fitBounds broken

[wipfli]

maplibre-gl-js version: 4.7.0

browser: chrome on ubuntu

Link to Demonstration

Correct behavior in v4.6.0:

https://jsbin.com/rexiseyeve/1/edit?html,output

Broken behavior in v4.7.0 (zooms to wrong location):

https://jsbin.com/ninokegeto/edit?html,output

Expected Behavior

Should zoom to bounds

Actual Behavior

Zooms somewhere else, maybe the antimeridian?

Probably related to #4620

[HarelM]

Thanks for taking the time to report this issue!
I think the problem is that the "direction" of the bounding box is reversed, and so the entire planet is shown, which I tend to think is the right behavior to allow selecting the non-smallest area as part of the above PR.
Calling the following fitBounds will do the requested behavior I believe (note that I have changed the order of the longitude between the two coordinates):

map.fitBounds([
        [9.031869499999999, 46.8322365],
        [9.289097166666666, 47.19989433333333]
      ], {
        padding: 50,
        maxZoom: 14,
      });

[wipfli]

The docs says

A LngLatBounds object, an array of LngLatLike objects in [sw, ne] order, or an array of numbers in [west, south, east, north] order.

so I guess you are right. But the change in behavior is still a bit inconvenient when updating the MapLibre version. If I understand correctly the old behavior was give me two points and I will make the map show both. So not exactly bounds but very convenient...

[wipfli]

If I take two points and put them in an order, they might have one of the following four meanings for bounding box corners

• [sw, ne]
• [ne, sw]
• [nw, se]
• [se, nw]

If only the first option should be given to fitBounds, how can I convert the other 3 to the first?

[wipfli]

OK I can just do this myself with something like this:

const getBoundingBox = (point1, point2) => {
  const [lng1, lat1] = point1;
  const [lng2, lat2] = point2;

  const sw = [Math.min(lng1, lng2), Math.min(lat1, lat2)];
  const ne = [Math.max(lng1, lng2), Math.max(lat1, lat2)];

  return [sw, ne];
};

and then I call

map.fitBounds(getBoundingBox(point1, point2));

That gives me roughly the v4.6.0 behavior I think.

[HarelM]

Exactly.

[wipfli]

Feel free to close this issue @HarelM. But maybe we can add a link to this workaround in the changelog. What do you think?

[HarelM]

I think adding some code example is a better approach, or adding an example in the examples with this code and linking to it.
Linking to a GitHub issue is less preferred from my point of view.
Closing... Feel free to submit a PR.

[wipfli]

But then again the above function does not work across the antimeridian. So tricky...