[clang] Allow the no_stack_protector attribute on local variables (PR #173311)

[via cfe-commits]

cooperp wrote:

> If I'm understanding correctly, if you specify this on a local variable, it doesn't directly impact how the stack protector is emitted, just the heuristic for whether we choose to enable the stack protector? i.e. it has no effect with -fstack-protector-all?

That's right yeah. For the `-fstack-protector-all` case for example i believe that is covered by the code in LLVM which looks for `F->hasFnAttribute(Attribute::StackProtectReq)`, which is done prior to any analysis on individual instructions. This PR and the equivalent LLVM one change that instruction based analysis.

> Maybe we should use a different name for this? Using an existing attribute name is convenient, but the meaning here seems different enough that we should use something different. (This also makes it easier for users can easily find relevant documentation.) Maybe stack_protector_ignore or something like that.

Yeah I have no problem with a new name.  Also considered something like `no_implicit_stack_protector` perhaps, but maybe `stack_protector_ignore` is better as otherwise folks have to think about what from my suggested name the "implicit" is even tied to.

Happy to introduce `stack_protector_ignore` and update this if you think that's the way to go.



https://github.com/llvm/llvm-project/pull/173311