map.getBounds does not account for horizon visibility and low zoom

[arindam1993]

mapbox-gl-js version: 2.0.1, 2.0.0

browser: All

Steps to Trigger Behavior

1. Call map.getBounds at low zoom
2. Invalid LatLng objects are returned

Link to Demonstration

https://jsfiddle.net/c3hyfme0/2/

[rreusser]

This is not the same issue, but is maybe closely related enough that it would make sense to investigate at the same time: #10447

[rreusser]

The above jsdfiddle is pinned to 2.0.1, but I'm not able to reproduce the specified behavior. I see:

{
  _ne: {
    lat: 89.99998979458857,
    lng: 1052.6131744340337
  },
  _sw: {
    lat: -89.9792927518562,
    lng: -140.77483253940437
  }
}

which, though not clipped to [-90, 90] x [-180, 180], is not invalid. If I zoom out all the way, I get:

{
  _ne: {
    lat: 90,
    lng: 13132.517765333585
  },
  _sw: {
    lat: -90,
    lng: -348.6930098987299
  }
}

I expected "invalid" to mean NaN values, but I just see numbers which reflect the wrapping going on. IMO this might be the preferred behavior since it's easy enough to mod 360 (or more precisely, ((lng + 180) % 360) - 180)) if you want the wrapped values.

[rreusser]

@arindam1993, do you by chance recall any additional context regarding this behavior? Was it returning NaN? Perhaps the window aspect ratio affects the result?

[SnailBones]

I think we really have two separate issues here. One is the unbounded longitude values, which as you mention might not be a serious problem.

What's unexpected about those values in this particular case (with high pitch) is that the values don't seem intuitively in line with he horizon. Longitude values from -140 to 1053 (wrapping the Earth ~3 times) are strange when the map appears to display just North America and a few pixels of Africa. I'd expect 180 degrees of longitude spread and we're seeing almost 7 times that.

I believe this is the issue with horizon visibility that @arindam mentions in the title.

Another related issue is #10610, though this one seems to be about actual failing values like (NaN, NaN) rather than just unexpectedly large ones.