We may well want to move past v13_0, in particular to a pybind11-based release.

Before this switch, we need to minimise the amount of C++ exposed to python! At present there are many functions that are either not used (especially instantiations), but also things written in C++ that could probably by python with no loss of speed (e.g. addFiberTraceToCcdArray)

I thought we decided to modernize in two steps, using pybind11 as the start of the second? That said I was mostly interested in getting obs_pfs updated, which was trivial and does not depend on swig.

I thought we decided to modernize in two steps...

I think Craig's recollection is correct — we agreed that moving to v13.0 was doable with minimal effort by Craig in the short term, while conversion to pybind11 would take more effort (and, as Robert and I discussed earlier, we'll likely call on Paul & Pim to do the bulk of the work).

I do think we should go to pybind11 as soon as we can reasonably schedule it, but I don't think that should stop Craig moving to 13.0 if he has time now.

One question: is drp_stella's master well enough merged, or are there some larger active branches one I should wait for?

After some discussion we decided to target hscPipe 5.0. 5.0 was branched off of LSST's w.2017.10 and was used for the HSC DR1 reductions. I believe the LSST 13.0 tag was applied to w.2017.6.

hscPipe 5.0 is also before LSST's pybind11 work, so we are deferring that for now. As of this afternoon we intend to track the hscPipe development. The specific tag we are starting with is hscPipe 5.0 beta8

The worst problem for me (and JHU and LAM) is the loss of the conda install.

This turned out to be trivial, only requiring tiny renames for pex_logging -> log and some minor work for daf_butlerUtils -> obs_base.

I think I should wait until any outstanding drp_stella ticket merges are done.

Re conda:

The conda stuff was really bad (a new EUPS_PATH for each upgrade). I believe that binary installs are now working, which should let you install into a pre-existing conda environment. Is that acceptable?

Superseded by PIPE2D-220 (go to LSST master)