encoding/wkb: POINT EMPTY encodes and decodes differently

[otan]

It seems as if POINT EMPTY when encoded to a WKB with encoding/wkb and then decoded will error with EOF as it tries to read more points, but there not any (with ReadFloatArray). Unfortunately this presents an annoying behaviour where Encode(Decode('POINT EMPTY')) != "POINT EMPTY')

From https://github.com/twpayne/go-geom/pull/161#issuecomment-631539451, it appears there's no WKB compliant way of representing POINT EMPTY.

Here are a few options we can do:

• Should POINT EMPTY error when decoding to a WKB?
• Should we return a POINT EMPTY if there are 0 bytes left to read? (not sure that's possible as the io.Reader has no BytesRemaining which is kind of annoying)?
• Should we introduce NaN, NaN syntax as previously put forward by https://github.com/twpayne/go-geom/pull/161?

I'm also for the argument that for EWKB, we should copy exactly what PostGIS does - after all, EWKB seems to be PostGIS specific. This backtracks my previous support for option 3 of https://github.com/twpayne/go-geom/pull/161#issuecomment-631539451.

[twpayne]

Iterating on this:

• I agree that encoding/ewkb should copy exactly what PostGIS does for exactly the reason you propose. Very good idea.
• Since WKB cannot represent POINT EMPTY, I think encoding/wkb should return an error cannot encode geometry or similar when asked to encode one. Similarly, since POINT EMPTY has no representation in WKB, encoding/wkb should under no circumstances return POINT EMPTY and instead return an error.
• encoding/ewkb could potentially wrap encoding/wkb, for example by translating POINT EMPTY into POINT(NaN NaN) and vice versa.

I'm idly wondering if we should create wkb.Encoder type with functional options, which would potentially allow a unification of the encoding/{wkb,ewkb,wkbcommon} code, but this requires more thought.