Remove reqwest dependency

[KendallWeihe]

No description provided.

[KendallWeihe]

Converting to a draft for the time being for further consideration

[KendallWeihe]

I think we'll probably move forward with this, but @diehuxx wondering if you can give it a look. Windows test seems to be failing so I'm guessing something in our TLS or TCP (or something else?) is causing issues specifically on Windows, so I'll get to the bottom of that.

I may rework the http.rs module to be a slightly different design, but I wanted to spike in the ground a functional solution where all tests are passing (except apparently Windows?).

The major win here is to get us off of OpenSSL entirely. Also it reduces the byte code file size by ~10% which is not insignificant. We're unable to get out of ring showing up in our Cargo.lock file, but @finn-tbd just recently added CI jobs which will fail if it slips into the compiled result.

[diehuxx]

@KendallWeihe The failure is that we expect Web5Error::Network("failed to lookup address information") but instead we're getting Web5Error::Network("No such host is known"). All that differs is the text of the error message returned to us by TcpStream::Connect in http.rs.

I suspect that Windows rust implementation of TcpStream::connect returns one error message and Mac and Linux return another. In both cases, the error message conveys the same thing, i.e. This host isn't valid.

I'd be comfortable just updating the test to accept either error message. I added a commit doing this.

[diehuxx]

feel like we could just call this get and it would be less confusing. get_bytes_as_http_response sounds like i'm parsing the url as an http response

[KendallWeihe]

I agree! and then do the same for the put_ function below too, yeah?

[nitro-neal]

I would always be in favor of using a library if possible, but removing openssl / reqwest moves us forward so onwards!

[KendallWeihe]

On second thought, I want to spend more time further contemplating the implications here. Moving back to a draft for the time being.

[KendallWeihe]

A leading alternative idea I have here: use reqwest inside the web5 crate, but compile it out during binding, and over the FFI expose the ability to inject an http client lib into web5 from the target languages native runtime. So in other words, for cross language bindings, we enable bring-your-own http client lib. This may be the best of both worlds; more overhead, but we compile away openssl/ring/etc so we reduce the byte code file size in the same way, and we're less likely to run into openssl bugs.

[andresuribe87]

A leading alternative idea I have here: use reqwest inside the web5 crate, but compile it out during binding, and over the FFI expose the ability to inject an http client lib into web5 from the target languages native runtime. So in other words, for cross language bindings, we enable bring-your-own http client lib. This may be the best of both worlds; more overhead, but we compile away openssl/ring/etc so we reduce the byte code file size in the same way, and we're less likely to run into openssl bugs.

FWIW, that's similar to the approach that ktor uses to support different platforms. See here, for more details.