[traines-source]

• revert winding correction, because winding fixes are handled by clipper
• clipper's polygon tree sometimes attaches holes to the wrong outer ring. Assign holes to smallest outer.

testing planet on test pending

[felixguendling]

maybe this is better in Clipper2?
#21

Clipper2 would speed up rectangular clipping (currently we use the generic clipping for polygons, no special case for rectangles).

[traines-source]

It appears that this is not better in clipper2, and that we don't actually want to use the rectangular clipping because it makes the flooding worse, meaning the speedup is also lost. for coastline import clipper2 does appear to make sense (and speeds up 2x), so it is probably still worthwile to upgrade. i would then proceed to fix the clipper2 PR?

[felixguendling]

#21 might be in a rough state - it was just an experiment and numbers so far looked good.

and that we don't actually want to use the rectangular clipping because it makes the flooding worse

maybe the to_fixed_polygon causes this?

not sure why rectangular clipping makes the issue worse? The "Assign holes to smallest outer." strategy does not help?

[traines-source]

I haven't looked into it in detail, but: "It only cuts rings at the tile edge and doesn't remove self-intersections. I tested RectClip followed by an even-odd union in Clipper2 to clean up afterwards. Quality was the same (10 bad features, worst 0.025%), but it was only about 5% faster (11.4 s vs. 12.0 s), and its output differs in 237 tiles. That's not worth the extra step. "

The custom hole assignment is still needed anyways.

[felixguendling]

5% faster

since we run this also in prod (tiles endpoint), I would say 5% is better than nothing

but for the sake of progress let's just merge the fix asap :)