[llvm] [MemCpyOpt] Require writable object during call slot optimization (PR #71542)

[Nikita Popov via llvm-commits]

nikic wrote:

> does alive model these sorts of issues?

I'm not really sure. I believe it doesn't reason about data races and doesn't support `noalias`, but maybe it has read-only memory? I think this would be a variant of a test case that doesn't rely on noalias: https://alive2.llvm.org/ce/z/ZDcarT And it doesn't verify ... but I'm not sure whether it fails to verify for the right reasons. I never know how to interpret these "Function triggered UB" results. cc @nunoplopes 

https://github.com/llvm/llvm-project/pull/71542