> On 7 Oct 2016, at 16:04, Gerrit Holl <g.h...@reading.ac.uk> wrote:
> On 7 October 2016 at 14:58, Lukas Kluft <lukas.kl...@gmail.com> wrote:
>> But I definitely support your point on unit testing.
> And which means that to do it properly, we need not only unit testing,
> but unit testing using (at least) the oldest officially supported
> versions of python, numpy, and other dependencies...
I think we have to stick to a more practical way of handling unit tests without
overengineering things. ;-) We decided for us to try to implement a simple test
for new functions and when a bug is discovered, add an appropriate test to
cover that issue for the future.
If someone is using an unsupported version of a package, even a simple test
should uncover that. So by just running the test suite, hopefully users get a
hint of any incompatibility problems. I don't think it's feasible to try to
find out which exact version for every package we depend on is the oldest we
can support. And always requiring the latest version of all packages would not
be a good experience for typhon users either. It's not practical/possible for
everyone to always keep their python environments updated to the minute.
typhon.mi mailing list