Avoid possible divide-by-zero in Vectors layer thumbnail update

[aganders3]

Description

Trying to run tests locally I ran into nine tests in layers/vectors/_tests/test_vectors.py tests consistently failing:

FAILED napari/layers/vectors/_tests/test_vectors.py::test_no_args_vectors - RuntimeWarning: divide by zero encountered in divide
FAILED napari/layers/vectors/_tests/test_vectors.py::test_no_data_vectors_with_ndim - RuntimeWarning: divide by zero encountered in divide
FAILED napari/layers/vectors/_tests/test_vectors.py::test_empty_vectors - RuntimeWarning: divide by zero encountered in divide
FAILED napari/layers/vectors/_tests/test_vectors.py::test_empty_vectors_with_property_choices - RuntimeWarning: divide by zero encountered in divide
FAILED napari/layers/vectors/_tests/test_vectors.py::test_empty_layer_with_edge_colormap - RuntimeWarning: divide by zero encountered in divide
FAILED napari/layers/vectors/_tests/test_vectors.py::test_empty_layer_with_edge_color_cycle - RuntimeWarning: divide by zero encountered in divide
FAILED napari/layers/vectors/_tests/test_vectors.py::test_no_data_3D_vectors_with_ndim - RuntimeWarning: divide by zero encountered in divide
FAILED napari/layers/vectors/_tests/test_vectors.py::test_empty_3D_vectors - RuntimeWarning: divide by zero encountered in divide
FAILED napari/layers/vectors/_tests/test_vectors.py::test_empty_data_from_tuple - RuntimeWarning: divide by zero encountered in divide

I will spare the repeated output because it was basically the same for each:

_________________________________________ test_no_args_vectors _________________________________________

    def test_no_args_vectors():
        """Test instantiating Vectors layer with no arguments"""
>       layer = Vectors()

napari/layers/vectors/_tests/test_vectors.py:40: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
napari/layers/vectors/vectors.py:248: in __init__
    self._update_dims()
napari/layers/base/base.py:704: in _update_dims
    self.refresh()  # This call is need for invalidate cache of extent in LayerList. If you remove it pleas ad another workaround.
napari/layers/base/base.py:1210: in refresh
    self._update_thumbnail()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <Vectors layer 'Vectors' at 0x287fdd7c0>

    def _update_thumbnail(self):
        """Update thumbnail with current vectors and colors."""
        # calculate min vals for the vertices and pad with 0.5
        # the offset is needed to ensure that the top left corner of the
        # vectors corresponds to the top left corner of the thumbnail
        de = self._extent_data
        offset = (np.array([de[0, d] for d in self._dims_displayed]) + 0.5)[
            -2:
        ]
        # calculate range of values for the vertices and pad with 1
        # padding ensures the entire vector can be represented in the thumbnail
        # without getting clipped
        shape = np.ceil(
            [de[1, d] - de[0, d] + 1 for d in self._dims_displayed]
        ).astype(int)[-2:]
>       zoom_factor = np.divide(self._thumbnail_shape[:2], shape).min()
E       RuntimeWarning: divide by zero encountered in divide

napari/layers/vectors/vectors.py:682: RuntimeWarning

The root of the problem is the undefined behavior in casting a NaN to int because self._extent_data is filled with NaN when the data is empty.

# natively on my Apple Silicon Mac
>>> np.float32("nan").astype(int)
0

# in an (emulated) x86-64 Ubuntu container on my machine
>>> np.float32("nan").astype(int)
-9223372036854775808

It looks like this or something like it was fixed or anticipated for the Shapes layer, so I copied the implementation style from there. I checked some of the other layer types and they look okay as well, though I'm still getting familiar with them.

Type of change

• Bug-fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

References

See numpy/numpy#14412
I think this is actually set to emit a warning in a coming version of numpy, and potentially raise an error.

How has this been tested?

Vector tests pass with my change on multiple platforms.

Final checklist:

• My PR is the minimum possible work for the desired functionality
• I have commented my code, particularly in hard-to-understand areas
• N/A I have made corresponding changes to the documentation
• N/A I have added tests that prove my fix is effective or that my feature works
• N/A If I included new strings, I have used trans. to make them localizable.
  For more information see our translations guide.

[codecov]

Codecov Report

Merging #5192 (6e9a71a) into main (2566980) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main    #5192   +/-   ##
=======================================
  Coverage   88.81%   88.81%           
=======================================
  Files         574      574           
  Lines       48910    48912    +2     
=======================================
+ Hits        43441    43443    +2     
  Misses       5469     5469           

Impacted Files	Coverage Δ
napari/layers/vectors/vectors.py	97.95% <100.00%> (-0.40%)	⬇️
napari/components/experimental/chunk/_pool.py	85.71% <0.00%> (-7.94%)	⬇️
napari/layers/image/image.py	95.89% <0.00%> (+0.24%)	⬆️
napari/utils/theme.py	89.79% <0.00%> (+0.68%)	⬆️
napari/utils/info.py	83.33% <0.00%> (+1.11%)	⬆️
napari/layers/image/_image_slice.py	100.00% <0.00%> (+2.38%)	⬆️
...layers/image/experimental/_chunked_image_loader.py	96.00% <0.00%> (+8.00%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[psobolewskiPhD]

Thanks for addressing this!
Nice detective work as to the real cause and why it passes on CI but crops up on arm64 macs!
We'd wondered about it before:
https://napari.zulipchat.com/#narrow/stream/212875-general/topic/failing.20test.20on.20latest.20main

Tested locally, all vector tests on my M1 pass and
viewer.add_vectors() no longer results in 3 warnings.
👍

For anyone reviewing, the diff is hard to follow, no idea why it's generated like that. Easier to look here perhaps:

napari/napari/layers/vectors/vectors.py

Lines 667 to 674 in dddbdcf

	def _update_thumbnail(self):
	"""Update thumbnail with current vectors and colors."""
	# Set the thumbnail to black, opacity 1
	colormapped = np.zeros(self._thumbnail_shape)
	colormapped[..., 3] = 1
	if len(self.data) == 0:
	self.thumbnail = colormapped
	else:

Those are the added lines, with the two colormapped lines being moved from 699-700. Then the rest is moved into the else: as is, with no changes but the indentation and black formatting.
VS Code does a better job of presenting it:

[brisvag]

Wow, that diff really is bad! Thanks for the screenshot @psobolewskiPhD, that helps a lot :D

@aganders3 nice catch, thanks for the PR! Looks good to me.

[aganders3]

Thanks for the reviews, and thanks @psobolewskiPhD for the clarification on the diff!

[psobolewskiPhD]

Everything looks good to me.
Edit: sorry didn't know how to do this properly—still learning!

[brisvag]

@aganders3 we have some merge conflicts, do you want to have a go at them?

[aganders3]

Thanks for the heads-up. The only difference I found was that self._dims_displayed became self._slice_input.displayed. I will request re-review if/when CI passes.

[andy-sweet]

As @psobolewskiPhD mentioned in Zulip, this is fairly niche, so I'm going to remove the v0.4.17 milestone from this to keep the RC changes small.

Thanks again for the fix @aganders3!