Implement RFC-725 and update the base scipipe conda environment to python 3.8.
Upgrade to newer pybind11
Changes to validate_drp to support treecorr > 4
afw fails to build with pybind11 2.5 (and python 3.8)
ingestIndexManager does not set up global multiprocessing value before forking, breaking multiprocessing after python 3.8. on mac
fgcm multiprocessing strategy with shared memory manager doesn't work well with python 3.8
pybind11 errors out at runtime with afw because Background and BackgroundMI use different holders
Upgrade to python 3.8
Update baseline python to version 3.8
astshim incompatible with pybind11 2.5
Update scipipe_conda_env to distribute rubin-env
ctrl_mpexec fails (or may sometimes fail) on mac with python 3.8 in test_mpexec_faillure_failfast
errors when compiling afw with pybind11 for pixel types
Upgrade to boost 1.72 or newer
Invetigate scarlet failures in python 3.8
use multiprocessing context with fork for ctrl_mpexec for python 3.8 support on mac
I think the following restrictions will need to be addressed to go to python 3.8:
• pybind11 < 2.3
• mpich = 3.2.1
• treecorr < 4
I will check on mpich because I think there was a bug but there's been new releases since then. pybind11 already has a ticket, but I don't think treecorr has a ticket right now.
I'm sorry, I had not realized you already had a ticket for 3.8 – you didn't mention it on the RFC. Shall I close this one?
Is the treecorr issue an API change that breaks our usage or a python 3.8 issue?
mpich may be blocked until 3.4 unless this issue is no longer relevant:
(we don't have a unit test for that, it is something we discovered at runtime)
I don't have any data on treecorr, I think it was an API change that broke things. I will test a build in jenkins with treecorr 4 right now
A ticket for the treecorr migration has been created - there is a small API change between 3 and 4 for ggcorrelation, which validate_drp uses.
Did treecorr get fixed?
Has this ticket been made irrelevant by DM-27005 ?
We're not putting this into scipipe_conda_env, but into rubin-env instead. Still marking this done.