d3.geo.path doesn't handle zero-area polygons correctly

[stvno]

d3.geo.path sometimes renders polygons where all points are in a straight line (so no inside) as an "infinite" area with a line shaped "hole". The polygons are created by TileStache from a postgis database and served as topojson. The polygons start out as valid RHR geometries and are turned into lines due to the shift from floating point space to integer space; see the drawing below.

If the order of the last three points in the topojson is changed the feature is drawn correctly. I've setup a page showing the difference here: http://research.geodan.nl/sites/vectortiling/test/faultytile.html

Is there a correct clockwise order for points on a line? If so it might be that TileStache is wrong here serving the points in the wrong order, but following the above drawing the points are in a valid order. I would think that if a polygon has an area of 0, the order of points doesn't matter and D3 shouldn't try to create an infinite area with a zero area hole in it.

[jasondavies]

Currently, “degenerate” polygons such as these cause problems for D3, because their winding orders are ambiguous and we need to improve the robustness of our winding order algorithm. I guess the most practical thing would be to treat such polygons as lines, rather than ignoring them completely.

[stvno]

I've been fiddling a bit and I discovered that the function d3_geo_pointInPolygon(point, polygon) returns 'true' where IMHO that is wrong, if you have a zero-area polygon there can't be a point inside. So I added a simple check whether or not the polygon had an area of zero. This solves this issue for me, but since I don't fully understand what the clipStartInside is supposed to mean, nor the other uses of pointInPolygon it might break other things.

You can check the fix here: https://github.com/stvno/d3/blob/master/src/geo/point-in-polygon.js

[mbostock]

if you have a zero-area polygon there can't be a point inside

As Jason alluded to, I believe the problem here is an ambiguity between a degenerate clockwise zero-area polygon (where the winding order is interpreted as intended) and a degenerate counterclockwise whole-sphere polygon (where the winding order is unintentionally inverted).

[jasondavies]

Exactly. But for such degenerate zero-area polygons, I think it would make more sense to interpret them as always being clockwise. After all, if you really need a whole-sphere geometry, you would simply use {type: "Sphere"}.

[mbostock]

Agreed.

[mbostock]

Fixed in 3.5.17; see #2784 by @lukasappelhans.