bpo-40409: Updates urlsplit scheme validation logic

* Checks that the first character of a scheme is a letter.
* Uses frozensets for scheme char validation instead of a string as the
  use of a string led to linear scanning when parsing
* Adds a unit test

Assuming this change is appropriate it should be easy to backport to older
versions of Python.

It is worth noting that this *does not* attempt to fix any other potential
parsing issues. While testing I did notice that `urlsplit` returns "paths"
that do not look valid at a glance both before AND after this change
(e.g. `+git+ssh:///+git+ssh://git@github.com/user/project.git"`).

Lib/test/test_urlparse.py

@@ -1027,6 +1027,17 @@ class UrlParseTestCase(unittest.TestCase):
                         with self.assertRaises(ValueError):
                             urllib.parse.urlsplit(url)
 
+    def test_urlsplit_invalid_scheme(self):
+        base_url = "git+ssh://git@github.com/user/project.git"
+        illegal_prefixes = "1234567890+-"
+        illegal_chars = "Å_/"
+
+        self.assertEqual(urllib.parse.urlsplit(base_url).scheme, "git+ssh")
+
+        for prefix in illegal_prefixes + illegal_chars:
+            split_url = urllib.parse.urlsplit(prefix+base_url)
+            self.assertEqual(split_url.scheme, "")
+
 class Utility_Tests(unittest.TestCase):
     """Testcase to test the various utility functions in the urllib."""
     # In Python 2 this test class was in test_urllib.

Lib/urllib/parse.py

@@ -73,10 +73,11 @@ uses_fragment = ['', 'ftp', 'hdl', 'http', 'gopher', 'news',
                  'file', 'prospero']
 
 # Characters valid in scheme names
-scheme_chars = ('abcdefghijklmnopqrstuvwxyz'
-                'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
-                '0123456789'
-                '+-.')
+scheme_alpha_chars = frozenset('abcdefghijklmnopqrstuvwxyz'
+                               'ABCDEFGHIJKLMNOPQRSTUVWXYZ')
+
+scheme_chars = scheme_alpha_chars.union('0123456789'
+                                        '+-.')
 
 # XXX: Consider replacing with functools.lru_cache
 MAX_CACHE_SIZE = 20
@@ -462,8 +463,8 @@ def urlsplit(url, scheme='', allow_fragments=True):
         clear_cache()
     netloc = query = fragment = ''
     i = url.find(':')
-    if i > 0:
-        for c in url[:i]:
+    if i > 0 and url[0] in scheme_alpha_chars:
+        for c in url[1:i]:
             if c not in scheme_chars:
                 break
         else: