I have a question, can I use tor as a middleware? for example, can I build a social network with Tor embedded into the client as the network layer?
This is the most interesting part of this announcement for me:
>> You can test Arti ... as an embeddable library (if you don't mind a little API instability).
My first thought was adding Tor as a transport for TCP DNS resolution for an existing recursive resolver like unbound. Or, a TOR proxy for DoH public recursive resolvers. Either would result in better privacy than directly using a centralized public resolver with DoH. For the former, you would need to send the query through multiple circuits to have confidence that a guard or an exit node wasn't modifying the query/result (with all exits for a particular query in the same country to minimize geo DNS load balancing causing different results-- this level of control would be easier with a library than a separate daemon communicated through via socks. For the latter, using a socks proxy would work, but the library would make for a simpler setup for the user.
Too bad rust doesn't really do dynamic linking. 'libtor' as a distribution maintained library that is automatically kept patched for security vulnerabilities would add piece of mind when running applications that embed Tor.
As far as I remember, you can actually do dynamic linking. But it has its caveats.
Maybe using two different crates, lib-internal and lib-external, where lib-internal compiles to an dylib/so that exposes a C-abi compatible interface. Lib-external it’s just a idiomatic Rust wrapper to that api. It’s a little bit wonky, but I’m pretty sure that it can work.
https://wiki.debian.org/StaticLinking#Rust https://lwn.net/Articles/797616/ https://github.com/rust-lang/rfcs/pull/2603
I note that the PR for Rust symbol mangling got merged, but it looks like it isn't the default yet, they are waiting on external tools supporting it.
Of course you can do dynamic linking without the way that I previously described, but that library will be highly tight to a specific version of the compiler. I think that the biggest problem is dealing with product types for that matter.
Not that wonky, it's really a natural consequence of the fact that Rust's stable ABI is the C ABI. Plus you can then use lib-internal from any language that supports FFI to C, not just from Rust.
(Of course, some things just cannot be supported across a dylib boundary, such as arbitrary monomorphized code. But this limitation applies to all such languages; it's why you have "header-only" libraries in C/C++ for example.)