I know that exists, but whats the point of that? You loose all advantages of rust when you use the library then because it cant predict application state with the library code. There is a reason all those rust libraries are compiled locally when you compile a rust application. Its a major lacking point for rust, and as long as it lacks that its dead in the water for big projects.
Again, i like strong type stuff and i like the ideas of rust but it is not grown up enough for me
It’s a lacking point yes but unless you want to use a closed-source library it’s also a non-issue, which is why it has never been given priority. It’s not like language semantics would prevent portable dylibs it’s that there’s more important fronts to improve Rust on. A proper solution would take quite some engineering effort, and do note that C doesn’t have a proper solution either it just lets you link stuff up willy-nilly and then crash. Rust is actually in a better position to implement a proper solution than C is.
The “big project” thing is a red herring given that rust compiles incrementally. I know it is technically possible to not rebuild everything from scratch in C but the code has to specifically written to not break assumptions your build system makes while rust is happily re-using the compilation results for one function in a file while discarding those of another because actual dependencies are actually tracked. Out of the box.
Speaking of large Rust projects and proper type-safe linking: The WebAssembly folks are hashing out their Component Model which isn’t really limited to compiling to wasm, in principle: Big picture it’s a way to programmatically specify ABIs and even derive ABI translation code. That might be a good option as a rust-specific solution would be, well, rust-specific and when you engineer something that can support multiple versions of a language you can just as well engineer a bit more and have something cross-language.
The Rustinomicon has a chapter on it. The basics are quite simple: Declare non-opaque types to use layout matching the C ABI, export/import functions, some wibbles around name mangling. Option<T> vs. null pointers. Where things get a bit more involved is unwinding, but then you’re at the end of it, nothing should be shocking to anyone having written C.
As to how Rusty it is… not very. I mean Rust has first-class FFI support, but the way FFI stuff is written is necessarily unidiomatic because you’re basically writing C in Rust syntax and you won’t get out of declaring your own functions `unsafe’ before you read the rest of the Rustinomicon to understand what properties you need to ensure because the nice and shiny parts of Rust assume them.
Hmm. So I guess it comes down to what OP is doing. They either want to write a Rust library, or something that uses a Rust library that may not be standardised or even exist yet. If the latter, they should stick with C.
You’re aware that Rust gives you access to the full C ABI?
What language are you going to use instead that has a better ABI story? Swift? Or maybe a dynamic language like Python?
I know that exists, but whats the point of that? You loose all advantages of rust when you use the library then because it cant predict application state with the library code. There is a reason all those rust libraries are compiled locally when you compile a rust application. Its a major lacking point for rust, and as long as it lacks that its dead in the water for big projects.
Again, i like strong type stuff and i like the ideas of rust but it is not grown up enough for me
It’s a lacking point yes but unless you want to use a closed-source library it’s also a non-issue, which is why it has never been given priority. It’s not like language semantics would prevent portable dylibs it’s that there’s more important fronts to improve Rust on. A proper solution would take quite some engineering effort, and do note that C doesn’t have a proper solution either it just lets you link stuff up willy-nilly and then crash. Rust is actually in a better position to implement a proper solution than C is.
The “big project” thing is a red herring given that rust compiles incrementally. I know it is technically possible to not rebuild everything from scratch in C but the code has to specifically written to not break assumptions your build system makes while rust is happily re-using the compilation results for one function in a file while discarding those of another because actual dependencies are actually tracked. Out of the box.
Speaking of large Rust projects and proper type-safe linking: The WebAssembly folks are hashing out their Component Model which isn’t really limited to compiling to wasm, in principle: Big picture it’s a way to programmatically specify ABIs and even derive ABI translation code. That might be a good option as a rust-specific solution would be, well, rust-specific and when you engineer something that can support multiple versions of a language you can just as well engineer a bit more and have something cross-language.
Are you suggesting OP write a C application and then compile it as Rust? I’m not a pro, but that sounds kind of janky.
I’m suggesting building a Rust library and exposing a C ABI. That’s what rsvg does for example.
Oh. There’s a still Rust-y way to do this? Nevermind.
OP wanted stability and predictability. I suppose we’ll see how entrenched one library can become.
The Rustinomicon has a chapter on it. The basics are quite simple: Declare non-opaque types to use layout matching the C ABI, export/import functions, some wibbles around name mangling.
Option<T>vs. null pointers. Where things get a bit more involved is unwinding, but then you’re at the end of it, nothing should be shocking to anyone having written C.As to how Rusty it is… not very. I mean Rust has first-class FFI support, but the way FFI stuff is written is necessarily unidiomatic because you’re basically writing C in Rust syntax and you won’t get out of declaring your own functions `unsafe’ before you read the rest of the Rustinomicon to understand what properties you need to ensure because the nice and shiny parts of Rust assume them.
Hmm. So I guess it comes down to what OP is doing. They either want to write a Rust library, or something that uses a Rust library that may not be standardised or even exist yet. If the latter, they should stick with C.
Writing C bindings to a Rust library is the easier scenario because you can rely on the safe code having nice and clean memory semantics.
Yeah, Rust has pretty good integration of it: https://doc.rust-lang.org/nomicon/ffi.html#calling-rust-code-from-c
You do lose some of the Rust-y-ness, because obviously the C ABI is much more simplistic, but in terms of a stable ABI, it’s impossible to beat C.