTML 5 algorithm did not result in a tree with
+    # insane nesting depths
+    for x in data.iterdescendants():
+        if len(x) == 0:
+            # Leaf node
+            depth = node_depth(x)
+            if depth > max_nesting_depth:
+                raise ValueError('html5lib resulted in a tree with nesting'
+                        ' depth > %d'%max_nesting_depth)
     # Set lang correctly
     xl = data.attrib.pop('xmlU0003Alang', None)
     if xl is not None and 'lang' not in data.attrib: