An experimental interpreter for the J programming language, built on top of NumPy.

Implements many of J's primitives and tacit programming capabilities, and can be extended to execute via other backends too.

Can be installed via PyPI:

pip install jjinx  # note the double 'j'

Start the interactive shell:

jinx

The shell prompt is four spaces, so commands appear indented. Internally, all multidimensional arrays are NumPy arrays. Verbs, conjunctions and adverbs are a mixture of Python and NumPy methods.

Here are some examples what Jinx can do so far:

• Solve the "trapping rainwater" problem (solution taken from here):

    +/@((>./\ <. >./\.)-]) 0 1 0 2 1 0 1 3 2 1 2 1
6

• Compute the correlation between two arrays of numbers (taken from here). This is a interesting combination of different trains of verbs, adverbs and conjunctions:

    2 1 1 7 9 (+/@:* % *&(+/)&.:*:)&(- +/%#) 6 3 1 5 7
0.721332

• Create identity matrices in inventive ways (see this essay):

    |.@~:\ @ ($&0) 3
1 0 0
0 1 0
0 0 1

    (i.@,~ = >: * i.) 3
1 0 0
0 1 0
0 0 1

    ((={:)\ @ i.) 3
1 0 0
0 1 0
0 0 1

• Solve the Josephus problem (see this essay). Calculate the survivor's number for a circle of people of size N. Note the use of the verb obverse and the rank conjunction:

    (1&|.&.#:)"0 >: i. 5 10    NB. N ranges from 1 to 50 here (arranged as a table)
 1  1  3  1  3  5  7  1  3  5
 7  9 11 13 15  1  3  5  7  9
11 13 15 17 19 21 23 25 27 29
31  1  3  5  7  9 11 13 15 17
19 21 23 25 27 29 31 33 35 37

• Build nested boxes containing heterogeneous data types and print the contents:

    (<<'abc'),(<(<'de',.'fg'),(<<i. 5 2)),(<(<"0 ] % i. 2 2 3))
┌─────┬──────────┬────────────────────────────┐
│┌───┐│┌──┬─────┐│┌────────┬────────┬────────┐│
││abc│││df│┌───┐│││_       │1       │0.5     ││
│└───┘││eg││0 1│││├────────┼────────┼────────┤│
│     ││  ││2 3││││0.333333│0.25    │0.2     ││
│     ││  ││4 5│││└────────┴────────┴────────┘│
│     ││  ││6 7│││                            │
│     ││  ││8 9│││┌────────┬────────┬────────┐│
│     ││  │└───┘│││0.166667│0.142857│0.125   ││
│     │└──┴─────┘│├────────┼────────┼────────┤│
│     │          ││0.111111│0.1     │0.090909││
│     │          │└────────┴────────┴────────┘│
└─────┴──────────┴────────────────────────────┘

Jinx is written in reasonably-readable Python, so should be easy to navigate. Adding new primitives is easy.

Update the primitives.py file with your new part of speech (e.g. a new verb such as +::). Write your implementation of this new part of speech in the relevant executor module (e.g. verbs.py) and then update the name-to-method mapping at the foot of that module. That's all that's needed.

Execution of sentences is backed by NumPy by default.

However, Jinx is designed so that it is possible to execute sentences using alternative frameworks too. Python has many Machine Learning and Scientific Programming libraries that could be used to execute J code, albeit with different sets of tradeoffs.

To prove this concept, Jinx currently has highly experimental and incomplete support for JAX:

jinx --executor jax

Primitive verbs are JIT compiled and nouns are backed by JAX arrays:

    mean =: +/ % #
    mean 33 55 77 100 101
73.2

This project is experimental and began as a project to learn J in more depth. There will be bugs, missing features and performance quirks.

You will notice that many key parts of J are not currently implemented (but might be in future). These include:

• Differences in how names are interpreted and resolved at execution time.
• Locales.
• Definitions and direct definitions (using {{ ... }}).
• Array types other than floats, integers and and strings.
• Executing J scripts.
• Control words.