The HTTP status code constants in Python 3.5 http.client are not compatible with the constants in Python 3.4, since the str() behaviour is different. This breaks code: 

srittau@moby:~$ python3.5
Python 3.5.1+ (default, Jan 13 2016, 15:09:18) 
[GCC 5.3.1 20160101] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import http.client
>>> str(http.client.OK)
'HTTPStatus.OK'

vs:

rittau@moby:~$ python3.4
Python 3.4.4 (default, Jan  5 2016, 15:35:18) 
[GCC 5.3.1 20160101] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import http.client
>>> str(http.client.OK)
'200'

These changes were made in issue21793.

Yes, this a side effect of using more human-friendly enums. This is not a problem if string representation is used in formatting human-readable messages, but if you need numerical representation, you could use str(int(code)), '%d' % code or '{:d}'.format(code).

It is no doubt that is easy to work around. Once I found the problem it took about five minutes to fix it and roll a new release. And of course for Python 3.5+ code it is better to use the enum http.HTTPStatus directly (I actually like that enum a lot). But this breaks existing code that should not break. It might be too late to change now, though.

As a side note, I would also have preferred str(HTTPStatus.OK) to return "OK" or "200" - the latter would mirror default format behaviour - instead of "HTTPStatus.OK", but this may be too late as well.

This affected the HTTP server log messages, discussed from <https://bugs.python.org/issue21793#msg235547> onwards. The first three patches proposed changed the HTTPStatus.__str__() implementation, which would have avoided the problem in general. But Demian’s final patch made a special case of checking for HTTPStatus objects in the logging code instead.

IMO it may have been better to change HTTPStatus.__str__() to use int.__str__() at the time.

Not using Enum's __str__ was discussed (I think during the initial Enum threads when PEP435 was being debated) and IIRC Guido was strongly against it as it took away half the purpose of using an Enum.

I take the liberty of closing this "wont fix". Changing the behaviour would most likely do more harm than good. If one of the maintainers disagrees, please reopen. :)

Thank you Sebastian.