CStr unchecked constructor can be a const fn

[abonander]

The current implementation of CStr::from_bytes_with_nul_unchecked() is not valid in a const context, but it can easily be tweaked so that it is:

pub const unsafe fn from_bytes_with_nul_unchecked(bytes: &[u8]) -> &Self {
    union BytesToCStr<'a> {
        bytes: &'a [u8],
        cstr: &'a CStr,
    }
    
    BytesToCStr { bytes }.cstr
}

[oli-obk]

At this point you can just enable the const_transmute feature gate and use transmute. There's no need for unions anymore

[abonander]

@oli-obk I thought transmute was explicitly not going to become const as per the discussion in #49450.

[abonander]

Alternatively, why can't we make DST constructors work for simple wrappers like this? Why can't it be, e.g.

&CStr{ bytes: *bytes }

Where it looks kind of like the opposite of a deref coercion, and the compiler just reinterprets the fat pointer? This would eliminate the need for any unsafe besides what's necessary for the API contract.

[oli-obk]

Yes I would like such constructors, too. But that needs and RFC and probably has a lot of caveats that we want to think about.

I did not know about that issue. I commented there.