Port the changes from iris.fileformats.grib back into iris_grib.

[pelson]

I've done this with very little understanding of the original rationale for integrating iris_grib back into iris, so I may have missed a key detail.

[coveralls]

Coverage increased (+0.3%) to 83.786% when pulling 4d0efeb on dask_iris into 1dd3aa6 on master.

[pelson]

Note: I accidentally pushed this branch to scitools/iris-grib, rather than my own fork. Apologies for that.

[coveralls]

Coverage increased (+0.3%) to 83.786% when pulling 6c9fb54 on dask_iris into 1dd3aa6 on master.

[pelson]

For the record, the pertinent commit in Iris that brought iris_grib back in was: SciTools/iris@e45acf7#diff-3967991fd23406c94b677376dc007fd8

[pelson]

This looks like an oversight in iris.

[lbdreyer]

I've done this with very little understanding of the original rationale for integrating iris_grib back into iris

You have been shown this before but in case you forgot, there was an attempt to open the discussion about the future of iris_grib on the goole groups.

I don't understand the purpose of this PR. As far as I'm concerned iris_grib is dead

[pelson]

Thanks @lbdreyer. The link was helpful - I've written something there, and suggested we get together to discuss iris' direction in a slightly higher-bandwidth form. (doodle poll for date/time: https://doodle.com/poll/yaxm4pqfkus96sm2)

Suffice to say, I personally do not feel that the subject has been considered sufficiently to call iris-grib a dodo just yet...

[pelson]

The thread was: https://groups.google.com/forum/#!topic/scitools-iris-dev/lFw2_CnaEHU
The direction is for SciTools/iris-grib to be the canonical GRIB loading/saving capability for Iris.

In light of this, could this please get a review? @SciTools/iris-grib-devs

[pp-mo]

I'm happy that this is equivalent to code in Iris at "version 2.0a0", i.e. SciTools/iris@7ec22fa.
There is a small code difference between the load_convert.ellipsoid routines, between use of if and elif :
( compare : https://github.com/SciTools/iris-grib/blob/6c9fb54b01eeac9ebeeec4a5337223159b31efae and /iris_grib/_load_convert.py#L387 )
I'm satisfied that that, and other differences, are immaterial.

Note that this version only works with current iris master branch, at the "2.0a0" version (which is not a tag, to avoid problems with conda-forge : see above SHA for reference).
This "2.0a0 alpha release" version contains the 'dask nanmask solution' code : It is ephemeral + will not be made available via a tag or release.
So, we need to fix again soon wrt the imminent mergeback of the feature branch https://github.com/SciTools/iris/tree/dask_mask_array