Make logic better match inspiration implementations#3
Conversation
…f topology-correction code that's no longer needed as a result
| [[3, 3], [9, 3], [9, 9], [3, 9], [3, 3]] | ||
| ] | ||
| } | ||
| }, |
There was a problem hiding this comment.
This linework should only produce three polygons; this fourth one overlapped existing ones and was not correct.
| polygons | ||
| } | ||
|
|
||
| fn multipolygon_to_geojson_string(multi: &MultiPolygon<f64>) -> String { |
There was a problem hiding this comment.
I got rid of the print statement that was the only consumer of this code.
| /// Minimal reproducer for split-touching-hole-drop: a hole that touches | ||
| /// the shell boundary must be cleanly split. | ||
| /// | ||
| /// Depends on passes: 3 (infer_contained), 7 (carve_contained, second). |
There was a problem hiding this comment.
These correction passes are all gone, so the comments aren't needed.
| assert_eq!( | ||
| actual_canonical, expected_canonical, | ||
| "fixture mismatch: split_touching_two_points" | ||
| ); |
There was a problem hiding this comment.
There was a bunch of code handling the processing of improperly noded lines, but we actually expect input linework to be noded, same as other implementers, and the existence of this improperly noded fixture was keeping us from realizing the simplification that allows. So: consistent with the contract, now we nodify it ourselves, and then we can get rid of the fallback code.
2 participants
My initial pass at this work was a rough port that didn't quite produce perfect results, and subsequent passes were mostly hamfisted attempts at repairing the output to make it more correct. This PR attempts to much more closely match the upstream (JTS, GEOS, etc.) algorithmic approaches and uncovers some early errors (mostly mistakes on my part due to differences in winding-order conventions between GEOS and
geo, the latter of which follows OGC rules) that are now corrected.The result is that almost all of the post-processing to correct the incorrect topology the early stages produce can be eliminated: the output is correct from the get-go. This eliminates lots of code, and eliminates a bunch of quite-expensive point-in-poly and area calculations, which makes everything dramatically faster. Plus I'm pretty sure it's more correct (hooray).