Closed josePhoenix closed 8 years ago
I don't understand - what requirements are forcing this downgrade?
I think it's not anything in webbpsf
or poppy
themselves, but has to be something in the astroconda-contrib
specifications. The requirements in poppy are just 'numpy>=1.8.0' so any version above that should be OK. Right? But conversely I don't see what's wrong with the meta.yaml
files for either in astroconda.
I think it's even more ridiculous. I think it might be that Continuum Analytics has not built a notebook
package with the appropriate magic bits that say it works with NumPy 1.11. @jhunkeler and I talked it over and iirc that's what we came up with.
Conversely, I suppose, we have not built a webbpsf/webbpsf-data/poppy package set for the new version that says it works with 1.10. That should be within our capabilities to fix.
But that's what confuses me. I don't see anything in https://github.com/astroconda/astroconda-contrib/blob/master/poppy/meta.yaml which indicates that it's not compatible with numpy 1.10. Same for https://github.com/astroconda/astroconda-contrib/blob/master/webbpsf/meta.yaml
The blame is split between Continuum and myself. I stopped building against numpy 1.10 in favor of 1.11 to cut down on the overall build-time. It is Continuum's fault for simply not keeping their packages up to date with the latest numpy.
Why did I go from 1.10 to 1.11 so suddenly? At the time, Continuum compiled Astropy 1.2.1 against numpy 1.11, but not 1.10 and some of our software already relied on features in Astropy 1.2.1.
Tough situation -- poor execution, I'll admit.
In any case, I posted a message on the Anaconda mailing list when this happened. It took a few days (a few days too long), but they ended up recompiling Astropy 1.2.1 against numpy >=1.8. We should probably just tell them about notebook
... there's high probability they have no idea it's even missing 1.11 support.
Scratch everything I just said... pywcs
was deprecated in stsci-hst
1.0.3 release, and thus was not compiled against numpy 1.11.
@josePhoenix @mperrin
Ok, it's definitely Monday.
When I upgraded the base packages to 1.11 the stsci
metapackage received a build
bump instead of a version
bump. So instead of cascading the numpy 1.11 packages it continued to provide numpy 1.10.
I suggest running conda update stsci
then conda install notebook
if you want to hop up to 1.11. Since notebook has no hard dependency on numpy, installing it by itself will not automatically upgrade stsci
. So if you try again, your conda client should go out and check the dependencies and be able to resolve notebook
without downgrading.
Hopefully this change clears up the mess. Let me know what happens on your end.
Thanks @jhunkeler ! Just built a new astroconda env with conda create --name astroconda python=3.5 stsci notebook
and successfully imported WebbPSF v0.5.0.
It appears that AstroConda does not provide a WebbPSF 0.5 package that is compatible with the version of the Jupyter Notebook available from Continuum Analytics. It's not clear whether this should be fixed by our Science Software Branch with a package stating compatibility with NumPy 1.10 (cc @jhunkeler) or by Continuum Analytics rebuilding the
notebook
package with the latest NumPy.¯¯_(ツ)_/¯¯
To reproduce:
conda create -n somename stsci
- install main AstroConda metapackageconda install notebook
- install the Jupyter Notebook (packaged by Continuum)Full log of replication below: