You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This follows up on this comment from @skirpichev: #23808 (comment) about f2py -c needing meson and ninja for Python 3.12 (or with --backend=meson on older Python versions).
In numpy it could have an extra f2py section, and that would allow installing the needed dependencies with pip install numpy[f2py]. Of course you have to know that that's possible, and at that point you can equally well do pip install meson ninja , so this isn't all that urgent. An extra complication is that ninja is a system dependency, and using it from PyPI is a bit hacky. That's for example why meson and meson-python do not have a direct dependency on ninja; meson-python first checks if it's installed on the system and dynamically adds it in non-isolated installs if it's missing.
I'd say we don't want to do this for 1.26.0, but we can consider populating the optional dependencies table for 2.0.
The text was updated successfully, but these errors were encountered:
I'm bumping this to the next release cycle, because I don't want to touch optional dependencies right before creating a release branch, and this isn't blocking/important for 2.0
We discussed this a triage meeting and decided to remove it from the milestone, but it would be nice to have this, as well as #26289 to ship tests separately.
This follows up on this comment from @skirpichev: #23808 (comment) about
f2py -c
needingmeson
andninja
for Python 3.12 (or with--backend=meson
on older Python versions).Right now we do not use
optional-dependencies
at all. It is also the future replacement fortest_requirements.txt
& co. Here is an example of how it can be used: https://github.com/scipy/scipy/blob/9f7549abcc61f5dc72ca000cef7bc43066c74527/pyproject.toml#L74-L109In numpy it could have an extra
f2py
section, and that would allow installing the needed dependencies withpip install numpy[f2py]
. Of course you have to know that that's possible, and at that point you can equally well dopip install meson ninja
, so this isn't all that urgent. An extra complication is thatninja
is a system dependency, and using it from PyPI is a bit hacky. That's for example whymeson
andmeson-python
do not have a direct dependency onninja
; meson-python first checks if it's installed on the system and dynamically adds it in non-isolated installs if it's missing.I'd say we don't want to do this for 1.26.0, but we can consider populating the optional dependencies table for 2.0.
The text was updated successfully, but these errors were encountered: