Having worked at Google for so long, I'm comfortable with having every dependency at the latest version. Internally at Google just about everything is built at HEAD.
That said, for libraries / core infrastructure having generous version ranges in setup.py is more typical. It makes me a little uncomfortable, but a lot of packages in the pandas ecosystem don't specify any max version in their setup.py files. To do either of these (always be at latest version or generous version ranges) safely, very thorough testing is needed. * • **Tim Swast* * • *Software Friendliness Engineer * • *Google Cloud Developer Relations * • *Seattle, WA, USA On Sat, Jan 12, 2019 at 8:21 AM Felix Uellendall <[email protected]> wrote: > Hey all, > > I have just created AIP-9 and would love to hear your feedback on this. > > It's about Dependency Management in Airflow and how we could improve it. > > > https://cwiki.apache.org/confluence/display/AIRFLOW/AIP-9+Automated+Dependency+Management > > Best regards, > Felix Uellendall > > >
