offset_of! overflows stack when called on BIG structs

[LoganDark]

I'm trying to initialize a struct on the heap, but field-offset just overflows the stack whenever I try to get the field offset xD

[Diggsey]

I'm guessing Self is an extremely large struct? Do you get the same stack overflow if you compile with optimizations enabled?

I think the answer here is going to be: don't use offset_of! with a struct so huge that it won't fit on the stack. In general such types are quite impractical to use with Rust anyway, since there's no stable in-place construction.

[LoganDark]

I'm guessing Self is an extremely large struct? Do you get the same stack overflow if you compile with optimizations enabled?

Looks like going into release mode fixes the stack overflow. So in order to test with debug assertions, I'd have to tweak an existing profile. I guess it works.

I think the answer here is going to be: don't use offset_of! with a struct so huge that it won't fit on the stack. In general such types are quite impractical to use with Rust anyway, since there's no stable in-place construction.

Isn't that a pretty big use-case for field-offset, though? Manually writing to struct fields that can't be accessed through a reference?

[Diggsey]

Manually writing to struct fields that can't be accessed through a reference?

I'm not sure what you mean by this? References can be used to access large struct fields, the problem is creating the struct on the heap without first blowing up the stack.

[LoganDark]

the problem is creating the struct on the heap without first blowing up the stack.

You cannot create a reference to uninitialized memory, that is undefined behavior. So you have to initialize the struct on the heap without creating a reference. In order to do this, to initialize memory, you have to write things into it. In order to write things into it, you need to know where the fields are that you're writing...

[Diggsey]

I see. Depending on the nature of your struct, there may be other ways to initialize it. For example, if it consists entirely of fields for which zero is a valid bit pattern, then you can initialize the memory with zeros, and then create valid references to fields within it.

The stack overflow is a problem with the underlying memoffset crate, see Gilnaa/memoffset#49 so it's not something I can fix here.

Also, there's actually no practical way to make this work without causing UB. Determining the offset requires a non-dangling pointer to the struct, and if we can't create it on the stack, then the only other option would be to allocate it on the heap, which means it wouldn't work without alloc, and it would be pretty bad performance wise.

[LoganDark]

if it consists entirely of fields for which zero is a valid bit pattern, then you can initialize the memory with zeros

That's actually what I needed! This works:

			let allocation = NonNull::new(alloc(Layout::new::<Self>()) as *mut Self)
				.expect("failed to allocate space");
			write_bytes(allocation.as_ptr(), 0, 1);
			Box::from_raw(allocation.as_ptr())

The stack overflow is a problem with the underlying memoffset crate, see Gilnaa/memoffset#49 so it's not something I can fix here.

Closing because it's an upstream issue then. Thanks!