Do not parse URL CGI part recursively.

doc/changelog.txt

@@ -3,6 +3,9 @@
 Fixes:
 - checking: Do not check content of already cached URLs.
   Closes: SF bug #1720083
+- checking: Do not parse URL CGI part recursively, avoiding maximum
+  recursion limit errors.
+  Closes: SF bug #3096115
 
 Changes:
 - cmdline: Don't log a warning if URL has been redirected.

linkcheck/url.py

@@ -251,11 +251,10 @@ def url_parse_query (query, encoding=None):
             encoding = url_encoding
         query = query.encode(encoding, 'ignore')
     # if ? is in the query, split it off, seen at msdn.microsoft.com
-    if '?' in query:
-        query, append = query.split('?', 1)
-        append = '?'+url_parse_query(append)
-    else:
-        append = ""
+    append = ""
+    while '?' in query:
+        query, rest = query.rsplit('?', 1)
+        append = '?'+url_parse_query(rest)+append
     l = []
     for k, v, sep in parse_qsl(query, True):
         k = url_quote_part(k, '/-:,;')

tests/test_url.py

@@ -526,6 +526,10 @@ class TestUrl (unittest.TestCase):
         u = "scid=kb;en-us;Q248840&b=c;hulla=bulla"
         self.assertEqual(linkcheck.url.url_parse_query(u), u)
 
+    def test_long_cgi (self):
+        u = "/test%s;" % ("?a="*1000)
+        self.assertEqual(linkcheck.url.url_parse_query(u), u)
+
     def test_port (self):
         is_numeric_port = linkcheck.url.is_numeric_port
         self.assertTrue(is_numeric_port("80"))