Cesium: Black terrain artifact

[YousefED]

See screenshot below:

Steps to reproduce:

• http://cesiumjs.org/Cesium/Apps/Sandcastle/index.html?src=Camera.html&label=Showcases
• change setHeadingPitchRoll to function below
• run
• click "Set camera with heading, pitch, roll" from dropdown to run code

function setHeadingPitchRoll() {
    Sandcastle.declare(setHeadingPitchRoll);

    var camera = viewer.camera;
    camera.setView({
        destination : new Cesium.Cartesian3(1288857.3292931253, -4338373.750901034, 4488506.669268853),
        orientation: {
            heading : 0.1765,
            pitch : -0.4667,
            roll : 0
        }
    });
}

OS: Windows 10
Browser: Chrome 51.0.2704.103 (also seen on firefox)

[hpinkos]

Thanks @YousefED! I was able to reproduce this based on the steps you provided

[YousefED]

Fyi, I',m seeing this bug for more locations, e.g.:

I'm open to looking into this, but have little experience with the internal 3d engine, so I'm not sure where to start

function setHeadingPitchRoll() {
    Sandcastle.declare(setHeadingPitchRoll);

    var camera = viewer.camera;
    camera.setView({
        destination : new Cesium.Cartesian3(4471707.323258337, 675395.3016673529, 4486939.158518577),
        orientation: {
            heading : 5.942565929933654,
            pitch : -0.4636476090008035,
            roll : 6.283185307179586
        }
    });
}

[pjcozzi]

The culling volume may not be properly synced with the camera view.

@YousefED you are welcome to look into this, but it might not be easy first issue to track down (instead check out these).

Also, are you using Cesium for TweetBeam? Do you have a demo?

[mramato]

I swear we have a very similar bug opened for this already that involved flying the geocoder to specific locations, but now I can't find it.

[mramato]

Nevermind, I was thinking of #3676, which is already fixed. Sorry for the noise.

[YousefED]

@pjcozzi Nope, for Relive.cc - I think you've already seen it on Twitter, and we're in touch with your team re. a showcase :)

[YousefED]

@pjcozzi Sorry to "bug" you about this, normally wouldn't bring it up again; but the bug seems to affect one of our users on a ride with Richard Branson 😅 If a quick look from your side could help assess whether a quick fix could be possible (and where I should look), or whether it affects a large part of the render engine (in this case I probably can't help) - that would be extremely helpful.

[pjcozzi]

@YousefED sorry, I am fully committed for quite a while to come; however, I labeled this as "bug bash" so perhaps someone will take a look then: http://cesiumjs.org/2016/09/19/Cesium-Bug-Bash/

[duvifn]

@mramato you probably meant #4698 which is the same bug and they both happen in lat 45

[duvifn]

Steps to reproduce:
1. Type 101 45 in the geocoder.
2. Zoom out until the scene level is 16 (as reported by Cesium Inspector's Show Tile Coordinates).
3. Move the camera slightly toward south .
4. The tiles north to the latitude 45 should disappear.

Workaround:
Turn off the fog:
viewer.scene.fog.enabled = false

[duvifn]

The problem is that the south plane of TileBoundingBox (in the northern hemisphere) is not bounding the tile, as in the following animation.

As a result the distance that is reported by TileBoundingBox.distanceToCamera might be wrong by up to ~250 Kilometers with tiles of the lower levels.
Therfore the fog is culling them and so their descendants are not selected for rendering either.

In the southern hemisphere the north plane is the problem.

These specs are failing in master.
This plane calculation first appeared in Cesium-b10.

For my own needs I fixed this in this branch. If you think this fix is good also for Cesium I'd be glad to open a PR.
My solution is to calculate the plane in a slightly different way.

[hpinkos]

Thanks @duvifn! It would be great if you want to open a pull request with that fix.

[duvifn]

Thanks @hpinkos!
I opened a PR.

[YousefED]

Great stuff, thanks @duvifn