#20531: Revert e20f98a8ed71, the 3.4 version of the #19063 fix.

Doc/library/email.message.rst

@@ -196,13 +196,7 @@ Here are the methods of the :class:`Message` class:
 
       Set the entire message object's payload to *payload*.  It is the client's
       responsibility to ensure the payload invariants.  Optional *charset* sets
-      the message's character set; see :meth:`set_charset` for details.  If
+      the message's default character set; see :meth:`set_charset` for details.
-      *payload* is a string containing non-ASCII characters, *charset* is
-      required.
-
-      .. versionchanged:: 3.4
-         Previous to 3.4 *charset* was not required when *payload* was a
-         non-ASCII string, but omitting it produced nonsense results.
 
    .. method:: set_charset(charset)
 

Lib/email/charset.py

@@ -378,19 +378,18 @@ class Charset:
             return None
 
     def body_encode(self, string):
-        """Body-encode a string, converting it first to bytes if needed.
+        """Body-encode a string by converting it first to bytes.
 
         The type of encoding (base64 or quoted-printable) will be based on
-        self.body_encoding.  If body_encoding is None, we perform no CTE
+        self.body_encoding.  If body_encoding is None, we assume the
-        encoding (the CTE will be either 7bit or 8bit), we just encode the
+        output charset is a 7bit encoding, so re-encoding the decoded
-        binary representation to ascii using the surrogateescape error handler,
+        string using the ascii codec produces the correct string version
-        which will enable the Generators to produce the correct output.
+        of the content.
         """
-        if not string:
+        # 7bit/8bit encodings return the string unchanged (module conversions)
-            return string
-        if isinstance(string, str):
-            string = string.encode(self.output_charset)
         if self.body_encoding is BASE64:
+            if isinstance(string, str):
+                string = string.encode(self.output_charset)
             return email.base64mime.body_encode(string)
         elif self.body_encoding is QP:
             # quopromime.body_encode takes a string, but operates on it as if
@@ -399,7 +398,15 @@ class Charset:
             # character set, then, we must turn it into pseudo bytes via the
             # latin1 charset, which will encode any byte as a single code point
             # between 0 and 255, which is what body_encode is expecting.
-            string = string.decode('latin1')
+            #
+            # Note that this clause doesn't handle the case of a _payload that
+            # is already bytes.  It never did, and the semantics of _payload
+            # being bytes has never been nailed down, so fixing that is a
+            # longer term TODO.
+            if isinstance(string, str):
+                string = string.encode(self.output_charset).decode('latin1')
             return email.quoprimime.body_encode(string)
         else:
-            return string.decode('ascii', 'surrogateescape')
+            if isinstance(string, str):
+                string = string.encode(self.output_charset).decode('ascii')
+            return string

Lib/email/message.py

@@ -301,23 +301,9 @@ class Message:
         Optional charset sets the message's default character set.  See
         set_charset() for details.
         """
-        if hasattr(payload, 'encode'):
+        if isinstance(payload, bytes):
-            if charset is None:
+            payload = payload.decode('ascii', 'surrogateescape')
-                try:
+        self._payload = payload
-                    payload.encode('ascii', 'surrogateescape')
-                except UnicodeError:
-                    raise TypeError("charset argument must be specified"
-                        " when non-ASCII characters are used in the"
-                        " payload") from None
-                self._payload = payload
-                return
-            if not isinstance(charset, Charset):
-                charset = Charset(charset)
-            payload = payload.encode(charset.output_charset)
-        if hasattr(payload, 'decode'):
-            self._payload = payload.decode('ascii', 'surrogateescape')
-        else:
-            self._payload = payload
         if charset is not None:
             self.set_charset(charset)
 
@@ -356,7 +342,7 @@ class Message:
             try:
                 cte(self)
             except TypeError:
-                self._payload = charset.body_encode(self.get_payload(decode=True))
+                self._payload = charset.body_encode(self._payload)
                 self.add_header('Content-Transfer-Encoding', cte)
 
     def get_charset(self):

