CXBQN

VM for BQN.

See CONTRIBUTING.md for information on contributing.

Building

Requires a compiler capable of C++20. GCC 9.3.1 is the oldest compiler I have tested it with.

mkdir build
cd build
# Use CXX to choose your compiler
CXX=g++-11 cmake ..
make -j12
./BQN -v
CXBQN 0.11.0 compiled on Dec 25 2021

Or if you want to use nix, CXBQN has a flake:

$ nix shell github:ashermancinelli/cxbqn
$ BQN -v
CXBQN 0.11.0 compiled on Jan  1 1980

Using

$ ./BQN -h
Usage: BQN [options] args

Positional arguments:
args                   	all remaining arguments are passed into the BQN program as •args [default: {}]

Optional arguments:
-h --help              	shows help message and exits [default: false]
-e --execute           	execute a string as BQN code
-v --version           	prints version information and exits [default: false]
-p --execute-and-print 	execute a string as BQN code, pretty print the result
-x --dump-cu           	dump compilation units after compiling [default: false]
-r --repl              	enter REPL [default: false]
-f --file              	execute a string as BQN code
$ ./BQN -e '•Show 5+5'
10
$ ./BQN -p ↕10
⟨ 0 1 2 3 4 5 6 7 8 9 ⟩
$ ./BQN
   foo←5
5
$ echo '•Show 3‿3⥊↕9' > foo.bqn && ./BQN -f foo.bqn
┌─
╵ 0 1 2
  3 4 5
  6 7 8
        ┘

See the values of •listSys to see what system functions are available in your build of CXBQN.

The REPL also has Readline support if you enable that option at configure time. When you run cmake, pass the argument -DCXBQN_READLINE=ON like so:

mkdir build && cd build
cmake .. -DCXBQN_READLINE=ON

If you have a readline library and development headers installed, this will use them and provide a much nicer REPL environment with variable name completion like so:

$ ./BQN -r
   foo←5
5
   foobar←10
10
   # Tab once for attempted completion
   fo<TAB>
   foo

   # Tab twice for all options if there is no single completion available
   foo<TAB><TAB>
foo     foobar
   foo

Configuring Options

Specification Feature Support

Nonstandard Extensions

GPU Execution

To build with CUDA, enable the CMake option CXBQN_CUDA at configure time. Support for CUDA execution is limited, but it exists! The primary function for CUDA execution is •_CUDAFor, which is a 1-modifier that performs the primitive 𝕗 on array 𝕩 and optionally 𝕨. If you give this functionality a try, feel free to open an issue for any bugs you encounter (there will be many). Here's an example:

$ cmake .. -DCXBQN_CUDA=ON
$ make -j 12
$ ./BQN -r
   (↕10) × •_CUDAFor ⌽↕10
⟨ 0 8 14 18 20 20 18 14 8 0 ⟩
   10 + •_CUDAFor ↕15
⟨ 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 ⟩

See the WIP design document for parallel execution here.

Plotting

CXBQN supports bindings to matplotlib via the matplotlib-cpp library. To enable plotting, enable the CMake option CXBQN_PLOT. You must have Python, matplotlib, and numpy installed.

See doc/plot.md for more information.

BQN -e '•cxbqn.plot.Plot ↕10⋄•cxbqn.plot.Show@'

FFI

See the documentation for FFI here.

Libraries and Packaging

There has been discussion on this topic on the Matrix server, but no consensus has been reached. CXBQN supports importing from locations defined in the environment variable BQNPATH, a :-delimited path just like $PATH.

The order in which •Import searches for files is:

1. Current working directory
2. The BQNPATH environment variable (: delimited)
3. <prefix>/share/bqn where <prefix> is the installation prefix of CXBQN
4. /usr/share/bqn

Custom Namespace •cxbqn

Everything under the namespace •cxbqn is specific to the CXBQN implementation of BQN. As of v0.10.4, only the value •cxbqn.config is available under this namespace:

$ ./BQN -p '•cxbqn.config'
⟨ ⟨ "CXBQN_READLINE" 1 ⟩ ⟨ "CXBQN_CUDA" 0 ⟩ ⟨ "CXBQN_FFI" 0 ⟩ ⟨ "CXBQN_PLOT" 0 ⟩ ⟩