Travis ran and failed, I don't know why it didn't post. https://travis-ci.org/github/SciTools/cartopy/builds/671037232

These were changed in 2f5e568, which I mostly reverted (x&y are the directions of the CRS, not necessarily lat/lon). Maybe we should revert these attribute name changes as well?

On the other hand, since gridlines now do bounding box intersections, it's not really guaranteed that x or y will hit just top/bottom or left/right. I'm not sure if we should warn more explicitly about it in the other setters.

I agree, projected coordinates/axes will make this much more confusing with what do x/y and left/right mean?

MPL uses ytick.right and ytick.labelright:
https://matplotlib.org/3.1.0/gallery/ticks_and_spines/tick_label_right.html

I personally think that ylabel_left (the older style) is actually doubly declaring the axes, i.e. can I don't think xlabel_left is even an option there. So, it is probably cleaner to just put label_left in there, rather than with the x/y qualifier.

I'm not tied to any particular attribute names, but think they should be consistent throughout the docs.

I think these names might be okay, but can you add the warning on the getter as well?

Sure, added that now.

Actually, looking at #1475, I think we might have too few options, but I'm not sure how best to specify things. On the one plot that is almost-rectangular, there is one latitude label that appears on the left side. So theoretically, we should have left/right/top/bottom options for x & y, but on the other hand we have some projections that don't really have defined sides.