I would suggest keeping one backport module per external module. So either cbook._backports_numpy, or if want to overengineer it, cbook._backports.numpy (anyways it's private :-))

I know right now it's only numpy (and I don't particularly hope that there would be more stuff) but from .cbook import _backports as _backports_np just seems wrong if _backports is supposedly generic.

Otherwise, +1 to the idea.

I'd hope that we only have numpy backport, but we do also have backports of the subprocess modules in the compat module. The problem is that we need a deprecation cycle to move it to the same place (it would have be nice to have that module private, but it ain't).

Note : @tacaswell it seems that the subprocess module has been renamed in 2.x without deprecation warning. Is that normal? If it is normal, shouldn't we move it to the cbook module to have everything in one place?

(other solution for 2.x, make it private, this way we can do whatever we want with it. But it's really short notice for the release…)

Please don't over engineer this 😄

I am +0.5 on this. Positive on the extra organization, but concerned of the added indirection going from a single file -> folder and the slight lose in ability to do git forensics (which maybe important for cbook as @anntzer sorts through what of that we are actually using and what of that we can deprecate).

@NelleV renaming the module was probably a mistake, where did that get done? It looks like it was renamed in f41000d which was pre 1.3, but I might be missing something.

mmh, maybe I am wrong. I am looking at git blame, and thus a filename change would not be visible in the blame.

I fixed @anntzer's comment.

We really need to fix our PEP8 compliance tests system… It's raising errors on code I have not modified.

You've renamed lib/matplotlib/cbook.py → lib/matplotlib/cbook/__init__.py but forgot to add the new name to the exception list. But fixing the PEP8 errors is better, anyway.

That's exactly my point. We should not have to do this.

Unfortunately, there's no such thing as a rename in native git, so I'm not sure looking at a diff will work either. I might not be trying it out right, but I wasn't able to get flake8-diff to work properly on the changes in this PR.

Current coverage is 62.20% (diff: 84.61%)

Merging #7830 into master will increase coverage by 0.03%

@@             master      #7830   diff @@
==========================================
  Files           174        175     +1   
  Lines         56078      56247   +169   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits          34862      34987   +125   
- Misses        21216      21260    +44   
  Partials          0          0          

Powered by Codecov. Last update 4c5fca7...aca81ac

Thanks!