7 ms·

I have long had the ambition to be the N:th guy to try Clojure on C++. For similar reasons. I have a pretty good idea how to do it, already got HAMT and some ma

I have long had the ambition to be the N:th guy to try Clojure on C++. For similar reasons. I have a pretty good idea how to do it, already got HAMT and some macroexpanded code to work. But then you remember how even ClojureCLR struggled at times, and that you'd be lightyears behind that, not to speak of CLJ/CLJS, and alone. Maybe one day? :)

jb1991 5y ago
It would seem to me the most significant challenge would be having functional data structures in C++, that would be a non-trivial effort I would think. Even if you were to use other libraries that experiment with this. Without those, the language just doesn’t make any sense.
- toymachine1974 5y ago
  My Park language does this https://github.com/toymachine/park-lang https://github.com/toymachine/park-lang Currently the syntax is mostly JS but the semantics is all clojure. I recreated the map and vector datatypes in c++ by adapting clojures Java code. it needs a GC though which is another big part of the implemention of Park
- vnorilo 5y ago
  Yeah, that was the problem with most attempts I saw, unordered_map won't work. I did write a persistent hash map and vector [1], but ended up mostly using them in C++ projects. I wouldn't advise anyone else to use them at this level of testing/maturity though. Writing them was super interesting. 1: https://hg.sr.ht/~vnorilo/pcoll/browse https://hg.sr.ht/~vnorilo/pcoll/browse (not production quality)
- bertmuthalaly 5y ago
  Immer [0][1] seems like a really serious attempt at bringing high-quality persistent data structures to C++. [0]: https://www.youtube.com/watch?v=sPhpelUfu8Q https://www.youtube.com/watch?v=sPhpelUfu8Q [1]: https://sinusoid.es/immer/ https://sinusoid.es/immer/
geokon 5y ago
I think recreating Clojure point by point is not optimal. There is some middle ground between full on Clojure and the straightjacket of the STL datastructures. - Clojure is kinda difficult to reason about performance wise. The lazyness.. some weird corner cases (like how `(first ..)` is slower than `(nth .. 0)` - The STL is very strict on the "zero cost abstraction" front. Maybe something like Immer but with some syntactic sugar to make it more light weight like Clojure? With hooking into GDB you could probably make a REPL as well.. GDB has live code reload, introspection - you even get state of crash and sane stack traces :)
- vnorilo 5y ago
  You see, I'd like to write more of my software in Clojure (or any functionally inclined Lisp). If there was a really good interop story with a non-gc:d native lang I could do more of it. Clojure is a hosted language, so hosting it on C++ (or Rust etc.) would be sweet for me personally. Also, I like compiler projects.
  - yawaramin 5y ago
    Check out https://github.com/carp-lang/Carp https://github.com/carp-lang/Carp From the readme: > The key features of Carp are the following: > - Automatic and deterministic memory management (no garbage collector or VM) > - Inferred static types for great speed and reliability > - Ownership tracking enables a functional programming style while still using mutation of cache-friendly data structures under the hood > - No hidden performance penalties – allocation and copying are explicit > - Straightforward integration with existing C code > - Lisp macros, compile time scripting and a helpful REPL
  - geokon 5y ago
    I've never had to use it myself, but what part of the JNI/JNA is problematic? One thing I'd sorta fantasized about is (in a very hand wavey way) using Clojure + ClojureCL to run OpenCL kernels. In theory if you were to bundled your JAR with POCL.. you could selectively run your tight-loops/kernels on the CPU or GPU. So it'd be effectively like launching little C snippets. Furthermore, from what I remember, kernels are compiled and executed at run time - so it'd really fit with the whole REPL workflow and they would effectively simply be inline strings of C code that get executed I think the main catch for the moment is that ClojureCL uses a MKL matrices to talk to the OpenCL driver (driver? dispatcher? I might be mixing up my terms) - but that could be modified to a JVM container Probably my own bias - but the only time I've really been let down by Clojure and missed native was when I need to optimize a tight loop. Once you optimized Clojure past a certain point (to eliminate reflection, boxing issues, destructuring etc.) it starts to get ugly.
pflanze 5y ago
I've recently started implementing a VM for Scheme [1] and was already envisioning to maybe add Clojure at some point. I'm writing it in plain C, since I don't see much reason to use C++ for the VM itself, and I like the simplicity and transparency of C. Also, I'm hoping to prove security at some point and C may make that easier (support from tooling like proof assistants). I don't think C++ would help enough to avoid security pitfalls by itself, unless one accepts the result to be very slow. But the VM does compile as C++, and complex data structures like HAMT may really better be written in C++, so I don't think I mind that. I think the approach to take for all speed-sensitive objects would be for them to be integrated directly with the GC (allocated from the GC heap, and traced via an object specific method). Let me know if you'd like to discuss this more. [1] https://github.com/pflanze/lilyvm https://github.com/pflanze/lilyvm

jb1991 5y ago

It would seem to me the most significant challenge would be having functional data structures in C++, that would be a non-trivial effort I would think. Even if you were to use other libraries that experiment with this. Without those, the language just doesn’t make any sense.

