search

probcomp / bdbcontrib

BayesDB contributions, including plotting, helper methods, and examples
http://probcomp.csail.mit.edu/bayesdb
Apache License 2.0
9 stars 6 forks source link

Avoid gratuitous incompatibilities on Ubuntu 14.04. #134

Closed riastradh-probcomp closed 8 years ago

riastradh-probcomp commented 8 years ago

This makes ./check.sh pass with the only additional package required from pypi being seaborn -- numpy, scipy, pandas, matplotlib, &c. from Ubuntu 14.04 all work.

This does not remove any functionality, nor do I expect it to create incompatibilities in newer versions.

gregory-marton commented 8 years ago

Supporting Ubuntu 14.04 without installing additional packages is an anti-priority. Rather than optimizing for distribution, we are now trying to optimize for making our internal developers' lives easier. If it makes their lives easier to use markdown2 or pandas 0.17, then we should do that. To the extent that pandas 0.17 fixed bugs in earlier versions of pandas, it is preferred. I don't wish to waste time reading about whether that's the case, and your time would be better spent elsewhere also.

riastradh-probcomp commented 8 years ago

I spent much of last week trying to help Leo and other users actually just use the software. I don't mean teaching BQL or the bayeslite API or anything, but simply running any of it at all in the face of gratuitous incompatibilities, which entailed hours and hours of flailing around trying to find exactly the right versions of everything to make anything work and order in which to install things, and then hours of working around the botched local upgrade to 2.7.11 on our machines.

If enabling people to use the software is an 'anti-priority', and if enabling me to support to them is an 'anti-priority', then I haven't the foggiest idea what our priorities are.

Right now I can't with a straight face claim this software works on any platform. If we were doing research on OS package management and repository curation, that would be another matter -- but we're not.