Graticule errors under gnomonic projection. about d3-geo

[mbostock]

Yep. This is almost certainly a bug in clipping because the points on the graticule exactly fall on the clip circle. It goes away if you change the clip angle to 60.0001°, for example.

FYI, I probably won’t be able to fix this bug and #55 anytime soon, since it would require a much deeper understanding of Jason’s geographic clipping code than I currently have.

[Fil]

In intersect(a,b,two) there is the possibility to return;, which makes point2 = undefined on lines 15090 and 15102 of the current d3.v4 build and throws the error.

Either intersectshould never return undefined, or we must check that point2 is defined in those lines. With both solutions the graticule does display correctly.

Note that, when I log the value of t2 = w * w - uu * (cartesianDot(A, A) - 1); when the bug happens, it is ~ - 1e-16. Adding some epsilon to this determinant computation when it is negative seems like a possible solution:

- if (t2 < 0) return;
+ if (t2 < 0 && (t2 += epsilon) && t2 < 0) return;

[Fil]

My understanding of the code is that this test on t2 < 0 is here only to avoid throwing an error on the next sqrt because of a rounding error. There is no case where this is "really" negative.

The test can be replaced instead by:
var t = sqrt$1(Math.max(t2, 0));

The test was introduced in d3/d3@d93f45e .

Side note: there is an utf8 ⨯ in one of the comments.

[Fil]

Amazingly I couldn't reproduce this bug anymore on any newer browser (Firefox, Safari Chrome). However, launching an old copy of "Brave.app" that hadn't updated, the bug was reproducible. Fixed by #91