Closed valeriupredoi closed 3 years ago
This is complicated by the fact that two of Iris' dependencies - cf-units and nc-time-axis - are also 'broken' by cftime 1.3, but no resource has so far been committed to sorting this nor making new releases once fixed. I would very much like to get this sorted in the near future!
cheers @trexfeathers - we are happy that so far only this pin was needed and are proceeding to fix the tests that the pin breaks, further on things look good :+1:
Hey @valeriupredoi
Yeah, we've pinned cftime<1.3.0
for iris
master
and v3.0.0
. As @trexfeathers mentions we're keen to fix this, but may require a bit of effort on several packages - hence the pin.
However, if we do unpin, it may result in an iris
v3.0.1
release - on the other hand, we may just fold that goodness into iris
v3.1.0
, which is scheduled for the summer 2021 (we'll firm the proposed release date soon)
But if you can work around this pin, that would be appreciated :wink:
cheers @bjlittle - we completely understand the issues with dependency packages that evolve differentially, in other words, they need to get their conflicts act together so we or you have a healthy up-to-date environment too :rofl: cftime=1.2.1
as it is pinned now in our environment to accommodate iris=3.0
has not given us any headaches and it's been around for a while now, so I think this is a non-issue for us. I'll have a look at the actual iris
environment myself, but I doubt I can fix anything that you guys didn't :grin: But will let you know if I manage anything!
I have opened https://github.com/SciTools/cf-units/pull/168 for getting cf_units
up to speed with cftime
v1.5 π
Edit: and it looks like we have a new contributor who has done the same for nc-time-axis
: https://github.com/SciTools/nc-time-axis/pull/59 π
We're getting there. As indicated we are dependent on nc-time-axis and cf-units first, so those are getting the initial attention.
hey guys, as I see things now cf-units
is doing fine at 3.0.1 (cheers, Ruth!) but nc-time-axis
is still at the ye olde 1.2.0 - what does the status: blocked mean for this issue? Lemme know if I can help :+1:
We are almost there: once #4222 is merged and we release Iris v3.0.4, we will have a full set of Scitools packages that work with cftime v1.5. In the meantime, we have
So these are incompatible π’
Latest cf_units works with the older or newer cftime π
cheers muchly @rcomer - apols for the late reply, too hot to think anything more complex than "water (or beer)" :sweat:
@valeriupredoi I'm now closing this issue as iris
3.0.4 unpins cftime
, finally allowing you to use the latest iris
along with the latest versions of nc-time-axis
, cf-units
and cftime
:+1:
iris
3.0.4 is already on PyPI, and I'm currently in the process of building the conda
recipe for conda-forge
... so that'll land imminently.
Thanks @trexfeathers, @rcomer and @pp-mo for all helping to get this across the line :partying_face: :beers: :rocket:
π° Custom Issue
Hey guys, congratulations for releasing
iris=3.0.0
:partying_face: :tada: We have tried to install it in the ESMValCore environment so we can grab it for our upcoming release but hit a stumbling block - it needscftime
lower or equal to (an old version) 1.2.1, see comment - you see, we had previously pinnedcftime
to >=1.3.0 because we noticed some test fails due to new years conventions when years < 1000 ie they write "0999" instead of " 999", anyways, we can fix those tests, but the more unnerving bit is the environment will forever be stuck with an oldcftime
- are there any plans to unpin that from your environment requirements soon? Cheers :beer:PS: here is our fully solved environment as lady Conda on Circle CI solved it: