AaronBallman wrote: > The pointer values in a specific `va_list` can and often do change when you > call `va_arg`, but I am not aware of any targets where they do not always > point to the same regions. In every ABI I've ever seen, `va_list` contains a > pointer to the stack argument area, and those pointers _must_ point there > because the stack argument area cannot be either copied or relocated. > `va_list` may also contain one or more pointers to separate "spill areas" > where the callee is required to spill certain argument registers, which it > generally does in its prologue. In theory, a compiler could produce multiple > copies of that spill area, or an ABI could require it to be copied by > `va_copy` or even allocated on the heap. However, there's no good reason to > do either of those things: the `va_list` can never validly escape the > variadic function because of the limitation on the stack argument area, and > so the spill area might as well also go in a unique location on the stack.
Ah! Thank you for the explanation, that makes a lot of sense. > I have no objection to requiring va_copy and documenting primitive copies as > undefined behavior. It is the easiest thing to do, and we can clearly revisit > it later if we want. Thank you for the clarification. So you're happy with the current state of the PR, or are there some changes you'd like to see made? https://github.com/llvm/llvm-project/pull/98146 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
