Modified attribute behaviour for netCDF save. #711
Conversation
| _CF_GLOBAL_ATTRS = ['conventions', 'history', 'title'] | ||
|
|
||
| # UKMO specific attributes that should not be global. | ||
| _UKMO_DATA_ATTRS = ['STASH', 'ukmo__um_stash_source', 'ukmo__process_flags'] |
There was a problem hiding this comment.
'STASH' seems unnecessary here as netcdf saving turns STASH into ukmo__um_stash_source
There was a problem hiding this comment.
I take it back. It turns out I did this for a reason. The cube.attributes dictionary key is 'STASH' not 'ukmo__um_stash_source'. I could in fact remove 'ukmo__um_stash_source' from this list, but I added it in case some one manually adds this attribute rather than using 'STASH'.
|
i would like to see an extra piece of functionality which enables a save operation to explicitly write global attributes to a file. This should be additive behaviour and can happily come in a new pull request |
|
I would like to see the iris.save documentation updated to give a little more help to users this can happily come in a new pull request |
| iris.save(self.cube, filename) | ||
| self.assertCDL(filename, ('netcdf', 'netcdf_save_attr.cdl')) | ||
|
|
||
| def test_conflicting_attributes(self): |
There was a problem hiding this comment.
Is this test really necessary? Are you not just testing dictionary behaviour here?
There was a problem hiding this comment.
Sorry. Missed the cube2 part. Doh.
|
Generally I have no problems with this. One thing that I really struggle with (as a user) is figuring out what the behaviour of the NetCDF saver is supposed to be without actually reading and understanding the fairly complex code. For instance, it'd be great to have some documentation on the behaviour of NetCDF attribute saving somewhere. Any thoughts on that? |
|
Yes I agree. The save documentation needs an overhaul. @marqh is working on a PR. |
|
@pelson - I've another commits of review actions. Let me know what you think. |
| # that should be attributes on data variables. | ||
| attributes = cubes[0].attributes | ||
| common_keys = set(attributes) | ||
| for cube in cubes[1:]: |
There was a problem hiding this comment.
Logic looks sound within this for loop (which is where all the complication relating to this PR exists). 👍
Modified attribute behaviour for netCDF save.
This PR modifies the treatment of cube attributes when saving to CF-netCDF. Currently cube attributes are saved as data variable attributes. However, this has the undesirable effect of routinely moving global CF attributes to data variable attributes when loading from, then saving to CF-netCDF. This PR modifies Iris so that it will save cube attributes as global, unless they are in a subset of CF-netCDF attributes that only apply to data variables (e.g.
standard_error_multiplier). When saving multiple cubes to a single file, only cube attributes that are common across the collection will be stored as global, the rest will be stored as data variable attributes (with warnings being issued when saving global-only or data variable-only attributes in the wrong place).