The conda package versions should be updated as part of general house keeping at the start of each dev/release cycle.
add scikit-learn to conda env
lsstsw & newinstall.sh should use named conda envs
lsstsw/bin/deploy can not update/switch conda package sets
Prepare and execute 15.0 release
Seemingly Large demo change with bleeding edge pipelines build
remove jenkins py2 builds - post v15 release
Upgrade to Anaconda 5.0
ci_hsc due to shebang length
pytest (and scons) are broken on the lsst-dev01 shared stack
jupyterlabdemo build failed on d.2018.07.10
documenteer build failed on d.2018.07.10
Update version checks in EUPS stub packages to match lsstsw minimums
Upgrade to Anaconda 5.0 (Miniconda 4.3.27)
Fix or disable healpy with Anaconda 5.0
Fix lsstsw installer to disable MKL
Fix newinstall installer to disable MKL
Make LSSTSW install work with tcsh
There is no point in updating the conda package set until after py2 support has been dropped.
Completing this ticket is hard blocked on DM-15044 as per discussion on slack, disabling the "demo" is not considered an option at this time.
ktl / tjenness It appears that updating the conda env should be unblocked? Do we need to go through a testing cycle or should be just go for it?
So long as you are not going to update to current versions of packages then the bump from a few weeks back is probably fine. DM-15462 demonstrates that numpy1.15 breaks things (although python 3.7 looks to be fine in general).
OK – I'll proceed with the PR as it currently stands as it's bumping numpy to 1.14.x.
It would be awesome if we had a single Jenkins job that ran once a week and that always built with bleeding edge packages so that we could spot oddities early.
All changes have been merged.