add valid_until to Client

[goenning]

fixes #1675

I found 3 different ways of doing this, but I'm not sure which one is the best.

This PR implements Option 1, but I'm fine with other options too (except maybe 3, my least favorite)

• Option 1: Fetch the identity_pem during the make_generic_builder and pass that onto both rustls and openssl HttpsConnector functions as an arg. This is a breaking change for those who are NOT using the make_generic_builder function.
• Option 2: Keep the exec_identity_pem call in both rustls and openssl connector builder as an arg, and bubble up the expiration datetime to the make_generic_builder function. This is also a breaking change for those who are using the make_generic_builder function as the return type for the HttpsConnector functions will change.
• Option 3: Call exec_identity_pem twice, once in the make_generic_builder function just to get the expiration date, and again inside the rustls and openssl HttpsConnector functions (as it is currently). This is NOT a breaking change, but it is inefficient as we are calling the exec_identity_pem function twice as this is a blocking system call.

[goenning]

will work on fixing tests soon, just wanted to get your thoughts on the options

[goenning]

Hey @clux just wondering if you had a chance to look at this PR. I’m happy to make any API that you think might be better.

[clux]

Hey! Sorry, I've been slowly working my way through the stack of PRs. From a quick look though, one thing does stand out to me though;

Breaking Client::new for this is not great. It's the main interface for creating any custom_client, and it inherently feels awkward to me to force this out in the public unless it's asked for. Same about ConfigExt if we can help it.

Option 2, you describe breaking make_generic_builder, but this is not public? Am I misunderstanding the concern?

AFAIKT, option 3, the inefficient double call might not be too bad? The functions it all call just seem to be re-juggling structs in memory a little bit? Maybe it's a sensible thing to start with. Sounds like maybe the date can be exposed through rustls more effeciently at a later date?

[goenning]

Option 2, you describe breaking make_generic_builder, but this is not public? Am I misunderstanding the concern?

That was a typo, I meant to say it's a breaking change for those NOT using make_generic_builder, the breaking change would be on rustls_https_connector_with_connector and openssl_https_connector_with_connector

AFAIKT, option 3, the inefficient double call might not be too bad? The functions it all call just seem to be re-juggling structs in memory a little bit? Maybe it's a sensible thing to start with. Sounds like maybe the date can be exposed through rustls more effeciently at a later date?

I've created #1707 as an implementation of option 3, let me know what you think!

[clux]

replaced by #1707