Make reproject work even with data not loaded

[adehecq]

This is mostly in reply to issue #273.

[adehecq]

Sorry @erikmannerfelt, another conflict with your PR...

[erikmannerfelt]

Looking good! There's just a test that needs fixing. Because you change the default reprojection algorithm (why by the way? I agree with the change but you haven't said why bilinear should be default!), some tests fail sporadically because the values are no longer exact. For some reason that I can't explain, only the Windows tests fail!

Here's the line that fails in the Windows CI:

https://github.com/GlacioHack/GeoUtils/blob/c101c3ed1e7424e0492b453b08f34aaa52b0569a/tests/test_georaster.py#L508

If you change this to use np.isclose instead, it should work. It's comparing 13.048369 == 13.0 so a one-decimal tolerance should be sufficient.

[adehecq]

Looking good! There's just a test that needs fixing. Because you change the default reprojection algorithm (why by the way? I agree with the change but you haven't said why bilinear should be default!), some tests fail sporadically because the values are no longer exact. For some reason that I can't explain, only the Windows tests fail!

You're right, that is the reason for the tests failing. I will revert to the previous behavior and make the change in a separate PR, to be more transparent.
The test also need to be updated to pass only for nearest interpolation.