[compiler-rt] [libcxxabi] [libunwind] [runtimes][PAC] Harden unwinding when possible (#138571) (PR #143230)

[Oliver Hunt via llvm-commits]

ojhunt wrote:

> > (if linux folk could retest to see how badly I have broken everything that would be great)
> 
> @ojhunt A couple of quick observations:
> 
> 1. It looks like that you forgot to add explicit declarations of copy constructor and assignment operator for `Registers_arm64`. Now, compiler complains about explicit definitions for these while they are implicitly defined.

Hmmm, what is the configuration you're building with? it's possible a mismerge, but also possible I just screwed up builds.

> 2. Why do we need `static_assert(!__has_extension(ptrauth_qualifier));` under ifdefs for `_LIBUNWIND_USE_DL_ITERATE_PHDR`, `_LIBUNWIND_SUPPORT_DWARF_INDEX` and `_LIBUNWIND_ARM_EHABI`? These are failing for me (particularly the first two ones), and I'm not sure I understand why these static assertions are needed.

I was verifying code I was building was being built as I expected, I'll remove those assertions - I realized while the original PR applied the `__ptrauth` qualifier to fields we don't have a testable configuration for, and I scattered a bunch of static asserts to verify what all our configurations were hitting because if they weren't I could not test them :-/ 

> 3. The PR title contains a reference to an old PR [[runtimes?][PAC] harden unwinding #138571](https://github.com/llvm/llvm-project/issues/138571), it's probably not needed

Righto

https://github.com/llvm/llvm-project/pull/143230