Add support for rustls-platform-verifier

[djc]

We recently released the rustls-platform-verifier crate, and the rustls team believes this is probably the best option for most client use cases (especially on platforms like Windows, macOS, iOS and Android), since it reuses the platform verifier instead of more naively trying to find installed roots (which might be subject to additional constraints when used in the platform verifier).

This has been used for years inside the 1Password multi-platform client codebase, so it is more mature than it might seem at first glance.

I actually started adding support for this, but unfortunately it doesn't currently have great support for adding additional roots on some platforms, so the current form of feature additivity that reqwest uses (adding both webpki-roots and rustls-native-certs) doesn't work on some platforms (it only works on Linux, where it is needed the most because there is no real platform verifier).

I think this should probably become the default going forward.

[djc]

@seanmonstar any feedback?

[seanmonstar]

Seems like a good goal. You mentioned that it can't currently support the feature additivity that reqwest exposes. Will it at some point? Or is that a non-goal?

[cpu]

Will it at some point? Or is that a non-goal?

There's an existing issue upstream for the missing feature: rustls/rustls-platform-verifier#58 I think we're all in favour, it just hasn't been implemented. IMO knowing it's a blocker for reqwest would help motivate getting around to it 👍

[djc]

Not sure it should be a blocker for reqwest -- but yes, we do want to support it eventually. While it is opt-in for reqwest, I don't think it needs to be a blocker?

[seanmonstar]

Just to ease finding the status, as commented in #2286: it's blocked on being an additive feature.

[djc]

(I think the work to enable that is pretty much done upstream, just needs a release.)

[ofek]

Is this the proper way to use the library until there is an official feature? The documentation mentions that applications should call the install default method early in main but I can't find examples on applications doing so. I can only find an example from rustup doing it manually.

[djc]

Is this the proper way to use the library until there is an official feature? The documentation mentions that applications should call the install default method early in main but I can't find examples on applications doing so. I can only find an example from rustup doing it manually.

Yup, that looks fine.

[ofek]

Great, thanks! Is there an example of setting it globally at the start of an application so one does not have to create a client manually?

[djc]

Great, thanks! Is there an example of setting it globally at the start of an application so one does not have to create a client manually?

That's not a thing for the ServerCertVerifier, only for the CryptoProvider, see rustls/rustls#1821.