[PATCH] D45639: [Driver] Support default libc++ library location on Darwin

[Louis Dionne via Phabricator via cfe-commits]

ldionne added a comment.

In https://reviews.llvm.org/D45639#1243017, @kristina wrote:

> I think on Darwin it would **not** make sense to have `libc++fs.a` ship in `libc++.dylib` especially considering that it ends up in the dyld cache and that has a lot of other implications. It would make sense to ship it as a separate library, perhaps as part of the SDK, at least for now.


Indeed, this makes a lot of sense and this is exactly what I was thinking about. I'd like to get a more holistic plan before we do that though.

> As far as making it a system dylib, it's a possibility as long as no core os components depend on it and it's there solely for consumers.

We can't do that unless that dylib is ABI-stable. Otherwise, imagine what happens when we update the OS (which would then contain a new version of that dylib): all consumers that link against that dylib will now crash at runtime because we'll have broken its ABI.


Repository:
  rC Clang

https://reviews.llvm.org/D45639