To test up a simple test case for processFile, should I use something from afwdata or obs_test? I’d like to make usre that it actually runs, but I want to keep dependencies minimized.
I’m looking for recommendations and your visions about how these things should look in the future.
[I summarize the HipChat discussion here]
@timj favored afwdata and argued that “an optional afwdata dependency is the lowest dependency solution”
@rowen proposed obs_test but explained that “obs_test brings in the butler machinery, so in that sense it may bring in more than afwdata. If you just need a few images to process and no butler, then Tim may be right. On the other hand, for an A-B test obs_test is nice because you can run it with obs_file and then again with the more normal code to compare results.”
Thanks for the discussion. I think the solution I’ll go with is that processFile will depend on something out of afwdata as an optional dependency for this (thus optional) test. This goes toward the minimal (and even optional) dependency as @timj suggests.
But lsst_ci will run obs_file:processCcd.py and processFile:processFile.py on something out of obs_test to do the A-B test that @rowen recommends.
I just wanted to take this opportunity to mention epic DM-1102, which I see as an effort to restructure and cleanup (or replace) afwdata to make it more compact and more useful - and to point out that I was hoping it might be something the SQuaRE scientist might someday take the lead in designing.
I agree with the goals https://jira.lsstcorp.org/browse/DM-1102. It was with the unresolved nature of this ticket in mind that I asked the question of this present topic.