Programs as Objects


License
MIT
Install
pip install programlib==12.0.1

Documentation

Programlib: programs as objects

Programlib is a tool that turns programs in any programming language into convenient Python objects, letting you run any string as a C++/Python/Clojure/etc program from a Python script. This project is aimed to help develop automatic programming and genetic software improvement systems, though many other applications are possible.

Installation

Programlib can be installed with

pip install programlib

However, you have to also make sure that the programming languages you want to use are installed. By default, programlib uses command line tools that come with the programming languages, i.e. python3 or javac.

Standard usage

Create a program object with

from programlib import Program
program = Program(source_code, language='C++')

This object has

  • a save method that will save the source code to a file at the specified path.
  • a run method that runs the program and returns a list of strings it printed to stdout. You can optionally provide a list of input strings as well.
  • a spawn method that launches the program in an interactive mode. It returns an Agent object with and act method that takes a sequence of input strings and returns a list of strings printed to stdout in response.
  • a test method that takes a list of test cases. A test case is a tuple of 2 lists: the first list is the input strings, the second is the expected output strings. The method returns a full log of all test runs and updates program.avg_score and program.test_pass_rate attributes.

To use the program to process some input lines, do

output_lines = program.run(input_lines)

To use the program as an agent in a Reinforcement Learning environment, do

env = gym.make('Env-vX')
agent = program.spawn().rl(env.action_space, env.obs_space)
obs, info = env.reset()
terminated = False
truncated = False
while not (terminated or truncated):
    action = agent.predict(obs, deterministic=True)
    obs, reward, terminated, truncated, info = env.step(action)

See also examples.

Currently supported programming languages out of the box are C++, Python, Java, Clojure, Ruby, Rust, Go, Haskell, Scala, Kotlin, PHP, C#, Swift, D, Julia, Clojure, Elixir and Erlang. See "Advanced usage" below for instructions on how to add other languages.

Advanced usage

Language configuration

When you create a program object with a language name like language='C++', programlib retrieves an appropriate language configuration from it's database. If you have a different opinion on how to compile or run in this language or want to use a language that is not supported out of the box, you can create your own language configuration object:

from programlib import Program, Language
language = Language(
        build_cmd='g++ {name}.cpp -o {name}',
        run_cmd='./{name}',
        source='{name}.cpp',
        artefacts=['{name}']
    )
program = Program(source_code, language=language)

source parameter describes the naming convention for the source file (usually {name}.extension). Make sure that this parameter contains a {name} placeholder, so that programlib can keep track of several source files at the same time. build_cmd and run_cmd respectively instruct programlib which commands to use to compile and run the program in this language. artefacts is a list of all the files produced by build_cmd command. It is needed to clean up the artefacts when the program object is destroyed.

Error handling

By default, any errors at build time or run time will lead to RuntimeError being raised, with 2 exceptions:

  • test function that catches exceptions during test cases execution and marks these tests as failed.
  • Setting program.run(force=True) or program.test(force=True) will make programlib ignore all errors.

You can check program.stdout and program.exitstatus to check the output during the last run (or, if in was never run, during build).