CODEWITHSUNDEEP
rustlifetime
raj
raj

Reputation: 6094

How can lifetime of reference larger than the lifetime of the type its pointing to?

I am playing with this code to to understand lifetimes and elision better. This code compiles.

struct Bytes<'a>(&'a [u8]);

impl<'a> Bytes<'a> {
    fn first_two<'b>(self: &'b Bytes<'a>) -> &'b [u8] 
        where 'b: 'a
    {
        &self.0[..2]
    }
}

How can the where bound 'b: 'a be acceptable for self ? Shouldnt it always be the other way for self to be valid?

And the return value points to slice inside Bytes whose lifetime is 'a . So how can the output have lifetime 'b if 'b can outlive 'a?

Upvotes: 3

Views: 605

Answers (1)

user4815162342
user4815162342

Reputation: 155246

It's important to notice that the code compiles even when you remove the where clause altogether, which should remove any relationship between the two lifetimes. Why is that allowed?

I think what happens is that the very existence of a self: &'b Bytes<'a> parameter implies the constraint of 'a: 'b. For any Bytes<'a> value to exist in the first place, the caller must ensure that 'a referenced by Bytes lives at least as long as the Bytes value itself - hence 'a: 'b. In other words, returning &'b [u8] actually reduces the lifetime compared to the original 'a, which is always allowed.

To return to your code, when you further add 'b: 'a, you are telling Rust that in addition to the requirement of 'a living at least as long as 'b, also 'b has to live at least as long as 'a. In other words, 'b and 'a must be the same lifetime - which again makes it ok to return &'b [u8], as it's the same as &'a [u8].

How can the where bound 'b: 'a be acceptable for self? Shouldnt it always be the other way for self to be valid?

Precisely - it should be the other way around for self to be valid, and that makes it ok to return &'b [u8]. As argued above, adding the seemingly "inverse" constraint doesn't make the first one go away, it just makes the relationship stricter. (It's like knowing that a >= b for two numbers, and further stipulating that b <= a - the two just boil down to a == b.)

Upvotes: 4

Related Questions