Explain default trait object liftime bounds.

[steveklabnik]

Fixes #31137

I feel a little weird about how short this is. It's really hard to not just say the exact rule, about T: 'a. It feels so arbitrary here, though.

r? @aturon @nikomatsakis

[nikomatsakis]

I'm not sure if this gets at the essence of the rule. For example, it doesn't example what happens with Box<Trait>. I think the rule is kind of aptly summarized as consulting, in order:

1. Lifetime bounds on the trait, if any. For example, trait Foo: 'static or trait Bar<'a>: 'a.
2. Lifetime of enclosing reference, if any. So, &Trait => &'a (Trait+'a)
3. Else, 'static.

[nikomatsakis]

Though I think there may be some subtleties about Ref<'a, Trait> -- I can't remember where we ended up on that (did we modify that case when we tweaked the default object lifetime rule?)

[steveklabnik]

@nikomatsakis updated, what do you think of this attempt?

[nikomatsakis]

@steveklabnik seems better. Something still gives me pause. I feel like the way the importance of the cases is the reverse of they are listed. That is, another way to think of it is "trait objects cannot contain borrowed data unless one of these cases is met":

• the trait declares a bound, e.g. trait Foo<'a>: 'a
• the object type appears underneath a reference, e.g., &Foo
• the object type has an explicit bound Box<Foo+'a>

It's just a matter of emphasis though. Thoughts?

[steveklabnik]

Ah, I like that.

[steveklabnik]

@nikomatsakis r?

[nikomatsakis]

feels weird to me that this sentence just... ends (and doesn't resume after the code snippet)

[steveklabnik]

@nikomatsakis after a very long time, nits addressed, i think.

[solson]

Why was this closed without merging? #31137 is still open as well.

[steveklabnik]

I believe I was cleaning up my branches and may have accidentally deleted it. It was also sitting in the queue for a very, very long time...

[solson]

@steveklabnik Would it be okay if I took your changes from this PR and submitted them as a new PR?