Are the `#![feature(layout_for_ptr)]` functions sound to call on invalid data pointers?

[repnop]

Reading the documentation for the #![feature(layout_for_ptr)] functions, it doesn't mention if it accepts invalid data pointer values (e.g. a pointer created with core::ptr::null()), but seems to suggest that it does:

If T is Sized, this function is always safe to call.

but there's a comment mentioning that the pointer must be valid inside the functions, e.g.:

rust/library/core/src/mem/mod.rs

Lines 375 to 378 in 456a032

	pub const unsafe fn size_of_val_raw<T: ?Sized>(val: *const T) -> usize {
	// SAFETY: the caller must provide a valid raw pointer
	unsafe { intrinsics::size_of_val(val) }
	}

I would think that the answer is yes since its operating on the metadata of the pointer and not the pointer value itself (and with my limited testing, this seems like the current behavior). However, in either case I think its worth explicitly mentioning the validity of the pointer in the documentation for the functions.

[memoryruins]

@rustbot label +A-docs +C-enhancement +T-libs-api +A-raw-pointers

[asquared31415]

playground link demonstrating that the metadata is read correctly despite the pointer being invalid in the traditional sense

This uses an unsized slice pointer though, where the requirements are

then the length of the slice tail must be an initialized integer, and the size of the entire value (dynamic tail length + statically sized prefix) must fit in isize.

[RalfJung]

but there's a comment mentioning that the pointer must be valid inside the functions, e.g.:

Yes, the caller must provide a valid raw pointer. Which values are valid for which type is documented in the reference.

despite the pointer being invalid in the traditional sense

The issue is that "(in)valid" are horribly overloaded terms. :/ But in this comment it is a technical term as explained in this blog post. Your pointer might be dangling, but as a raw pointer, it is still "valid" (as a reference, it would not be "valid" -- validity is a type-relative property).