Build statically salt against libsodium

[benoitc]

Since libsodium is not widely deployed on the systems right now, it is useful to just embed libsodium in the sources and build statically against it. Also since salt is now built statically against libsodium it's easier to deploy with an erlang release.

For now the patch is only building libsodium on UNIX (and cygwin probably), build with MVS is not yet supported.

[freza]

Thanks, I like the idea of making salt more self-contained. Reviewing the changes now.

[benoitc]

@freza what's the status about this PR. Anything I can do?

[freza]

Hi, sorry been busy -- I'll merge over the weekend. Wanted to make sure libsodium tests are always run after build (unconditionally), and that 32bit vs. 64bit build is selected correctly.

[freza]

As you certainly noticed I've been reluctant to do this -- I wanted a self-contained library from the start, but just including a snaphost of libsodium at some arbitrary point in time didn't feel right.

Now I've finally found the right way to do this: http://tweetnacl.cr.yp.to/software.html
It's NaCl in portable C, one small source file plus one small header file.

So I'll look into integrating that, would be good to also carry over tests from full NaCl.

[benoitc]

Does it means you wil rewrite the nif from scratch?

[freza]

Of course not. TweetNaCl is API compatible with NaCl and libsodium, or so the paper suggests. So users can choose between small portable implementation that's included, or opt for faster external one.

[benoitc]

@freza did you start any work about it?