Ellipsoid.geodeticSurfaceNormal returns wrong result if the cartesian is not on the ellpsoid surface

[duvifn]

I encountered this while working on #4622.

Sandcastle for this issue

The following code log false to console:

var latitude = 45.0;
var longitude = 90;

var cartographic = Cesium.Cartographic.fromDegrees(longitude,latitude);
var cartesian = Cesium.Ellipsoid.WGS84.cartographicToCartesian(cartographic,new Cesium.Cartesian3());
var surfaceNormal =  Cesium.Ellipsoid.WGS84.geodeticSurfaceNormal(cartesian,new Cesium.Cartesian3());

cartographic.height= 2000000;
var cartesian2 = Cesium.Ellipsoid.WGS84.cartographicToCartesian(cartographic,new Cesium.Cartesian3());
var surfaceNormal2 =  Cesium.Ellipsoid.WGS84.geodeticSurfaceNormal(cartesian2,new Cesium.Cartesian3());


//log 'false'
console.log(Cesium.Cartesian3.equalsEpsilon(surfaceNormal,surfaceNormal2, Cesium.Math.EPSILON8));

However, Ellipsoid.geodeticSurfaceNormalCartographic returns accurate result in this case:

var cartographic = Cesium.Cartographic.fromDegrees(longitude,latitude);
var surfaceNormalFromCartographic = Cesium.Ellipsoid.WGS84.geodeticSurfaceNormalCartographic(cartographic);

cartographic.height= 2000000;
var surfaceNormalFromCartographic2 = Cesium.Ellipsoid.WGS84.geodeticSurfaceNormalCartographic(cartographic);

//log 'true'
console.log(Cesium.Cartesian3.equalsEpsilon(surfaceNormalFromCartographic,surfaceNormalFromCartographic2, Cesium.Math.EPSILON8));

[hpinkos]

Thanks @duvifn

[pjcozzi]

For performance, this might just be a doc fix so that geodeticSurfaceNormal requires a height of zero. @duvifn do you have another suggestion?

[hpinkos]

@pjcozzi if that's the case, shouldn't we throw a DeveloperError if the height is greater than zero?

[pjcozzi]

I am suggesting that the performance for the conversion is not worth it.

[duvifn]

Thanks @pjcozzi and @hpinkos for reviewing this.

@duvifn do you have another suggestion?

Not really. Maybe additional parameter that explicitly indicates if the input value is on the surface.

Anyway, please consider reviewing of the code that currently uses this function (js and glsl). I saw some places where the input is not necessarily on the ellipsoid surface.

I believe that for most of the needs the bias is not significant, but there can be places in which this could lead to error