Checks for bbox CRS definition are generally invalid

[aaime]

Describe the bug BBoxCrsParameter#verifyBboxCrsParameterWithDefaultCrs performs two requests with two different bounding boxes:

• The first one in the "default CRS"
• The second one in the same bbox transformed onto a different, target CRS

The test then goes to check the list of returned feature ids in both cases, and checks they are the same set of ids.

However, there are a few issues in this:

• If the original BBOX covers the entire world (maxExtent, lin 144), then the bbox is "reduced a bit", 5 degrees per side. This reduction implies features returned by the original bounds might not be in the reprojected result. (Also, the test is very weak, for example, a Countries dataset will touch the south pole, due to Antarctica, but not the north pole, as there is not land there... leading to south pole being projected to infinite if one is using 3857).
• The requests performed are limited to the first 10 items. But it's not safe to assume that two different requests against the same spatial area will return the same first 10 features, depending on how the spatial filter execution is performed, the order of the results could be different (one should really check the entire possible set of features instead)
• As reported in the past already, envelope reprojection is a difficult matter, as the lines of the bbox could be turned into curves during reprojection, while the test suite is just reprojecting the corners of the envelope instead. This is especially true if the datasets involved have large area coverage (eventually the whole world), while this issue would be less of a problem over small areas (e.g, a city).

I don't believe the current test logic can be easily fixed. I would suggest to change it:

• Add the "crs" parameter so that the returned geometries are in the target CRS
• Verify that the returned features footprints are crossing the reprojected bbox

Even this will be subject to issues (e.g Antartica will fail to reproject to 3857).

Maybe the test suite should ask for a "safe bbox" for all layers instead, and use it to issue requests? Just thinking out loud, mind.

[dstenger]

Thank you for reporting. We will do further investigation and check if the logic of the test can be updated.