Esmvaltool: [PI] Test ESMValTool with Iris 3.0.0rc0

[bjlittle]

The release candidate 3.0.0rc0 for iris is now available, see here.

I wanted to propose this issue as a parent issue [PI], to ensure,

• that esmvaltool testing has been performed with iris v3.0.0rc0
• capture and track any subsequent outcomes or follow-on tasks resulting from testing esmvaltool with the iris release candidate

I'm particularly keen that any changes required have enough time to be addressed by the iris development team before iris v3.0.0 is tagged and released. There is always the opportunity to issue a follow-on patch as an iris v3.0.1 point release, but I'm keen that v3.0.0 is as fit for purpose as possible for esmvaltool.

ESMValCore issue: https://github.com/ESMValGroup/ESMValCore/issues/378
ESMValCore pull request: https://github.com/ESMValGroup/ESMValCore/pull/819
ESMValTool issue: https://github.com/ESMValGroup/ESMValTool/issues/1862
ESMValTool pull request: https://github.com/ESMValGroup/ESMValTool/pull/1874

[valeriupredoi]

@bjlittle we've already started looking into this in ESMValCore, see https://github.com/ESMValGroup/ESMValCore/pull/819

[bouweandela]

I tested:

• examples/recipe_python.yml
• examples/recipe_preprocessor_test.yml

with the iris3 branch from https://github.com/ESMValGroup/ESMValTool/pull/1874 and compared the runtime and results with the master branch.

The runtime is similar and the resulting netcdf files are too, though there was some minor difference in the time axis of the multimodel statistics of diagnostic_1 of the preprocessor test recipe (different calendar chosen from input cubes) and differences at the numerical accuracy in the data, but I suspect that is related to how the multi model statistics preprocessor function orders the input cubes and not an iris issue.

[valeriupredoi]

So here's what I propose we do:

• we have tested quite extensively the integration functionality with iris3rc but apart from Bouwe's tests, we have not made much benchmarking vs iris2; in any case that's a bit of a fait accompli since we'll not keep using iris2 and the tests made by Bouwe sound promising;
• we need to test all Python recipes since there may be a creeping one using bits of iris2 that are not compatible with iris3;
• if we finish that before the next release date in February, we make a 2.1.1 release to include iris3 as dependency (and a bunch of other bugfixes);

How's this sound @bouweandela ?

[valeriupredoi]

we have now merged the branch https://github.com/ESMValGroup/ESMValTool/pull/1874 with production iris3 into master :partying_face:

[valeriupredoi]

if we hit issues related to diag-stuff and iris3 we will open individual issues for each (hopefully none) :+1: