Available optimizers

Available optimizers#

Early-stage optimizers (order 100)#

These run first and delete entire files or directories, avoiding unnecessary processing by later optimizers.

remove_tests#

Removes Python test files that are confirmed to contain test framework usage. Requires both a test naming convention (test_*.py, *_test.py, conftest.py, or files inside tests//test/ directories) and test framework signals (import pytest, import unittest, def test_, class Test) before deleting.

  • Only removes .py files

  • Keeps __init__.py, helper modules, fixture data, and non-Python files

  • conftest.py only removed if it imports pytest

remove_typestubs#

Removes .pyi type stub files and py.typed PEP 561 markers. Type stubs are only used by type checkers and IDEs, never at runtime.

remove_pycache#

Removes files inside __pycache__/ directories that sometimes get included in wheels by buggy build systems.

Normal-stage optimizers (order 500)#

These transform .py source files in place.

remove_docstrings#

Strips docstrings from module, class, function, and async function bodies using AST parsing and token manipulation. Preserves all runtime behavior since docstrings are only used for documentation.

remove_type_annotations#

Strips type annotations using an AST NodeTransformer and ast.unparse(). Handles all annotation locations:

  • Function parameters (positional-only, regular, keyword-only, *args, **kwargs)

  • Return type annotations

  • Annotated assignments (x: int = 5 → x = 5; bare x: int → pass)

Preserves annotations in:

  • @dataclass classes

  • NamedTuple subclasses

  • TypedDict subclasses

  • Protocol subclasses

These classes depend on annotations for their runtime behavior.

remove_assertions#

Replaces assert statements with pass using an AST NodeTransformer. This is similar to running Python with the -O flag but applied at the source level.

remove_comments#

Strips # comments from .py files using the tokenize module. Correctly preserves strings that contain # characters and docstrings.

minify_whitespace#

Reduces indentation from 4 spaces (or whatever the file uses) to 1 space per level and removes blank lines. Uses tokenize to detect the indent unit, then rewrites line-by-line. String contents are preserved. This is analogous to whitespace minification in JavaScript bundlers.

Late-stage optimizers (order 900)#

These run last, after all source transforms are complete.

compile_pyc#

Compiles .py files to .pyc bytecode using py_compile and removes the original .py files. The .pyc file is placed next to the original (not in __pycache__/). Files with syntax errors are skipped.