trim dist

[JeffBezanson]

There is a lot of unnecessary stuff in make dist:

• static libraries. I don't think they are needed.
• readline libraries. I believe we statically link them so they shouldn't need to be distributed.
• 3 copies of libarpack named .so, .so.2, and .so.2.0.0. Might be a problem in the arpack build, copying instead of making links perhaps.
• 20MB of unicode test files. make install specifically runs make -C test/unicode. Why?

We could also remove the docs, since they are available online.

[tshort]

i prefer leaving the docs in, so they're usable locally (for searching for example).

[ViralBShah]

Yes, we should include the docs in dist, but rest all should be fixed. For docs, we should also include a pdf of the docs in the distribution.

[ViralBShah]

I think we can close this, since we got our major gains from generating the unicode files on the fly, rather than packaging them. @JeffBezanson Unless you strongly feel about all the other stuff you mention, could you close this?

[JeffBezanson]

No, we can't close this until we stop packaging multiple copies of libraries, or anything else that is not needed.

[ViralBShah]

For the static libraries, the simplest thing to do is to rm *.a in the directory containing the tarball. readline could also be deleted at the same time.
The multiple copies are probably due to the use of cp -R in make install.

[JeffBezanson]

cc @staticfloat
This is still happening. make dist ships 2 copies of openblas, 3 copies of fftw, 4 copies of GMP, etc.
see also #6711

[IainNZ]

related perhaps: I just downloaded rc-4 today (on OSX) and it was ~50MB bigger than a 60 day old nightly I had been using

[Keno]

We also recently added debug symbols for release builds. I like this when building from source, but for distributions it might make sense to strip them.

[tkelman]

Is there a point to having separate julia and julia-debug if both have debug symbols included? I think the osx nightly from 60 days ago also was using accelerate instead of openblas

[tknopp]

@tkelman: Sorry if this is a dump question but are the debug symbols in a release build really valid? How can that be when the optimizer moves statements wildly around...