v2/client: move to async reqwest

[lucab]

Now that reqwest has an async module, we should use that instead of plain hyper. Unfortunately, this means switching from rustls to native-tls, thus getting a FFI dependency to openssl (or similar).

---

Progress

• src/v2/blobs.rs
• src/v2/tags.rs
• src/v2/manifest/mod.rs
• src/v2/auth.rs *: substitute hyper for reqwest #95
• src/v2/catalog.rs *: substitute hyper for reqwest #95
• src/v2/mod.rs *: substitute hyper for reqwest #95
• src/v2/config.rs *: substitute hyper for reqwest #95
• src/lib.rs *: substitute hyper for reqwest #95
• src/errors.rs *: substitute hyper for reqwest #95

[steveej]

Could you describe the benefits of switching to reqwest over using hyper? Do you expect the reqwest API to be more stable than the hyper one?

[lucab]

@steveej both are still in development and will likely see a major rework once futures land in stdlib. However hyper is the wrong layer to use when implementing an HTTP client, which means that we poorly re-implement here some middleware logic (e.g. following redirections).

This crate status quo is is purely a result of historical timing, as at that point there was no higher-level async HTTP client.

[steveej]

However hyper is the wrong layer to use when implementing an HTTP client, which means that we poorly re-implement here some middleware logic (e.g. following redirections).

This crate status quo is is purely a result of historical timing, as at that point there was no higher-level async HTTP client.

I understand your concern. If we need to extend the functionality of dkregistry we can refactor the code to use reqwest along that. Personally I don't have the urge to do it earlier, but of course I'm not stopping you from doing it ;-)

[steveej]

The close was unintentional via a stale Fixes .. in a PR.
Marking this as a bug because it causes the system certificate store to be ignored, which is the default behavior of hyper-rustls.