This is related to https://bugs.python.org/issue27321 but a different exception is thrown for a different reason. This is caused by a defective spam message. I don't actually have the offending message from the wild, but the attached bad_email_2.eml illustrates the problem.

The defect is the message declares the content charset as us-ascii, but the body contains non-ascii. When the message is parsed into an email.message.Message object and the objects as_string() method is called, UnicodeEncodeError is thrown as follows:

>>> import email
>>> with open('bad_email_2.eml', 'rb') as fp:
...     msg = email.message_from_binary_file(fp)
... 
>>> msg.as_string()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3.5/email/message.py", line 159, in as_string
    g.flatten(self, unixfrom=unixfrom)
  File "/usr/lib/python3.5/email/generator.py", line 115, in flatten
    self._write(msg)
  File "/usr/lib/python3.5/email/generator.py", line 181, in _write
    self._dispatch(msg)
  File "/usr/lib/python3.5/email/generator.py", line 214, in _dispatch
    meth(msg)
  File "/usr/lib/python3.5/email/generator.py", line 243, in _handle_text
    msg.set_payload(payload, charset)
  File "/usr/lib/python3.5/email/message.py", line 316, in set_payload
    payload = payload.encode(charset.output_charset)
UnicodeEncodeError: 'ascii' codec can't encode characters in position 31-33: ordinal not in range(128)

What would you like to see happen in that situation?  Should we use errors=replace like we do for headers?  (That seems reasonable to me.)

Note that it can be re-serialized as binary.

Yes. I think errors=replace is a good solution. In Mailman, we have our own mailman.email.message.Message class which is a subclass of email.message.Message and what we do to work around this and issue27321 is override as_string() with:

    def as_string(self):
        # Work around for https://bugs.python.org/issue27321 and
        # https://bugs.python.org/issue32330.
        try:
            value = email.message.Message.as_string(self)
        except (KeyError, UnicodeEncodeError):
            value = email.message.Message.as_bytes(self).decode(
                'ascii', 'replace')
        return value

I do wonder where you are using the string version of messages :)

I actually thought I'd already done this (errors=replace), but obviously not.  I don't have time now to work on a patch for this, and the patch in the other issue hasn't be updated to reflect the review I did :(

> I do wonder where you are using the string version of messages :)

Probably some places where we could use bytes, but one of the problem areas is where we save the content of a message held for moderation.