Since TakeOwnCell doesn't put its data inside a MaybeUninit, this means that that data always has to satisfy the validity requirements of that type.
But since that value can be moved away, it might not.
For example, if we create a TakeOwnCell<Box<u32>>, take the box, and then deallocate the box, the data inside the empty TakeOwnCell would hold a pointer to freed memory, which makes it invalid.
I tried to confirm this with miri, but my miri tests passed. I'm not sure why - Maybe I missed something, or maybe this causes the violation to not be found because it's not marked dereferencable because of the UnsafeCell (But I also tried with an older version of miri and it also passed).
However If I didn't misunderstand anything this is still UB.
This can be fixed by putting the data inside a MaybeUninit, since moved away values are basically the same as uninitialized values.
Since
TakeOwnCelldoesn't put its data inside aMaybeUninit, this means that that data always has to satisfy the validity requirements of that type.But since that value can be moved away, it might not.
For example, if we create a
TakeOwnCell<Box<u32>>, take the box, and then deallocate the box, the data inside the emptyTakeOwnCellwould hold a pointer to freed memory, which makes it invalid.I tried to confirm this with miri, but my miri tests passed. I'm not sure why - Maybe I missed something, or maybe this causes the violation to not be found because it's not marked
dereferencablebecause of theUnsafeCell(But I also tried with an older version of miri and it also passed).However If I didn't misunderstand anything this is still UB.
This can be fixed by putting the data inside a
MaybeUninit, since moved away values are basically the same as uninitialized values.