Specialized Option implementations

[blonk]

I watched jonhoo's stream and while I was distracted with other things I heard it mentioned that Option<Vec<T>> can be optimized so that it uses the vector's internal list pointer to indicate whether it's Some or None. Unfortunately because I was focused on other things, I didn't quite get the context under which this happens.

Are these types of quote-specializations-endquote something that is completely built into the compiler (i.e. it knows about Option<Vec<T>> specifically), or is this something one could hint the compiler about in library code? Conceptually:

struct MyThing {
  #[if_this_is_null_it_can_be_regarded_as_none]
  ptr: *const u32;
}

Or is it based on things like NonNull?

[cole-miller]

This is called a "niche" or "niche-filling" optimization. There's some information in the Unsafe Code Guidelines here:

https://rust-lang.github.io/unsafe-code-guidelines/layout/enums.html#discriminant-elision-on-option-like-enums

(Incidentally you can observe this in action quite easily, e.g. by verifying that size_of::<Option<Box<u32>>>() == size_of::<Box<u32>>().)

[cuviper]

@blonk

Are these types of quote-specializations-endquote something that is completely built into the compiler (i.e. it knows about Option<Vec<T>> specifically), or is this something one could hint the compiler about in library code? Conceptually:

struct MyThing {
  #[if_this_is_null_it_can_be_regarded_as_none]
  ptr: *const u32;
}

Or is it based on things like NonNull ?

NonNull uses unstable attributes for this:

#[stable(feature = "nonnull", since = "1.25.0")]
#[repr(transparent)]
#[rustc_layout_scalar_valid_range_start(1)]
#[rustc_nonnull_optimization_guaranteed]
pub struct NonNull<T: ?Sized> {
    pointer: *const T,
}

And there's similar on the NonZero* types.

[H2CO3]

So, no, it's not specialization, and yes, it's built into the compiler, but it's not hard-coded for Option.