No. The correct way to write that code is to use .spare_capacity_mut() to get a &mut [MaybeUninit<T>], then write your Ts into that using .write_copy_of_slice(), then .set_len(). And that will not be any slower (though obviously more complicated) than the original incorrect code.

Oh this is very nice, I think it was stabilized since I wrote said code.

write_copy_of_slice doesn't look to be stable. I'll mess around with godbolt, but my hope that whatever incantation is used compiles down to a memcpy

As I wrote in https://news.ycombinator.com/item?id=44048391 , you have to get used to copying the libstd impl when working with MaybeUninit. For my code I put a "TODO(rustup)" comment on such copies, to remind myself to revisit them every time I update the Rust version in toolchain.toml

In other words the """safe""" stable code looks like this:

    let mut data = Vec::with_capacity(sz);
    let mut dst_uninit = data.spare_capacity_mut();
    let uninit_src: &[MaybeUninit<T>] = unsafe { transmute(buf) };
    dst_uninit.copy_from_slice(uninit_src);
    unsafe { data.set_len(sz) };

That's correct.