[rbarrois]

That's quite interesting.

Now, I'm left to wonder if this could have been used to better handle the Python 2-to-3 transition, e.g '# coding: six.python2' would adjust Python2 code to be valid Python3, or '# coding: six.python3' would adjust Python3 code to run under Python2 - e.g adding/removing the b"..." or u"..." prefixes!

[dgoldstein0]

It could help, but the parts it would help with are the easy parts. The challenges of py2 to py3 were the runtime behavior change: a Unicode and regular string containing ASCII were the same string in py2, as in you could use them as keys of a dict and they'd key the same entry. In py3 a bytes and str of the same ASCII would identify different entries in the same dict.

Some more nasty changes: various built ins like .keys() and .values() return lists in py2 but iterators in py3. Code gets very verbose if you use the six utilities or other workarounds to translate code safely - most times those are called they are used once, but every once in a while they are used twice.

Imo if you have such a tool that can rewrite at import time you should just commit the transformed code, and clean it up incrementally. The hard parts are the behavior changes that can cross long distances like the str v bytes behaving so different than py2 Unicode v str