We currently claim to support 2.6 in the setup.py classifiers and our TravisCI config is testing against it (our tox config also includes py26 but I don't think this is used anywhere). The only reason our unit tests pass on 2.6 is because they have very weak coverage.
There are 2 issues with 2.6 that I know of:
It does not include the argparse module that our scripts rely on.
It does not support format un-indexed placeholders {}.
We are already using astropy.utils.compat.argparse to work around the first problem, but adding indices to every {} would be painful (and perhaps make the code less readable).
There are probably other 2.6 incompatibilities that we don't know about since nobody is actually using it.
Should we give up on 2.6 or get more serious about testing with it?
I am now leaning towards abandoning 2.6 since I don't think we have any users and just realized that this would speed up our continuous integration tests by a factor of 2.
We currently claim to support 2.6 in the
setup.py
classifiers and our TravisCI config is testing against it (ourtox
config also includespy26
but I don't think this is used anywhere). The only reason our unit tests pass on 2.6 is because they have very weak coverage.There are 2 issues with 2.6 that I know of:
argparse
module that our scripts rely on.format
un-indexed placeholders{}
.We are already using
astropy.utils.compat.argparse
to work around the first problem, but adding indices to every{}
would be painful (and perhaps make the code less readable).There are probably other 2.6 incompatibilities that we don't know about since nobody is actually using it.
Should we give up on 2.6 or get more serious about testing with it?