Prepare for Numpy 2.0 #2338
Prepare for Numpy 2.0 #2338
Conversation
lib/cartopy/mpl/gridliner.py
Outdated
| if longitude != 180: | ||
| # Wrap the longitude to the range -180 to 180, keeping positive 180s | ||
| longitude = ((longitude + 180) % 360) - 180 |
There was a problem hiding this comment.
There's a minor difference here in that 180 + 360 *n was formerly considered +180, whereas now it'll be -180.
There was a problem hiding this comment.
True, I was ignoring the larger multiples. I think we need an intermediate variable to handle that as before, I just pushed up a new change.
There was a problem hiding this comment.
Thinking more generally here... this is only in the W/E labels function. Do we actually want a label on the 180s, or should we just strip that off altogether like we do for 0 already? Remove the E/W ambiguity.
There was a problem hiding this comment.
Yea, that might be something to do instead. I feel like there are probably differing opinions on which one to use, too.
There was a problem hiding this comment.
OK, there was one image failure with that update, so I pushed the change to remove the E/W label as that simplifies the logic a bit. I'm happy to remove that and go back to the previous logic if preferred too.
I looked up a few examples and it seems there are versions with and without the labels on 180, so no standard I could see at a quick glance.
Matlab does produce E/W labels on 180: https://www.mathworks.com/help/map/the-map-grid.html
Images without the E/W labels on 180: https://manoa.hawaii.edu/exploringourfluidearth/physical/world-ocean/locating-points-globe
Wikipedia seems to indicate you should label both the prime and antimeridian with either E or W: https://en.wikipedia.org/wiki/180th_meridian
The longitude at this line can be given as either east or west.
There was a problem hiding this comment.
I was too undecided. I now went to the old version and kept the E/W labels as they were so we aren't changing anything anymore. This is the simple change for now to get ready for numpy 2.0 and we can re-evaluate the label naming in a follow-up PR later.
There was a problem hiding this comment.
I think we'd eventually want to go for something "smarter", that adds the direction if there are (potential for) two 180 labels. But this can be a separate PR.
I installed the latest pre-release of numpy 2.0 to see if we had any failures appear and these were the only ones that showed up. Pretty simple fixes to all internal/test machinery.
Here is the quick command to help others wanting to test this out by upgrading some local packages as well if wanted.
python -m pip install numpy matplotlib contourpy --index-url https://pypi.anaconda.org/scientific-python-nightly-wheels/simple --pre --upgrade numpy