Lib/test/test_email/test_contentmanager.py

@@ -208,11 +208,12 @@ class TestRawDataManager(TestEmailBase):
                          "Basìc tëxt.\n")
 
     def test_get_text_plain_utf8_base64_recoverable_bad_CTE_data(self):
-        m = self._bytes_msg(textwrap.dedent("""\
+        m = self._str_msg(textwrap.dedent("""\
             Content-Type: text/plain; charset="utf8"
             Content-Transfer-Encoding: base64
 
-            QmFzw6xjIHTDq3h0Lgo""").encode('ascii') + b'\xFF=\n')
+            QmFzw6xjIHTDq3h0Lgo\xFF=
+            """))
         self.assertEqual(raw_data_manager.get_content(m, errors='ignore'),
                          "Basìc tëxt.\n")
 

Lib/test/test_email/test_email.py

@@ -92,44 +92,6 @@ class TestMessageAPI(TestEmailBase):
         msg.set_payload('This is a string payload', charset)
         self.assertEqual(msg.get_charset().input_charset, 'iso-8859-1')
 
-    def test_set_payload_with_8bit_data_and_charset(self):
-        data = b'\xd0\x90\xd0\x91\xd0\x92'
-        charset = Charset('utf-8')
-        msg = Message()
-        msg.set_payload(data, charset)
-        self.assertEqual(msg['content-transfer-encoding'], 'base64')
-        self.assertEqual(msg.get_payload(decode=True), data)
-        self.assertEqual(msg.get_payload(), '0JDQkdCS\n')
-
-    def test_set_payload_with_non_ascii_and_charset_body_encoding_none(self):
-        data = b'\xd0\x90\xd0\x91\xd0\x92'
-        charset = Charset('utf-8')
-        charset.body_encoding = None # Disable base64 encoding
-        msg = Message()
-        msg.set_payload(data.decode('utf-8'), charset)
-        self.assertEqual(msg['content-transfer-encoding'], '8bit')
-        self.assertEqual(msg.get_payload(decode=True), data)
-
-    def test_set_payload_with_8bit_data_and_charset_body_encoding_none(self):
-        data = b'\xd0\x90\xd0\x91\xd0\x92'
-        charset = Charset('utf-8')
-        charset.body_encoding = None # Disable base64 encoding
-        msg = Message()
-        msg.set_payload(data, charset)
-        self.assertEqual(msg['content-transfer-encoding'], '8bit')
-        self.assertEqual(msg.get_payload(decode=True), data)
-
-    def test_set_payload_to_list(self):
-        msg = Message()
-        msg.set_payload([])
-        self.assertEqual(msg.get_payload(), [])
-
-    def test_set_payload_with_non_ascii_and_no_charset_raises(self):
-        data = b'\xd0\x90\xd0\x91\xd0\x92'.decode('utf-8')
-        msg = Message()
-        with self.assertRaises(TypeError):
-            msg.set_payload(data)
-
     def test_get_charsets(self):
         eq = self.assertEqual
 
@@ -596,10 +558,20 @@ class TestMessageAPI(TestEmailBase):
         self.assertIsInstance(msg.defects[0],
                               errors.InvalidBase64CharactersDefect)
 
+    def test_broken_unicode_payload(self):
+        # This test improves coverage but is not a compliance test.
+        # The behavior in this situation is currently undefined by the API.
+        x = 'this is a br\xf6ken thing to do'
+        msg = Message()
+        msg['content-type'] = 'text/plain'
+        msg['content-transfer-encoding'] = '8bit'
+        msg.set_payload(x)
+        self.assertEqual(msg.get_payload(decode=True),
+                         bytes(x, 'raw-unicode-escape'))
+
     def test_questionable_bytes_payload(self):
         # This test improves coverage but is not a compliance test,
-        # since it involves poking inside the black box in a way
+        # since it involves poking inside the black box.
-        # that actually breaks the model invariants.
         x = 'this is a quéstionable thing to do'.encode('utf-8')
         msg = Message()
         msg['content-type'] = 'text/plain; charset="utf-8"'