geoTransverseProjection?

[mbostock]

It’s pretty easy to do generically:

projection = d3.geoProjection((x, y) => d3.geoEquirectangularRaw(-x, y).reverse()).rotate([0, 0, 90])

[Fil]

Testing at https://observablehq.com/@fil/transverse-projections

[Fil]

A much simpler solution:

d3.geoEquirectangular()
  .rotate([0, 0, 90])
  .angle(-90)

Should we add it to https://github.com/d3/d3-geo/blob/master/README.md#projection_angle ?

[mbostock]

That worked for Atlantis, too:

https://observablehq.com/@d3/atlantis

Maybe it was just Mercator that needed this more elaborate solution because of its automatic clipping behavior.

[Fil]

Yes, I guess it should be requalified as a bug of projection.angle

[Fil]

The issue seems to lie in clipAntimeridianInterpolate which takes large strides along the antimeridian line. This doesn't work well later with clipExtent.

A way to fix it is to take smaller steps; but as it adds intermediate points, it breaks about a dozen unit tests which expect an exact path, and makes a heavier SVG path.

--- a/src/clip/antimeridian.js
+++ b/src/clip/antimeridian.js
@@ -68,17 +68,21 @@ function clipAntimeridianIntersect(lambda0, phi0, lambda1, phi1) {
 }

 function clipAntimeridianInterpolate(from, to, direction, stream) {
-  var phi;
+  var phi, e;
   if (from == null) {
     phi = direction * halfPi;
     stream.point(-pi, phi);
-    stream.point(0, phi);
+    for (e = -pi; e < pi; e+=0.01)
+      stream.point(e, phi);
     stream.point(pi, phi);
-    stream.point(pi, 0);
+    for (e = -halfPi; e < halfPi; e+= 0.01)
+      stream.point(pi, -direction * e);
     stream.point(pi, -phi);
-    stream.point(0, -phi);
+    for (e = -pi; e < pi; e += 0.01)
+      stream.point(-e, -phi);
     stream.point(-pi, -phi);
-    stream.point(-pi, 0);
+    for (e = -halfPi; e < halfPi; e+= 0.01)
+      stream.point(-pi, direction * e);
     stream.point(-pi, phi);
   } else if (abs(from[0] - to[0]) > epsilon) {
     var lambda = from[0] < to[0] ? pi : -pi;