Address new nightly unsafe precondition check panics

[Rafael Rivera (riverar)]

Passing size=0, ptr=0 into std::slice::from_raw_parts is undefined behavior and is no longer safe to rely upon. Also addresses an unaligned pointer used by Event.

See also:
rust-lang/rust#120902
rust-lang/rust#120594

[Chris Denton (ChrisDenton)]

Thank you for looking into this. It seems there's still one more failure:

thread 'add_remove' panicked at library\core\src\panicking.rs:155:5:
unsafe precondition(s) violated: slice::from_raw_parts requires the pointer to be aligned and non-null, and the total size of the slice not to exceed `isize::MAX`
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
thread caused non-unwinding panic. aborting.
error: test failed, to rerun pass `-p test_event --test tests`

Which indirectly points to

windows-rs/crates/libs/core/src/event.rs

Lines 161 to 167 in 40f671b

	fn as_slice(&self) -> &[Delegate<T>] {
	if self.is_empty() {
	&[]
	} else {
	unsafe { std::slice::from_raw_parts((*self.buffer).as_ptr(), self.len) }
	}
	}

But that looks totally fine on its own. The empty case is checked and I very strongly doubt the size exceeds isize::MAX. Which just leaves alignment of the buffer.

So looking at Buffer::new:

windows-rs/crates/libs/core/src/event.rs

Lines 205 to 216 in 40f671b

	fn new(len: usize) -> Result<*mut Self> {
	if len == 0 {
	Ok(std::ptr::null_mut())
	} else {
	let alloc_size = std::mem::size_of::<Self>() + len * std::mem::size_of::<Delegate<T>>();
	let header = crate::imp::heap_alloc(alloc_size)? as *mut Self;
	unsafe {
	header.write(Self(crate::imp::RefCount::new(1), std::marker::PhantomData));
	}
	Ok(header)
	}
	}

It uses HeapAlloc, which aligns to 16 (or 8 on 32-bit) but then it writes a header which, is essentially an AtomicI32. Meaning the data itself is only 4 byte aligned.

Quick fix here would be to add an #[repr(align(8))] to Buffer to add padding to the header. Something a bit more involved would be to fully account for the type's alignment when determining the offset of the buffer's data.

[Kenny Kerr (kennykerr)]

Thanks all. #[repr(align(8))] should do the trick for Event.

[Kenny Kerr (kennykerr)]

Thanks!

[Kenny Kerr (kennykerr)]

I'll probably publish an update to crates.io soon to avoid dependents having issues with nightly.

[Kenny Kerr (kennykerr)]

But we should probably get the feature doc search feature integrated before then. Let me know if there's anything else top of mind.