{% raw %}

{% endraw %}

All the clever stuff in this library is provided by Python's builtin ast module and compilation/exec/eval system, along with IPython's CachingCompiler which does some deep magic. tinykernel just brings them together with a little glue.

Install

With pip:

pip install tinykernel

With conda:

conda install -c fastai tinykernel

How to use

This library provides a single class, TinyKernel, which is a tiny persistent kernel for Python code:

{% raw %}

k = TinyKernel()

{% endraw %}

Call it, passing Python code, to have the code executed in a separate Python environment:

{% raw %}

k("a=1")

{% endraw %}

Expressions return the value of the expression:

{% raw %}

k('a')

{% endraw %}

All variables are persisted across calls:

{% raw %}

k("a+=1")
k('a')

{% endraw %}

Multi-line inputs are supported. If the last line is an expression, it is returned:

{% raw %}

k("""import types
b = SimpleNamespace(foo=a)
b""")

namespace(foo=2)

{% endraw %}

The original source code is stored, so inspect.getsource works and, tracebacks have full details.

{% raw %}

k("""def f(): pass # a comment
import inspect
inspect.getsource(f)""")

'def f(): pass # a comment\n'

{% endraw %}

Acknowledgements

Thanks to Christopher Prohm, Matthias Bussonnier, and Aaron Meurer for their helpful insights in this twitter thread.