Leaks

[197g]

parcode/src/reader.rs

Lines 902 to 918 in 1cdc49d

	// Prevent double‑free of the source items
	let src_items = ManuallyDrop::new(items);
	#[allow(unsafe_code)]
	unsafe {
	// Reconstruct pointers
	let dest_base = buffer_addr as *mut MaybeUninit<T>;
	let dest_uninit = dest_base.add(start_idx);
	// Cast MaybeUninit<T> -> T (Layout compatible)
	let dest_ptr = dest_uninit as *mut T;
	let src_ptr = src_items.as_ptr();
	// Efficient copy
	std::ptr::copy_nonoverlapping(src_ptr, dest_ptr, count);
	}
	Ok(())

This leaks memory, quickly. Also, definitely not zero-copy. Sidenote: The previous unsafe block was terse, easier to verify, and just as correct: https://github.com/RetypeOS/parcode/pull/1/files#diff-267c9f4da66412a9f439ac08d224356fe24265b5e1cebb6c44c2d55b96414513R393-R400 Just please tell your LLM about pointer::expose_provenance / with_exposed_provenance as alternatives to as usize.

---

Some aspects of this are neat, there's an idea where to take the project and what is necessary to make a usable API. I'd have more comments but I don't see myself spending more time reading code and writing issue comments than an LLM query takes to write. Siempre es necesario entender el código fuente, especialmente el que no escribiste.

[Retype15]

Thanks for review. You are right, ManuallyDrop on the vector leaks the heap buffer. We've fixed it by using set_len(0) to free the buffer without dropping elements, and updated the pointer logic to use .addr() and with_exposed_provenance_mut properly. This error is my fault, because have been used for testing core lines on main branch. Thank you to review and detect this problem!
#12