[RanaPriyansh]

Closes #712.
Closes #861.

Implement From<Cow<'static, [u8]>> and From<Cow<'static, str>> for Bytes. Borrowed values use static storage. Owned values use the existing Vec<u8> and String conversions.

Tests cover UTF-8 content, borrowed pointers, cloning and slicing, empty inputs, and owned pointer/capacity round-trips. The string fixture fails on baseline for the missing conversion and passes with the implementation. The tests use explicit Cow types where the two conversions make inference ambiguous.

Local validation: formatting, git diff --check, RUSTFLAGS=-Dwarnings bash ci/test-stable.sh test, and cargo hack check --feature-powerset --optional-deps --rust-version with Rust 1.57.0.

Pointer and capacity checks establish payload storage reuse for the tested cases. They do not establish zero total allocation. Nightly, Miri, sanitizers, cross-platform checks, and hosted CI remain separate.

Codex used.

[NfNitLoop]

I'm not a maintainer (or even a contributor. yet? 😅), but this looks good to me! Thanks for picking up the str variant too!

Only thing I can think of is, I wonder if there's a way to generalize this, so that it automatically picks up all Cow<'static, T> where both the borrowed and owned types already implement Into<Bytes>?

[RanaPriyansh]

Thanks for taking a look! I tried the generic version locally:

impl<T> From<Cow<'static, T>> for Bytes
where
    T: ToOwned + ?Sized + 'static,
    &'static T: Into<Bytes>,
    T::Owned: Into<Bytes>,
{
    fn from(cow: Cow<'static, T>) -> Bytes {
        match cow {
            Cow::Borrowed(borrowed) => borrowed.into(),
            Cow::Owned(owned) => owned.into(),
        }
    }
}

It compiles in place of the two impls, and the existing test suite plus the new Cow tests pass unchanged on Rust 1.94.1. Today it covers exactly the same types, [u8] and str, because those are the only T with From<&'static T> for Bytes. Its main gain is for downstream crates: a type that implements From<&'static T> and From<T::Owned> for Bytes would get Cow support automatically.

I kept the concrete impls because the blanket impl is a broader API commitment. It shows up in the docs as a where-clause rather than two concrete conversions. Any future From<&'static X> for Bytes would also extend Cow support to X. Starting with concrete impls leaves room to switch to the generic form later. Removing a blanket impl would be a breaking change. I haven't checked the generic version against the 1.57 MSRV.

I'm happy to switch if the maintainers prefer the generic form.

Codex used.