Add -no-toolchain-stdlib-rpath flag.

[dan-zheng]

Add -no-toolchain-stdlib-rpath flag: the negative version of
-toolchain-stdlib-rpath.

Make -no-toolchain-stdlib-rpath be the default: use /usr/lib/swift as
default RPATH on Darwin platforms instead of toolchain standard library.

---

Adapted from #27206.

tensorflow branch requires the opposite default (use toolchain standard
library as RPATH) because some stdlib modules like TensorFlow do not exist in
/usr/lib/swift.

[dan-zheng]

Patch suggested by @brentdax, to minimize the diff between master and tensorflow branches.
I'm not sure how to add tests, suggestions welcome.

[beccadax]

To test this, I would add test cases to test/Driver/linker-rpath.swift which duplicate the existing x86_64-apple-macosx10.14 and x86_64-apple-macosx10.15 test cases but pass -no-toolchain-stdlib-rpath. Your change in the TensorFlow branch should probably require you to modify the originals of these test cases; just leave the copies testing the original results.

[dan-zheng]

To test this, I would add test cases to test/Driver/linker-rpath.swift which duplicate the existing x86_64-apple-macosx10.14 and x86_64-apple-macosx10.15 test cases but pass -no-toolchain-stdlib-rpath. Your change in the TensorFlow branch should probably require you to modify the originals of these test cases; just leave the copies testing the original results.

Done in 9a1b3f2.

I added -no-toolchain-stdlib-rpath test cases for x86_64-apple-macosx10.9 and x86_64-apple-macosx10.15, following existing tests in the same file for -no-stdlib-rpath and -toolchain-stdlib-rpath.

I added mirror tests in the tensorflow branch patch: 1ad2aa4.

[jrose-apple]

This seems all right to me. I'm not quite sure what the future of -toolchain-stdlib-rpath mode is on Apple platforms at all, but it seems like the same things we need for testing the master stdlib also apply to using the TensorFlow stdlib.

(A bunch of this might get reorganized too as part of the "generalize cross-compilation and SDK layouts" work described on the forums. But for now this is good.)

[beccadax]

We should settle the help text question, but otherwise this LGTM.

[dan-zheng]

This seems all right to me. I'm not quite sure what the future of -toolchain-stdlib-rpath mode is on Apple platforms at all, but it seems like the same things we need for testing the master stdlib also apply to using the TensorFlow stdlib.

(A bunch of this might get reorganized too as part of the "generalize cross-compilation and SDK layouts" work described on the forums. But for now this is good.)

Thank you for the helpful context.

---

Thanks to reviewers!

[dan-zheng]

@swift-ci Please test

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 0596b20

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 0596b20