One option to handle this issue would be to use Option<&mut [u8]> instead of a plain slice in ops.

In V8, empty ArrayBuffers will (always, I think) return a null pointer for the data. In Rust, zero-sized slices can't be created from a null pointer, but they can be created from a dangling pointer. In fact, we fixed this same bug in the Deref implementation of v8::BackingStore by using dangling pointers: https://github.com/denoland/rusty_v8/blob/main/src/array_buffer.rs#L315-L320.

I opened denoland/rusty_v8#1131 to have ArrayBuffer::data return an Option<NonNull>, like BackingStore::data does. And with it, you could do:

let store = b.data().unwrap_or_else(NonNull::dangling).cast::<u8>();
// SAFETY: rust guarantees that lifetime of slice is no longer than the call.
unsafe { ::std::slice::from_raw_parts_mut(store.as_ptr(), byte_length) }

In V8, empty ArrayBuffers will (always, I think) return a null pointer for the data. In Rust, zero-sized slices can't be created from a null pointer, but they can be created from a dangling pointer. In fact, we fixed this same bug in the Deref implementation of v8::BackingStore by using dangling pointers: https://github.com/denoland/rusty_v8/blob/main/src/array_buffer.rs#L315-L320.

I opened denoland/rusty_v8#1131 to have ArrayBuffer::data return an Option<NonNull>, like BackingStore::data does. And with it, you could do:

let store = b.data().unwrap_or_else(NonNull::dangling).cast::<u8>();
// SAFETY: rust guarantees that lifetime of slice is no longer than the call.
unsafe { ::std::slice::from_raw_parts_mut(store.as_ptr(), byte_length) }

The unfortunate thing is that for FFI at least, we need the null pointer. A dangling pointer will cause the UnsafePointer.of static method to lie about the pointer address of a given buffer.

So, to some degree the solution cannot be quite so clear cut, we also need something done in the ops macro stage.

I think this is now fixed with #16826 and #16814.

I think this is now fixed with #16826 and #16814.

Yup, I think so. I rebased the PR, dropping out the asserts... Well, this PR essentially no longer serves a purpose I guess :D