Fix polar limits

[hoxbro]

Fixes #5396

[codecov-commenter]

Codecov Report

Merging #5397 (0400851) into master (8b2ee78) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #5397      +/-   ##
==========================================
- Coverage   88.07%   88.07%   -0.01%     
==========================================
  Files         301      301              
  Lines       61969    61991      +22     
==========================================
+ Hits        54579    54596      +17     
- Misses       7390     7395       +5     

Impacted Files	Coverage Δ
holoviews/plotting/mpl/element.py	90.08% <100.00%> (+0.05%)	⬆️
...iews/tests/plotting/matplotlib/test_elementplot.py	98.42% <100.00%> (+0.07%)	⬆️
holoviews/util/transform.py	85.25% <0.00%> (-0.75%)	⬇️
holoviews/core/data/pandas.py	93.30% <0.00%> (-0.10%)	⬇️
holoviews/tests/core/test_utils.py	99.20% <0.00%> (+<0.01%)	⬆️
holoviews/core/util.py	86.06% <0.00%> (+0.03%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[jlstevens]

Thanks for the quick fix!

The diff looks fine to me and the additional unit test gives me confidence. Could you explain why this bug manifested itself for the scatter element but not for a curve? And why did it seem to work up to a certain angle? I'm guessing the polar axis wasn't being defined for a full circle.

Why would the polar axes be ok for one element and not another?

Once my curiosity is satisfied, I am happy to merge. :-)

[hoxbro]

I don't think any of the plots worked there was always a slice missing.

holoviews/holoviews/plotting/mpl/element.py

Line 321 in 4c118b3

extents = self.get_extents(view, ranges)

Returns the following for the three examples in #5396:

a = [-0.53, -0.2, 5.83, 2.2]
b = [-0.52, -0.2, 5.72 2.2]
c = [0.0, -0.2, 5.3, 2.2]

So the reason why the Curve plot is fine is that the extents is not padded.

Then somewhere, a round-trip of going from coordinates back to radians is done, which can be faked with something like this np.arcsin(np.sin(x))) which for a gives l=-0.53 and r=-0.45, which gives the small slice. For b we get l=-0.52 and r=-0.56, which gives the almost full circle.

I hope this satisfies your curiosity.

[jlstevens]

I hope this satisfies your curiosity.

Yes thanks. merging!