Implementation of UGRID (part 1 of 2)

[davidhassell]

Fixes #696

Changes

• Implementation of new constructs
• Implementation of mesh regridding
• Implementation of mesh collapsing
• Refactor to move weights code out of the Field class
• New Data methods arctan2 and masked_values

New data model constructs that describe meshes

• cellconnectivity.py
• domaintopology.py

Misc.

• auxiliarycoordinate.py Inherit __repr__
• bounds.py Inherit __repr__
• cellmeasure.py Inherit __repr__
• cfdmimplementation.py Support for the new constructs and their (compressed) data
• count.py Inherit everything from cfdm
• dimensioncoordinate.py Inherit __repr__
• domain.py Inherit __repr__
• domainancillary.py Inherit __repr__
• field.py Inherit __repr__; move underscore methods higher up the file
• fieldancillary.py Inherit everything from cfdm
• index.py Inherit everything from cfdm
• interiorring.py Inherit everything from cfdm
• interpolationparameter.py Inherit everything from cfdm
• list.py Inherit everything from cfdm
• nodecountproperties.py Inherit everything from cfdm
• partnodecountproperties.py Inherit everything from cfdm
• tiepointindex.py Inherit everything from cfdm

Collapse

• field.py Move weights-related functionality to weights.py
• weights.py Worker functions for cf.Field.weights that have been refactored out of field.py. Includes new methods for UGRID mesh weights, and bug-fixes the calculation of geometry cell weights.

Regrid

• regrid.py Extend to allow the regridding of data on UGRID meshes.
• regridoperator.py New mesh-related attributes; refactored tosparse so that it also works with UGRID meshes.

Data

• data/data.py New arctan2 and masked_values methods; new masked_value keyword to __init__; Update the calculation of regridded chunksizes to allow for dropped and new axes.
• data/utils.py New normalize_chunks function to replace the dask one.
• data/dask_regrid.py Update to allow to the regridded data having more/fewer axes than the source data; Refactor the adjustment of weights to account for missing data.
• data/collapse/dask_collapse.py Move the checking of weights to compute time, away from definition time in Field.collapse.
• A whole bunch of changes for representing mesh arrays found in UGRID netCDF files, following the compressed array framework

[sadielbartholomew]

About half-way through (by file number total) eyeballing the code changes. Here are the (minor) comments so far. Functionality review TODO.

[sadielbartholomew]

More minor comments, including those below. I am mostly done with the functional review and eyeballing the code, but still need to look over the regrid and weights modules. I'll submit my final review on Monday.

---

1. Some new methods are missing from the API reference at present:

$ ./check_docs_api_coverage 
Method cf.Field.del_mesh_id not in docs/source/class/cf.Field.rst
Method cf.Field.get_mesh_id not in docs/source/class/cf.Field.rst
Method cf.Field.has_mesh_id not in docs/source/class/cf.Field.rst
Method cf.Field.set_mesh_id not in docs/source/class/cf.Field.rst
Method cf.Domain.del_mesh_id not in docs/source/class/cf.Domain.rst
Method cf.Domain.get_mesh_id not in docs/source/class/cf.Domain.rst
Method cf.Domain.has_mesh_id not in docs/source/class/cf.Domain.rst
Method cf.Domain.set_mesh_id not in docs/source/class/cf.Domain.rst
Traceback (most recent call last):
  File "docs/source/check_docs_api_coverage.py", line 77, in <module>
    raise ValueError(
ValueError: Found 8 undocumented methods and 0 missing .rst files

2. There are some of my old commented-out references to arctan2 still floating about. We should remove them to tidy whilst we are finally adding our Dask-enabled arctan2:

$ pwd
/home/sadie/cf-python/cf
$ git grep "AT2"
data/data.py:    # `arctan2`, AT2 seealso
mixin/propertiesdata.py:    # `arctan2`, AT2 seealso
mixin/propertiesdata.py:    # `arctan2`, AT2 seealso
mixin/propertiesdatabounds.py:    # `arctan2`, AT2 seealso
mixin/propertiesdatabounds.py:    # `arctan2`, AT2 seealso

[davidhassell]

1. Some new methods are missing from the API reference at present:

Ah yes, that's deliberate, as noted in NCAS-CMS/cfdm#272 (comment):
The new recording of a "mesh_id" (netcdfready.py, fielddomain.py, field.py) is forward looking to the time when we write these things out (see NCAS-CMS/cfdm#271). However, it's possible that when it comes to it we find that we don't need it. As a result it is deliberate that, for now, the various X_mesh_id methods remain undocumented in cfdm/docs/

2.There are some of my old commented-out references to arctan2 still floating about

71c6e93

[sadielbartholomew]

Some new methods are missing from the API reference at present:

Ah yes, that's deliberate, as noted in NCAS-CMS/cfdm#272 (comment):
The new recording of a "mesh_id" (netcdfready.py, fielddomain.py, field.py) is forward looking to the time when we write these things out (see NCAS-CMS/cfdm#271). However, it's possible that when it comes to it we find that we don't need it. As a result it is deliberate that, for now, the various X_mesh_id methods remain undocumented in cfdm/docs/

Aha, apologies I do remember that from reviewing the cfdm PR but it slipped my mind here. All good, in that context.

RE (2), thanks for removing the old refs. Also all good now.

About to finish the review...

[sadielbartholomew]

Leaving my final batch of minor comments, as usual mostly inevitable typos appearing across lots of new text.

Overall, I am happy, so please merge once you have considered the batch of comments here. I've done a sanity check at this stage after some new post-review commits and everything works as it should, but am happy to do another later before we merge if you wish.

[sadielbartholomew]

All feedback addressed really well, thanks. I have just pulled down the latest state of the PR and run the test suite locally, with all passes. As far as I am concerned, this is all ready to merge so please do the honours if you agree!

[davidhassell]

Great - thanks again for your reviewing!