Dan Baston

No, only the ones listed in the exactextractr docs.

 Overall this change causes no performance regression, and a small gain in a few cases.

> What needs to be done to make things "2d-safe"? Need to stop reading 3D coordinates (`Coordinate`) by reference without checking whether the sequence is actually 3D or 4D. Not...

> Can we mostly start reading CoordinateXY by reference instead? It's always safe to read `CoordinateXY&`, though the assumption that we have a `Coordinate&` shows up in a surprising number...

Here's another test I did: - Removed the part of this PR that copies an external XY buffer to XYZ. This should give best case performance but could crash if...

About a 10% gain on `ST_Centroid`. I have never been too worried about `ST_Centroid` as a bottleneck, but maybe better performance for very fast functions could still be useful for...

It makes me nervous despite seeing the tests pass. I would rather creation sites clearly indicate that they do not require initialization by using the constructor that provides this parameter.

Yes, see https://github.com/libgeos/geos/blob/4c3bd72ba73e9688b0712d9878d073559c84e4d6/include/geos/geom/CoordinateSequence.h#L85-L96

I disabled precompiled headers, since they were causing problems outside of Linux/gcc that I don't have time to track down. Remaining failures are: MSVC: ``` cl : command line error...

I suspect the "real solution" to improving the build time is to use a unity build, but that doesn't play well with TUT's use of typedefs: ``` typedef test_group group;...