Cleanup make dist

[staticfloat]

• No longer install anything related to libuv with make dist
• Use custom install imposter; copy symlinks correctly to drastically reduce make dist size.

Tested on OSX and Linux (Ubuntu)

[tkelman]

would this be needed on freebsd too?

[staticfloat]

Looking at this again, I completely flummoxed this check. I have a much more resilient method in mind.

[tkelman]

Oh and for BSD's this shouldn't rely on bash being installed. Posix sh only.

[staticfloat]

Good reminder. I always default to bash for some reason. :P

[tkelman]

Travis says line 10 is a bashism

[staticfloat]

It wasn't the only one! :P I think I've got everything working now.

[JeffBezanson]

Sweet!

[tkelman]

Wait, do we want to install versioned symlinks at all? Maybe install is doing what we want it to but we're just calling it on things we shouldn't be?

[staticfloat]

It is best practice to install the versioned symlinks. I see no reason not to, since it's completely valid to try and dlopen libopenblas.so.0.2.10, and without these files here, that will fail.

[tkelman]

As long as the symlinks are all relative-paths, get skipped in the Windows build (I can't find any on master so I think we're okay), and get properly preserved by archives, I guess it's fine then.

[staticfloat]

Yes, they are definitely all relative. I think they don't get installed on Windows because the individual build processes for OpenBLAS, openlibm, etc... just don't create them on Windows. (At least, that's my guess without looking into it). Symlinks are preserved in .tar files.

[tkelman]

Oddly libgmp does make a symlink at usr/lib/libgmp.lib (funny that a GNU library would get their autotools so wrong there both in symlinking and the static library extension for mingw), but we don't install it or put that file in the binaries. MSYS2's tar has trouble with symlinks (see libgit2/libgit2#2526 for example), but there's not too much reason I can think of to need to work with the OSX/Linux binaries from within MSYS2.

[JeffBezanson]

So we will not have this for 0.3?

[staticfloat]

Ack, I thought this was merged days ago. We should probably have this to reduce distribution size, huh.

[staticfloat]

Let me squash this, and then I'll merge and do some rebase magic.

[ivarne]

Isn't this what we have milestones for?

[staticfloat]

Yes, I should have put this on a milestone.