Imperative modifications of immutable collections

Overview

The pyrsistent-mutable package presents a decorator that will transform a decorated function to use the pyrsistent API.

This means that a set of specific operations are transformed:

Construction of literal sets, dicts and lists are transformed into calls to pset, pvector and pmap.
Assignments are rewritten to handle:
- Assignments to attributes become evolve calls; nesting is handled correctly.
- Augmented assignments are transformed into regular assignments.
Standalone method invocations are transformed into assignments.

from pyrsistent_mutable import pyrmute
from pyrsistent import PRecord, field

class Simple(PRecord):
    attr = field()
    other = field()

@pyrmute
def example_func():
    # Built in referential integrity
    save_vector = my_vector = [0, 1, 2, 3, 4]  # Mapped to a pvector
    del my_vector[3]  # Does *not* change save_vector

    # Evolve nested attributes
    my_precord = Simple(attr=Simple(), other=[])
    my_precord.attr.attr = 5
    my_precord.other.append(20)

    # Transforms literals and comprehensions
    my_maps = [{'filling': key} for key in ('apple', 'banana')]
    my_maps[0]['crust'] = 'flaky'

    return my_vector, save_vector, my_precord, my_maps

This example is tested in tests/test_readme.py.

What's the point?

It's entirely that the imperative form is easier to read, and that pyrsistent's API is tedious for nested collections, at least compared to native Python syntax.

Also, I'm working on a language that uses this technique more extensively, so this was an opportunity to turn a prototype into something more generally useful.

Installing

Installation should just be:

# Install via pip, preferred.
pip3 install pyrsistent-mutable

# Install traditionally.
python3 setup.py install

Usage

Beyond the example shown above, the main things to keep in mind when using this module:

You function still needs to return values.
A "copy" can be made by simple assignment.
Lists, dicts and sets literals and comprehensions are transformed.
Tuples are not transformed, nor are generators.
Method calls are only transformed if they are standalone expressions.
Rewritten operations should fall back to normal behavior for non-pyrsistent values.
The decorated function can't allow nonlocal names.
global may not work.

Troubleshooting

This is really just trying to take a prototype and do something useful with it.

If a function isn't calling something in a useful manner, the culprit is probably my very lame implementations in pyrsistent_mutable.globals.

Don't forget to `return`

This only munges assignments and expression statements.

Read the `source`

The transformed code is written into your function under __source__ which may be helpful in debugging.

Known limits

Most of these are because I've only done very preliminary work to map imperative operations to pyrsistent values.

Assignment of slices uses the evolver framework, which doesn't handle complex slices.
Deletion of slices similarly doesn't work.
Augmented assignment generally requires a pyrsistent value on the rhs.
- This is mitigated now that the module translates literals.
It is not tested on asynchronous functions or generators. It shouldn't care about them, though.
It's all or nothing.
The top level function can't have nonlocal names. Embedded functions can, though.

Debugging

By default, the decorator will write the transformed source to your function as __source__. I just pulled that name out my hat. You can call the decorator with write_source=False to disable this.

Package maintainer notes

pip install twine
python setup.py bdist_wheel
twine upload dist/pyrsistent_mutable-0.0.x-py3-none-any.whl

scooby/pyrsistent-mutable