Joe Birr-Pixton

> Are you thinking the demultiplexing shouldn't be done in Rustls and we should continue to expect it to be done donwstream ala the approach used in [rustls/tokio-rustls#54 (comment)](https://github.com/rustls/tokio-rustls/issues/54#issuecomment-2027496525) ?...

wrong butan

> * Ureq depends on both `rustls` and `rustls-webpki`, which means if we do nothing, we get both `aws-lc-rs` and `ring` (which goes against the idea of having minimal amount...

Memory allocations that only happen in error branches seems like a recipe for memory leaks (though OpenSSL does that, it hangs on to the allocations in a global and lets...

I don't think this is necessary? There's no reason that a crate which takes a dependency must re-export or otherwise control the dependency's features. If you want to set `pki-types/web`...

> I agree it's not strictly necessary, but it usually makes things easier for the consumer at little cost to the maintainer? I think a basic level of testing is...

That's very weird. %zu is the c99-standard format for a size_t, and has been supported by GCC since version ~3.

I would suggest using ring instead of this crate; it should have pretty competitive performance to this one and is easier to build.

> It would be nice to add the GOST cipher suite But... why? It's not widely deployed, and not really meaningfully better than the widely deployed alternatives (though admittedly I...

rustls can be asked how much data it would write into an infinite size buffer provided to `reader().read()` -- `self.session.process_new_packets().unwrap().plaintext_bytes_to_read`[^1] -- and I think it would be reasonable to expose...