Will Barnes

@jessevdk I've successfully run cldoc under 3.6 with the same fixes that @fknorr has applied here. Do you know what the timeline might be for merging this fix and doing...

Thanks for the detailed response @jrbourbeau! I suspected there may have been some sort of special casing for serializing numpy arrays, but had no idea of how to diagnose that....

Things that should be moved from `sunpy.util` up into the top `sunpy` namespace: - `Metadict` - `NoMapsInFileError` - `SunpyDeprecationWarning` - `SunpyMetadataWarning` - `SunpyPendingDeprecationWarning` - `SunpyUserWarning` I may have missed some....

Yep so could `Metadict` actually

And the original astropy issue: astropy/astropy#4935 though the consensus seems to be this is an upstream issue with Sphinx. Either way, not unique to SunPy.

One possible way of doing this is with the `make_lupton_rgb` function in astropy (e.g. this gist https://gist.github.com/wtbarnes/62ef2995369a17eead6c3ce331985cd0). I think this is a good idea. I wonder if it would be...

I wonder if this could be folded into the PR template itself, with maybe some links to the relevant parts of the dev guide.

Where would this check happen? Do you mean when constructing the default normalization on `GenericMap`? Since we construct a colormap normalization for each source individually, I'm just a bit confused...

We are actually still using `exp_time` by default instead of the FITS-4 `xposure` key for exposure time.

Another motivator for this proposed change: this function uses `skimage.transform.warp` and thus rewriting it is a necessary step in removing our dependency on `scikit-image`.