Sanity check: Transform results are always 64-bit floats

[djhoese]

I was adding some dtype checks to some tests that use pyproj and just realized something that should have been obvious a long time ago: does the transformation of an array of coordinates (Transformer.transform) always result in np.float64 results? Or is there a way to provide 32-bit floats and get 32-bit floats back? Or is the loss of precision so great that that would be foolish?

[jjimenezshaw]

It depends on the precision you want. For a surveyor it makes no sense to use float32 for instance with UTM.

[snowman2]

It is a double internally, so returning float64 gives you the same information that was calculated by PROJ.

Probably opening a can of worms to try to return the same dtype as the precision isn't dependent on the input array dtype.

[djhoese]

That works for me. Expect double for all inputs is an easy enough thing to code for my use case.