nikic wrote: > Otherwise, it's not. We are working with pointers result from int2ptr, so p+1 may point to a different object. Waaat? Surely int2ptr cannot bypass the provenance model like that? https://github.com/llvm/llvm-project/pull/158062