I love properties and think they should be everywhere. But consistency is more important, so I suspect that EmailMessage.is_attachment should be demoted to a normal method. Why? Because if it remains a property then I am likely to first write:

    if msg.is_attachment:
        ...

and then later, when doing another bit of email logic, write:

    if msg.is_multipart:
        ...

Unfortunately this second piece of code will give me no error and will appear to run just fine, because bool(a_method) always returns True without a problem or warning or error. But the result will not be what I expect: the if statement's true block will always run, regardless of whether the message is multipart.

Since EmailMessage is still provisional, and since no one can use is_attachment yet anyway because it is broken for nearly all attachments, mightn't we make these two features consistent before calling it official?

Patch to change message.is_attachment from a property to a normal method. I've updated the doc and all calls to is_attachment.

This is your call David - I agree consistency is highly desirable, and having a chance to find and fix this kind of discrepancy is a large part of why we introduced provisional APIs.

Unfortunately this will silently break existing code because msg.is_attachment will be always true. But it is possible to make EmailMessage.is_attachment a property which returns special callable with the __bool__() method which will emit deprecation warning and call the __call__() method. After some intermediate period (one or two releases) it can be replaced by regular method.

The alternative is to make EmailMessage.is_multipart a property.

Very good point, Serhiy.

Here is an alternative patch, which instead changes Message.is_multipart from a method to a property as per your suggestion. This way incorrect usage should fail noisily.

This patch is against the relevant docs, tests, is_multipart and calls to is_multipart only.

Here is a patch with more soft transition. Message.is_multipart() still works but emits deprecation warning.

Based on the provisional API status, a faster deprecation plan could be to
do Serhiy's patch in a 3.4 maintenance release and the hard break in 3.5

is_multipart is *not* part of the provisional API, though; only is_attachment is.

So per my understanding of the provisional rules, we should either make is_attachment a method in both 3.4 maint and 3.5, or make is_multipart emit a deprecation warning in 3.5.  

I lean toward the former for backward compatibility reasons, with Serhiy's addition of making it emit a warning if not called in 3.4.

Here is a patch.  Sorry for leaving it until the last minute...maybe someone can review it, but it is simple enough I'll commit it soon regardless.

New changeset a3df1c24d586 by R David Murray in branch '3.4':
#21091: make is_attachment a method.
https://hg.python.org/cpython/rev/a3df1c24d586

New changeset f7aff40609e7 by R David Murray in branch 'default':
Merge: #21091: make is_attachment a method.
https://hg.python.org/cpython/rev/f7aff40609e7