impl Clone for Cow

[japaric]

Now we can use #[deriving(Clone)] on structs that contain Cow.

r? @aturon or anyone else

[japaric]

Just want to add that due to the bad interaction of struct bounds and the deriving syntax extension, in the case of CowVec<T> you are forced to bound T: Clone with a where clause:

This compiles:

use std::vec::CowVec;

#[deriving(Clone)]
struct Foo<'a, T: 'a> where T: Clone {
    bar: CowVec<'a, T>,
}

fn main() {}

But this does not:

use std::vec::CowVec;

#[deriving(Clone)]
struct Foo<'a, T: 'a + Clone> {
    bar: CowVec<'a, T>,
}

fn main() {}

See #19361 and #19358 for more details

[Gankra]

I must confess that it feels weird to me to always into_owned a Clone of an owned Cow. I imagine that always producing a Borrow would have unfortunate consequences? Can you provide an example?

[japaric]

If Self = SendStr = CowString<'static>. Cloning that can't return a Borrowed value, since that would need an inner &'static str, and you can't produce &'static str from String.

[Gankra]

Should we explicitly document the behaviour, since it's subtle and might be unexpected? Or do you think it can be considered an implementation detail?

[alexcrichton]

Another option would be to require T: Clone instead of B: BorrowFrom<T> so you could just call .clone() on the owned variant (which may be more expected)

[japaric]

require T: Clone instead of B: BorrowFrom

Note that the Cow struct enforces the B: BorrowFrom<T> bound, so it can't be removed. I wrote the impl in this way to reduce the numbers of bounds on the input parameters.