Different studies have different biases.
E.g.
http://blog.frite-camembert.net/python-survey-2014.html says that
there are actually more 3.2 coders than 2.5 ones.
(Not that I care much about 2.5. Just about biases.)
BTW original data is on
https://docs.google.com/forms/d/1DqxkNi4GvyTCu54usSdE1DjW29zw1tc52iMeH3z4heg/viewanalytics
, here are the numbers for "which Python versions do you regularly use":
2.5 89 1.3%
2.6 609 9.0%
2.7 5501 81.4%
3.2 191 2.8%
3.3 797 11.8%
3.4 2916 43.1%
3.5 238 2.0%
Seems like 3.2 is indeed not much in use anymore.
Still, it's really easy to cut out too much. If you drop support for
anything that's less than 5%, you already lose 6.1% of the user base, or
one out of 16 potential users. If we go just by the numbers, we should
probably rather drop 3.5...
> I believe tests should not run on Python 3.2, it's just a waste of time for
> a Python version that hardly anyone is using.
I'm not feeling very strongly either way, actually.
It might be a good idea to do the tests in order of descending relevance.
I.e. first 2.7 and 3.4 (stable versions), 3.5 (upcoming version), then
3.3, 2.6, and 3.2 (deprecated but still in use, in descending order of
usage).