VoronoiTessellator._enforce_barriers splits Voronoi cells crossed by barrier lines — the feature the README describes as "enforce barrier features by cutting through cells". It selects which lines to cut with:
barriers_to_cut = self.cm.clean_lines[mask_barrier & mask_no_straddle & mask_no_quad_buffer]
That is: is_barrier=True and no straddle_width and no quad_buffer. Straddled and quad-buffered barriers are already aligned by the mesher, so they are correctly excluded.
Every is_barrier=True in the test suite is paired with quad_buffer=True:
tests/test_buffer.py:25 is_barrier=True, -> quad_buffer=True
tests/test_buffer.py:119 is_barrier=True, -> quad_buffer=True
tests/test_buffer.py:151 is_barrier=True, -> quad_buffer=True
tests/test_buffer.py:159 is_barrier=True, -> quad_buffer=True
tests/test_buffer.py:196 is_barrier=True, -> quad_buffer=True
tests/test_buffer.py:290 is_barrier=True, -> quad_buffer=True
tests/test_buffer.py:327 is_barrier=True, -> quad_buffer=True
All seven are filtered out by mask_no_quad_buffer, so the cell-splitting body never executes in the test suite — not the shapely.ops.split call, not the largest-piece-keeps-the-id rule, not the new-id allocation, not the except path.
The path does work; a manual run with is_barrier=True and no quad_buffer cuts cells as expected. It is simply unguarded against regression.
Why it matters now
The companion issue about max_id += 1 and uint64 promotion is a suspected NumPy 1.x bug living entirely inside this untested body. The minimum-dependencies CI job pins numpy==1.24.0 and would be the natural place to catch it — but it cannot, because nothing reaches the code.
Suggested test
def test_plain_barrier_splits_cells():
cm = ConceptualMesh(crs="EPSG:3857")
cm.add_polygon(Polygon([(0,0),(20,0),(20,20),(0,20)]), zone_id=1, resolution=4.0)
cm.add_line(LineString([(2,10),(18,10)]), line_id="fault",
resolution=2.0, is_barrier=True) # no quad_buffer, no straddle_width
...Assert that no cell geometry crosses the barrier, that ids stay unique, that node_id keeps an integer dtype, and that total area is conserved.
Pre-existing gap on develop — raised separately from #12, which adds 204 tests but does not cover this path.
VoronoiTessellator._enforce_barrierssplits Voronoi cells crossed by barrier lines — the feature the README describes as "enforce barrier features by cutting through cells". It selects which lines to cut with:That is:
is_barrier=Trueand nostraddle_widthand noquad_buffer. Straddled and quad-buffered barriers are already aligned by the mesher, so they are correctly excluded.Every
is_barrier=Truein the test suite is paired withquad_buffer=True:All seven are filtered out by
mask_no_quad_buffer, so the cell-splitting body never executes in the test suite — not theshapely.ops.splitcall, not the largest-piece-keeps-the-id rule, not the new-id allocation, not theexceptpath.The path does work; a manual run with
is_barrier=Trueand noquad_buffercuts cells as expected. It is simply unguarded against regression.Why it matters now
The companion issue about
max_id += 1anduint64promotion is a suspected NumPy 1.x bug living entirely inside this untested body. Theminimum-dependenciesCI job pinsnumpy==1.24.0and would be the natural place to catch it — but it cannot, because nothing reaches the code.Suggested test
Assert that no cell geometry crosses the barrier, that ids stay unique, that
node_idkeeps an integer dtype, and that total area is conserved.Pre-existing gap on
develop— raised separately from #12, which adds 204 tests but does not cover this path.