[clang] [PAC][clang] Add new features to pauthtest ABI (PR #113150)

[Daniil Kovalev via cfe-commits]

kovdan01 wrote:

> Driver options need to have some stability guarantee, while CC1 options can be more freely changed. So I'd be concerned if we commit to certain forms of driver options with known ergonomics issues. The Clang driver options should come last in the patch stack, with other patches reviewed first. If cc1 options are already available, it's not a blocker for end users to use them (experimental feature).

@MaskRay I totally get your point, but the actual option implementation, cc1 flags and even driver flags are already available in mainline for pretty long time. The only thing this patch does is making several options (which are already implemented at all the stages) enabled as part of pauthtest ABI.

> Have you thought about `-fptrauth-something=call,return,indirect-goto`?

We have an old issue for discussing that #97320, but it seems that there was no actual discussion. I don't mind using the approach you've suggested, but it needs to be discussed with guys from Apple since they are a big user of pauth stuff.

Given that this particular patch changes behavior only for pauthtest ABI (which is, by definition, intended to be used in test purposes), I believe it should be safe to merge that. I would be happy to discuss different approaches to implementing driver options in #97320, it just seems a bit unrelated to this particular PR.

https://github.com/llvm/llvm-project/pull/113150