MAINT: skip setting rpath if it's not needed

[FFY00]

I also fixed ELF.rpath, which is needed for the tests, and added a test for setting the rpath in the first place in this PR.

Fixes #125

[rgommers]

Thanks @FFY00. There are some issues shown by the CI results. The code change looks like an improvement, and will solve the most important part of gh-125. It looks like it won't yet solve the unnecessary dependency issue. That can be left for later if you prefer.

[rgommers]

Ugh, this still depends on other fixes:

patchelf: getting info about '/tmp/pytest-of-runner/pytest-0/test_unneeded_r"
 b"path0/plat.cpython-39-x86_64-linux-gnu.so': No such file or directory

That's because the test package it's built on is still xfailed:

@pytest.mark.xfail(reason='Meson bug')
def test_purelib_and_platlib(wheel_purelib_and_platlib):
    artifact = wheel.wheelfile.WheelFile(wheel_purelib_and_platlib)

    expecting = {
        f'plat{EXT_SUFFIX}',

This passes locally with Meson master. After the other open PR for tags is merged, it should pass with released Meson as well.

[FFY00]

As per https://github.com/FFY00/meson-python/pull/143#issuecomment-1261084168, I rebased this PR on main to get rid of the non-linear commit history as I'd prefer the PR to be rebase-merged.

[rgommers]

You just lost the commits that made this PR pass CI. I'll re-add them.

[FFY00]

Yeah, sorry, they got lost in the rebase. I have just added the fixes directly to the affected commits.

[FFY00]

Sorry 😅

[rgommers]

I think that's it, but hard to be sure since I had to manually re-apply fixes. Let's first ensure CI is green, and then if desired at the end rewrite history to the desired commit granularity.

[FFY00]

I have squashed the fix with the affected commit, the tests seem to be passing, are you okay with me pushing?

[rgommers]

Sure, go ahead. CI is green (codecov isn't relevant, and Sage isn't currently testing the right thing). So rewrite history as you wish and then merge it I'd say.