The torque project CLI toolchain
The tourque project is an LLVM based clojure compiler. This repository is the main entry point to clone and build the project.
The project started out in 2016 because of some frustration with the slowness of ClojureScript in particular. The goal of this project was to bootstrap a working REPL that compiles to machine code using LLVM, in less time than it takes to compile a simple ClojureScript application. In general this goal was met. The torque compiler currently manages to compile the c++ boot strap code in <8s on my mac book, interprets the clojure.core code, and compiler once, and then starts a repl, which in turn compiles the clojure core code. The boot strapping process until a repl is available currently takes less than 10 seconds.
Besides this purely "for fun" goal, the theory behind the project was that a single compiler that targets different back ends would have worked better than Clojure's current approach, of hosting the language of different platforms. Especially considering the possibility to target web assembly, or GPUs using LLVM, makes using LLVM attractive. Clojure's initial goal to support interaction with the JVM could have easily been supported once compilation to llvm byte code works.
Fun things this might do in the future.
- Compile Clojure to GPU code.
- Compile Clojure to Web Assembly via LLVM.
- Interface with Java via JNI.
If you don't want to build the project yourself by hand, you can just run
docker build . -t torque/demo
docker run --rm -it --entrypoint /bin/zsh torque/demo
in the project root to get a docker container with a built torque environment.
repl
will get you into a repl that compiles to native code. Note that the
repl has to get bootstrapped every time right now, it might take a
couple of seconds, since all of clojure.core has to be compiled on
each startup. From here on, you can use the repl, like you normally would
in Clojure.
user> (+ 1 1)
user> 2
The toolchain currently works on Ubuntu and OS X. Make sure that LLVM 18.0+ is installed from either the llvm debian repository or via homebrew on OS X.
Then clone this reposity with the --recursive flag.
git clone --recursive -b main git@github.com:torque-project/torque.git
This will clone all dependencies of this project. Next, we need to build the libffi version in lib/rev/lib/libffi. This is required since the version in homebrew and the ubuntu repository is outdated. Install libffi to an appropriate location like /usr/local.
Next, we need to set some environment variables for the build, and later running a repl.
. ./activate
On platforms, other than Mac, with homebrew, you need to provide the path to the LLVM installation, if this isn't simply installed in the search path.
export LLVM_PATH=/opt/llvm/18.0.0/
After that simply run make in the project root. Once this is done you should source the activate script found in the project root. Now, simply run the repl script in the bin folder.
bin/repl
This will bootstrap the compiler and start a LLVM compiled REPL. The repl currently doesn't cache the result of the bootstrapping process, so the REPL start time is currently a couple of seconds long on each run.
The code base is split up into several sub projects, providing the core libraries, I/O, and compiler infrastructure.
- https://github.com/torque-project/torque.core
- https://github.com/torque-project/torque.reader
- https://github.com/torque-project/torque.analyzer
- https://github.com/torque-project/torque.compiler
- https://github.com/torque-project/torque.io
I'm currently in the process of updating this project. The original version didn't feature garbage collection, since I wanted to write a sytem that doesn't rely on this. Rusts borrow system provides a nice alternative. There are also other optimizations, and a clean up that are necessary.