Feature/upstream update

[eddelbuettel]

Also adds experimental Hamming distance support (closes #25)

[erikbern]

Just out of curiosity why do you include the full source code distribution of annoy here? Can you use a submodule or something similar?

[eddelbuettel]

Couple of reasons, in random order:

• I never really got comfortable with submodules
  • I see them used only sparingly in the R world
  • They seem a little out of fashion
• Having a degree of freedom allows me to add additonal defines etc
  • That has come in handy eg when I carried over CNPy for my RcppCNPy
  • R is older than Python and has fewer base types (ie natively only int in 32 bit and double in 64
  • So the mapping sucks
• I'm lazy, and Annoy is small enough so that I can just catch up once every few months
• I don't need all your files (ie can skip Lua, Go, ...)

[eddelbuettel]

Oh, and while I have you here: should Hamming work? I got weird results when I carried the unit test over and "parked" two of the tests. Will try to look more closely another day.

[erikbern]

thanks for the explanation!

Hamming should work, but it's admittedly not as thoroughly tested as the other distance functions... so I wouldn't be shocked if there's some wacky errors somewhere