wsgi: Work around CPython bug when parsing non-ASCII headers

eventlet/wsgi.py

@@ -674,10 +674,13 @@ def get_environ(self):
 
         ct = self.headers.get('content-type')
         if ct is None:
+            ct_was_none = True
             try:
                 ct = self.headers.type
             except AttributeError:
                 ct = self.headers.get_content_type()
+        else:
+            ct_was_none = False
         env['CONTENT_TYPE'] = ct
 
         length = self.headers.get('content-length')
@@ -694,16 +697,33 @@ def get_environ(self):
         env['REMOTE_PORT'] = str(client_addr[1])
         env['GATEWAY_INTERFACE'] = 'CGI/1.1'
 
-        try:
+        if six.PY2:
             headers = self.headers.headers
-        except AttributeError:
-            headers = self.headers._headers
-        else:
             headers = [h.split(':', 1) for h in headers]
+        else:
+            headers = self.headers._headers
+            payload = self.headers.get_payload()
+            if payload:
+                # There shouldn't be a message associated with the headers;
+                # must've bumped up against https://bugs.python.org/issue37093
+                for line in payload.rstrip('\r\n').split('\n'):
+                    if ':' not in line or line[:1] in ' \t':
+                        # Well, we're no more broken than we were before...
+                        # Should we support line folding? Should we 400 a bad header line?
+                        break
+                    header, value = line.split(':', 1)
+                    headers.append((header, value))
+                    if ct_was_none and header.lower() == 'content-type':
+                        env['CONTENT_TYPE'] = value.strip(' \t\n\r')
+                    elif length is None and header.lower() == 'content-length':
+                        length = env['CONTENT_LENGTH'] = value.strip(' \t\n\r')
 
         env['headers_raw'] = headers_raw = tuple((k, v.strip(' \t\n\r')) for k, v in headers)
         for k, v in headers_raw:
-            k = k.replace('-', '_').upper()
+            if six.PY2:
+                k = k.replace('-', '_').upper()
+            else:
+                k = k.replace('-', '_').encode('latin1').upper().decode('latin1')
             if k in ('CONTENT_TYPE', 'CONTENT_LENGTH'):
                 # These do not get the HTTP_ prefix and were handled above
                 continue

tests/wsgi_test.py

@@ -1534,6 +1534,59 @@ def wsgi_app(environ, start_response):
         assert isinstance(g[1], str), msg
         assert g[1] == '/\xbd\xa5\xe5\xa0\xbd\xe4'
 
+    def test_headers_latin1(self):
+        g = []
+
+        def wsgi_app(environ, start_response):
+            g.append(environ)
+            start_response("200 OK", [])
+            return [environ['wsgi.input'].read()]
+
+        self.site.application = wsgi_app
+        sock = eventlet.connect(self.server_addr)
+        sock.sendall(b'PUT / HTTP/1.1\r\n'
+                     b'Snow-Man: ' + u'\N{SNOWMAN}'.encode('utf8') + b'\r\n'
+                     b'Utf-8-' + u'\U0001F334'.encode('utf-8') + b': palm tree\r\n'
+                     b'Content-Length: 4\r\n'
+                     b'\r\ndata')
+        result = read_http(sock)
+        assert result.status == 'HTTP/1.1 200 OK'
+        assert result.body == b'data'
+
+        assert 'HTTP_SNOW_MAN' in g[0]
+        # WSGI demands native strings, either as bytes or decoded-Latin-1
+        assert isinstance(g[0]['HTTP_SNOW_MAN'], str)
+        assert g[0]['HTTP_SNOW_MAN'] == '\xE2\x98\x83'
+
+        assert 'HTTP_UTF_8_\xF0\x9F\x8C\xB4' in [h for h in g[0] if h.startswith('HTTP_')]
+        assert isinstance(g[0]['HTTP_UTF_8_\xF0\x9F\x8C\xB4'], str)
+        assert g[0]['HTTP_UTF_8_\xF0\x9F\x8C\xB4'] == 'palm tree'
+
+        sock.sendall(b'PUT / HTTP/1.1\r\n'
+                     b'Connection: close\r\n'
+                     b'Snow-Man: ' + u'\N{SNOWMAN}'.encode('utf8') + b'\r\n'
+                     b'Utf-8-' + u'\U0001F334'.encode('utf-8') + b': palm tree\r\n'
+                     b'Content-Type: foo/bar\r\n'
+                     b'Transfer-Encoding: chunked\r\n'
+                     b'\r\n'
+                     b'e\r\n'
+                     b'Hello, world!\n\r\n'
+                     b'0\r\n\r\n')
+        result = read_http(sock)
+        assert result.status == 'HTTP/1.1 200 OK'
+        assert result.body == b'Hello, world!\n'
+
+        assert 'HTTP_SNOW_MAN' in g[1]
+        # WSGI demands native strings, either as bytes or decoded-Latin-1
+        assert isinstance(g[1]['HTTP_SNOW_MAN'], str)
+        assert g[1]['HTTP_SNOW_MAN'] == '\xE2\x98\x83'
+
+        assert 'HTTP_UTF_8_\xF0\x9F\x8C\xB4' in [h for h in g[1] if h.startswith('HTTP_')]
+        assert isinstance(g[1]['HTTP_UTF_8_\xF0\x9F\x8C\xB4'], str)
+        assert g[1]['HTTP_UTF_8_\xF0\x9F\x8C\xB4'] == 'palm tree'
+
+        assert g[1]['CONTENT_TYPE'] == 'foo/bar'
+
     @tests.skip_if_no_ipv6
     def test_ipv6(self):
         try: