Fix many things have have been slightly broken in OSX

[jpakkane]

Basically this steals even more stuff from CMake's implementation:

• pad rpath entry to maximum size
• use absolute paths for rpaths

Still needs to be added:

• version flags (see macOS: add libtool compatible versioning and fix rpath issues #2577)
• do the otool magic on install

[mojca]

I tested this with a simple setup:

testbin.cpp

#include "testlib.h"
#include <iostream>
int main()
{
    std::cout << answer() << std::endl;
    return 0;
}

testlib.h:

#pragma once
int answer();

testlib.cpp:

int answer() { return 42; }

meson.build:

project('Test libraries on mac', 'cpp')

testlib = shared_library('testlib', 'testlib.cpp', install: true)
executable('testbin', 'testbin.cpp', link_with : testlib, install: true)

And running

meson --prefix /tmp/inst <sources>
DESTDIR=/tmp/dest ninja install
# as well as once without `DESTDIR`

I end up with

/tmp/inst/lib/testlib.dylib
/tmp/inst/bin/testbin

with

> otool -L /tmp/inst/bin/testbin
 /tmp/inst/bin/testbin:
	@rpath/libtestlib.dylib (compatibility version 0.0.0, current version 0.0.0)
	/usr/lib/libc++.1.dylib (compatibility version 1.0.0, current version 400.9.0)
	/usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1252.0.0)

>  /tmp/inst/bin/testbin
dyld: Library not loaded: @rpath/libtestlib.dylib
  Referenced from: /tmp/inst/bin/testbin
  Reason: image not found
Abort trap: 6

Note that the binary works as long as libtestlib.dylib is in the current folder which is why you don't notice the failure right away.

So basically this doesn't seem to work for me. (I would also like to occasionally keep the liberty to keep using absolute paths, but that's a lower priority. Current solution would also not work on 10.4, but well ...)

[jpakkane]

Can you try writing the same in CMake and trying it out? It would be interesting to know if and how they do it. Thanks.

[mojca]

I need some more time to figure out what's the proper CMakeLists.txt file since a very simple example results in an equally broken library. I've put the (so far incomplete) sources here.

[jpakkane]

Thanks. I'll merge this now because it improves things and does not seem to regress anything. We can fix the other issues in other MRs.