I've written a custom Cow implementation that's meant to be (mostly) a drop-in replacement for std::borrow::Cow.

The #[derive(Deserialize)] macro currently does specialization for Cow by ident name when using #[serde(borrow)], which is smart, but it also means we get the false positives, as it is in this case:

use beef::Cow;

#[derive(Deserialize)]
struct Foo<'a> {
    #[serde(borrow)]
    foo: Cow<'a, str>,
}

Will produce following compile error:

#[derive(Deserialize)]
         ^^^^^^^^^^^
         |
         expected struct `beef::generic::Cow`, found enum `std::borrow::Cow`
         in this macro invocation

This PR modifies serde::private::de::borrow_cow_* functions to return a generic type R with a bound R: From<std::borrow::Cow>. It would allow any custom type named Cow that also implements From<std::borrow::Cow<T>> (which it probably should) to be consistent with std::borrow::Cow in deserializing behavior. This should obviously just add a no-op for std Cow.

Breakage?

All the usage of the function within the derive macros should continue working since types will be always easily inferred. Given this function is not part of public documentation I reckon this should be fine even though it's technically a breaking change. Thoughts?