Since the test tolerances were changed, the expected values were updated. From a first look, it's not obvious that the current results, either in Python 2 or Python 3, are within the old tolerances of the currently expected values. I'll have to do some more digging, but it looks like this isn't just a simple case of restoring the old places values.

I was wrong about the above: just resetting places seems to do the trick.

PR: https://github.com/lsst/pipe_tasks/pull/96

Waiting for DM-8030 to merge before I send this for review, and also planning to check this works on a Mac (which means rebuilding the stack locally).

Jenkins:

• lsst_py3: https://ci.lsst.codes/job/lsst_py3/420/
• lsst_distrib: https://ci.lsst.codes/job/lsst_distrib/625/

Building on a Mac to test there ran into DM-9316, hence the delay.

Finally managed to demonstrate that these numbers are correct on a Mac, too.

Fred Moolekamp, I think you made the original changes here during the Python 3 port. Are you willing to review this work? Should be quick. PR here.

Sure, I should be able to have a look at it tonight or tomorrow morning.

This looks good, and I see that you reverted to the previous tolerance of 1e-7. That's fine, although at some point it would be nice to have justifications in our tests as to why particular tolerances were chosen. But of course that would be outside the scope of this ticket.