toymachine1974 5y ago
My Park language does this https://github.com/toymachine/park-lang https://github.com/toymachine/park-lang Currently the syntax is mostly JS but the semantics is all clojure. I recreated the map and vector datatypes in c++ by adapting clojures Java code. it needs a GC though which is another big part of the implemention of Park
vnorilo 5y ago
Yeah, that was the problem with most attempts I saw, unordered_map won't work. I did write a persistent hash map and vector [1], but ended up mostly using them in C++ projects. I wouldn't advise anyone else to use them at this level of testing/maturity though. Writing them was super interesting. 1: https://hg.sr.ht/~vnorilo/pcoll/browse https://hg.sr.ht/~vnorilo/pcoll/browse (not production quality)
bertmuthalaly 5y ago
Immer [0][1] seems like a really serious attempt at bringing high-quality persistent data structures to C++. [0]: https://www.youtube.com/watch?v=sPhpelUfu8Q https://www.youtube.com/watch?v=sPhpelUfu8Q [1]: https://sinusoid.es/immer/ https://sinusoid.es/immer/

geokon 5y ago

I think recreating Clojure point by point is not optimal. There is some middle ground between full on Clojure and the straightjacket of the STL datastructures. - Clojure is kinda difficult to reason about performance wise. The lazyness.. some weird corner cases (like how `(first ..)` is slower than `(nth .. 0)` - The STL is very strict on the "zero cost abstraction" front. Maybe something like Immer but with some syntactic sugar to make it more light weight like Clojure? With hooking into GDB you could probably make a REPL as well.. GDB has live code reload, introspection - you even get state of crash and sane stack traces :)

vnorilo 5y ago
You see, I'd like to write more of my software in Clojure (or any functionally inclined Lisp). If there was a really good interop story with a non-gc:d native lang I could do more of it. Clojure is a hosted language, so hosting it on C++ (or Rust etc.) would be sweet for me personally. Also, I like compiler projects.
- yawaramin 5y ago
  Check out https://github.com/carp-lang/Carp https://github.com/carp-lang/Carp From the readme: > The key features of Carp are the following: > - Automatic and deterministic memory management (no garbage collector or VM) > - Inferred static types for great speed and reliability > - Ownership tracking enables a functional programming style while still using mutation of cache-friendly data structures under the hood > - No hidden performance penalties – allocation and copying are explicit > - Straightforward integration with existing C code > - Lisp macros, compile time scripting and a helpful REPL
- geokon 5y ago
  I've never had to use it myself, but what part of the JNI/JNA is problematic? One thing I'd sorta fantasized about is (in a very hand wavey way) using Clojure + ClojureCL to run OpenCL kernels. In theory if you were to bundled your JAR with POCL.. you could selectively run your tight-loops/kernels on the CPU or GPU. So it'd be effectively like launching little C snippets. Furthermore, from what I remember, kernels are compiled and executed at run time - so it'd really fit with the whole REPL workflow and they would effectively simply be inline strings of C code that get executed I think the main catch for the moment is that ClojureCL uses a MKL matrices to talk to the OpenCL driver (driver? dispatcher? I might be mixing up my terms) - but that could be modified to a JVM container Probably my own bias - but the only time I've really been let down by Clojure and missed native was when I need to optimize a tight loop. Once you optimized Clojure past a certain point (to eliminate reflection, boxing issues, destructuring etc.) it starts to get ugly.

pflanze 5y ago

I've recently started implementing a VM for Scheme [1] and was already envisioning to maybe add Clojure at some point. I'm writing it in plain C, since I don't see much reason to use C++ for the VM itself, and I like the simplicity and transparency of C. Also, I'm hoping to prove security at some point and C may make that easier (support from tooling like proof assistants). I don't think C++ would help enough to avoid security pitfalls by itself, unless one accepts the result to be very slow. But the VM does compile as C++, and complex data structures like HAMT may really better be written in C++, so I don't think I mind that. I think the approach to take for all speed-sensitive objects would be for them to be integrated directly with the GC (allocated from the GC heap, and traced via an object specific method). Let me know if you'd like to discuss this more. [1] https://github.com/pflanze/lilyvm https://github.com/pflanze/lilyvm