Repository navigation
Conversation
jedi-base-env pulled in 12 python packages, so every environment built on it inherited python whether or not it needed it -- including the model interface environments jedi-mpas-env and jedi-fv3-env. Move 11 of them into a new jedi-python-env bundle, following the existing neptune-env / neptune-python-env split. py-pybind11 stays in jedi-base-env because core JEDI packages link its headers. jedi-mpas-env and jedi-fv3-env are left with no python dependency other than py-pybind11. The other eight jedi-base-env dependents (ai-env, ewok-env, gmao-swell-env, jedi-geos-env, jedi-neptune-env, jedi-ufs-env, jedi-um-env, soca-env) gain an explicit jedi-python-env dependency so their content is unchanged. jedi-python-env is added to the unified-dev and skylab-dev templates.
b833ed6 to
5db8091
Compare
| # (shared with py-pyogrio) so only py-versioneer itself is duplicated. | ||
| - ^py-versioneer@0.28 | ||
| prefer: | ||
| - '~performance' |
There was a problem hiding this comment.
This got added in the last change and ended up setting py-versioneer@0.28 as preferred, I had to move it down to make sure py-versioneer@0.28 stayed in the require: block.
|
@eap Be wary that now that you did this, I might ask/annoy/bother you for a similar (But that's 2.3+ timeframe) |
So far my process has been.
I'm probably sneaking |
climbfuji
left a comment
There was a problem hiding this comment.
This is fine with me. I may take jedi-python-env out of the jedi-neptune-env, because most if not all packages are already in neptune-python-env. Or I'll simplify it in different ways. But that's a later change after the 2.2.0 release.
This change moves scientific python dependencies out of the primary c++/fortran build path for some JEDI components.
Motivation: AI work often is picky about python requirements and can clash with our primary environment. Many jedi components can build with a minimal python environment.
Fixes: https://github.com/JCSDA-internal/jedi-bundle/issues/146
Fixes: #1888