Closed xylar closed 5 years ago
@zshaheen, I'm not going to be able to make a new version of e3sm-unified until this issue gets fixed. Is there any chance you could try a build of e3sm_diags
with a newer version of cdms2
and perhaps other cdat
packages?
@dnadeau4, @doutriaux1, didn't you guys have a similar issue like this? I think you guys said that for some reason, the newest version on conda-build
gave issues, specifically regarding cdms2
. Any input?
It's been fixed on conda-forge (not sure about the cdat channel). It's just that v.3.0.1 hasn't been rebuilt with the fix. Is the easiest solution to rebuild the old version with the new recipe? Is there a good reason not to update e3sm_diags
to a more recent version?
Digging into this a bit further, I now see that the latest release of cdat (8.0) only works with v3.0.1, so unless we're good with using a nightly build of cdat, there's not a good option to upgrade. I'll request a rebuild of 3.0.1
@xylar, I can build e3sm_diags
with the latest version of of cdms
, but I don't know if that would fix it. @dnadeau4, is whatever causing this issue be fixed in the later versions of cdms
?
I'm asking because building v3.0.1
with an older version of conda-build
worked fine for me. So this issue might not have been addressed. If not, like Xylar said, we'd need a rebuild.
I have a PR here for rebuilding: https://github.com/conda-forge/cdms2-feedstock/pull/27
I'm not saying that's the best option but it might work.
I'm asking because building
v3.0.1
with an older version ofconda-build
worked fine for me. So this issue might not have been addressed. If not, like Xylar said, we'd need a rebuild.
Yes, I can build a metapackage with just cdms2 ==3.1.2
just fine. That's because the unsupported syntax for specifying the versions of esmf
and esmpy
were removed in later versions. The problem is finding a version of all the other cdat packages that works with that version of cdms2
. If you have such a set, @zshaheen, I think that's by far the better solution and I can abandon my branch on cdms2-feedstock.
Okay, so I think I now understand the problem better. The latest conda
/conda-build
doesn't think that esmf=7.1.0r
satisfies esmf>=7.1.*
, presumably because it is a release candidate. Presumably, this is a new behavior and release candidates were handled okay before. I'll see if there's anything more I can figure out about this but currently there is no such release version of ESMF (or ESMPy) available. So conda
doesn't think cdms2
is supported.
Ah, there's already an issue about this: https://github.com/conda/conda/issues/8309
If I go back to conda=4.5.12
, the issue goes away when I use conda create...
. I may have to go back to an earlier conda-build
as well, we'll see. But maybe this is the solution for now.
@xylar For your info, it seems that conda=4.5.13
(i.e. the latest 4.5.x) was the latest version where it worked, and conda=4.6.1
was the first version where it didn't.
(By the way, sorry, I am not going to follow this thread.)
The conda recipe for
cdma2=3.0.1
has syntax that the latest conda-build doesn't support:The newest versions of
cdms2
onconda-forge
(3.1.2) don't have this problem but isn't compatible withe3sm_diags
.