Explicitly set nopython=False in numba.jit

[robna]

Hello! Please make sure to check all these boxes before submitting a Pull Request
(PR). Once you have checked the boxes, feel free to remove all text except the
justification in point 5.

1. You have run tests on this submission locally using pytest on your changes. Continuous integration will be run on all PRs with GitHub Actions, but it is good practice to test changes locally prior to a making a PR.
2. This pull request is directed to the pysal/master branch.
3. This pull introduces new functionality covered by
   docstrings and
   unittests?
4. You have assigned a
   reviewer and added relevant labels
5. The justification for this PR is: just doing what the deprecations warning says: @jit decorators in alpha_shapes.py that are currently without any arguments now explicitly specify nopython=False

Closes #526

[codecov]

Codecov Report

Merging #535 (0281ca6) into main (a2e3902) will increase coverage by 0.1%.
The diff coverage is 100.0%.

@@           Coverage Diff           @@
##            main    #535     +/-   ##
=======================================
+ Coverage   79.9%   79.9%   +0.1%     
=======================================
  Files        113     113             
  Lines      13135   13133      -2     
=======================================
+ Hits       10491   10498      +7     
+ Misses      2644    2635      -9     

Files Changed	Coverage Δ
libpysal/cg/alpha_shapes.py	87.2% <100.0%> (ø)
libpysal/cg/voronoi.py	78.6% <100.0%> (-0.4%)	⬇️

... and 1 file with indirect coverage changes

[knaaptime]

thanks for this!

[robna]

It was apparent that some functions inside alpha_shapes.py already carry a @jit decorator with explicit nopython=True

I.e.:

libpysal/libpysal/cg/alpha_shapes.py

Lines 77 to 78 in a2e3902

	@jit(nopython=True)
	def r_circumcircle_triangle_single(a, b, c):

libpysal/libpysal/cg/alpha_shapes.py

Lines 129 to 130 in a2e3902

	@jit(nopython=True)
	def r_circumcircle_triangle(a_s, b_s, c_s):

libpysal/libpysal/cg/alpha_shapes.py

Lines 682 to 683 in a2e3902

	@jit(nopython=True)
	def _construct_centers(a, b, radius):

These are not changed by this PR. However, someone more knowledgeable in numba and in this project should review and decide, whether the lines where I now specified nopython=False should better also receive a nopython=True.

[robna]

I added one more commit to the PR regarding another deprecation warning:

voronoi.py imports geopandas (if available) and creates geodataframes, where a 'geometry' column is manually assigned. This will lead to errors with future versions of geopandas.

libpysal/cg/voronoi.py:173: FutureWarning: You are adding a column named 'geometry' to a GeoDataFrame constructed without an active geometry column. Currently, this automatically sets the active geometry column to 'geometry' but in the future that will no longer happen. Instead, either provide geometry to the GeoDataFrame constructor (GeoDataFrame(... geometry=GeoSeries()) or use `set_geometry('geometry')` to explicitly set the active geometry column.

Instead, the geometry is now specified in the gdf creation via gpd.GeoDataFrame(geometry = gpd.GeoSeries( ...

Please review,
thank you.

[martinfleis]

nopython=True is recommended and we should use nopython=False only if the former fails to compile. (source). So I would try to set them all to True and see if CI complains.

[knaaptime]

the default is switching from nopython=False to nopython=True, so these are the ones that raise the warning. In the upcoming versions of numba, the default will switch so the tests will fail for anything that requires python (I played around with this in esda yesterday, and i actually couldn't induce any failures when setting nopython=True, and havent been able to reproduce pysal/esda#240)

[robna]

actually, both, @jit (without arguments) and @jit(nopython=False) are raising a warning.

NumbaDeprecationWarning: The 'nopython' keyword argument was not supplied to the 'numba.jit' decorator. The implicit...

and

NumbaDeprecationWarning: The keyword argument 'nopython=False' was supplied. From Numba 0.59.0 the default is...

But also for me, no matter what way I set them, there are no severe problems. Just the warnings.

[knaaptime]

explicit is better than implicit, so we'll want to include nopython=False anyway. I vote to merge

[martinfleis]

Let's give it a go. We can always revise that if it causes issues (which it should not imho).