python-obfuscator

python-obfuscator

A Python source-code obfuscator built on the standard-library ast module. It applies multiple independent techniques — each individually togglable — to make code harder to read while keeping it fully executable. See Known limitations before use.

If this project is useful to you, consider sponsoring development.

Installing

pip install python-obfuscator

Requires Python ≥ 3.10.

From source (contributors)

pip install -e ".[dev]"

Or with Poetry:

poetry install

(poetry install pulls dev dependencies from the lockfile; use poetry run <command> to run tools inside that environment.)

Quick start — CLI

# Writes obfuscated/your_file.py (path structure is preserved)
pyobfuscate -i your_file.py

# Print to stdout
pyobfuscate -i your_file.py --stdout

# Disable specific techniques
pyobfuscate -i your_file.py --disable dead_code_injector --disable exec_wrapper

# Show version
pyobfuscate --version

Python API

One-shot helper

from python_obfuscator import obfuscate

source = "x = 1\nprint(x + 2)"
result = obfuscate(source)
print(result)

Selective techniques

from python_obfuscator import obfuscate, ObfuscationConfig

# All techniques except dead-code injection
config = ObfuscationConfig.all_enabled().without("dead_code_injector")
result = obfuscate(source, config=config)

# Only string encoding
config = ObfuscationConfig.only("string_hex_encoder")
result = obfuscate(source, config=config)

Reusing across multiple files (caches the pipeline)

from python_obfuscator import Obfuscator, ObfuscationConfig

obf = Obfuscator(ObfuscationConfig.all_enabled())
for path in my_files:
    path.write_text(obf.obfuscate(path.read_text()))

Config combinators

cfg = ObfuscationConfig.all_enabled()   # every registered technique
cfg = ObfuscationConfig.only("variable_renamer", "exec_wrapper")
cfg = cfg.without("exec_wrapper")       # returns a new frozen config
cfg = cfg.with_added("dead_code_injector")

Techniques

Name	Priority	What it does
variable_renamer	10	Renames local variables, function names, parameters, and class names to visually ambiguous identifiers (lIIllIlI…). Excludes builtins, imports, dunders, and attribute-accessed names.
string_hex_encoder	20	Replaces every string literal "hi" with bytes.fromhex('6869').decode('utf-8'). Skips f-strings.
dead_code_injector	30	Injects dead variable assignments at every scope level — module body, function bodies, class bodies, if/for/while/try/with branches. Some assignments reference other dead variables to simulate computation.
exec_wrapper	100	Wraps the entire module in a single exec("…") call, reducing the top-level AST to one statement. Runs last.

Techniques are applied in priority order (lowest first).

Example

Input

def greet(name):
    msg = "Hello, " + name
    print(msg)

greet("world")

Output (all techniques enabled — abridged)

exec('def lIlIllI(IIlIlII):\n    lIllIlI = bytes.fromhex(\'48656c6c6f2c20\').decode(\'utf-8\') + IIlIlII\n    ...\nlIlIllI(bytes.fromhex(\'776f726c64\').decode(\'utf-8\'))')

Performance overhead

Benchmarks run on an Apple M-series machine, 20 iterations each. The test programs cover OOP, algorithms, functional patterns, number theory, and string processing.

Total overhead (all techniques)

Program	Original	Obfuscated	Overhead
algorithms.py	0.94 ms	1.98 ms	+112%
cipher.py	1.37 ms	2.67 ms	+95%
data_structures.py	0.72 ms	2.20 ms	+207%
functional.py	0.67 ms	1.71 ms	+155%
number_theory.py	1.84 ms	3.16 ms	+72%
oop_simulation.py	0.68 ms	1.66 ms	+144%

Per-technique contribution (average across all programs)

Technique	Avg overhead	Notes
variable_renamer	~5%	Pure rename — negligible at runtime
string_hex_encoder	~12%	bytes.fromhex call per string literal
dead_code_injector	~85%	Dominant cost — dead assignments execute every iteration
exec_wrapper	~2%	Single extra exec layer

The dead-code injector's overhead scales with the number of scopes and loop iterations in the original program. Programs with tight inner loops see the most overhead.

Known limitations

• Class method names are not renamed. Attribute-accessed names (obj.method) cannot be safely renamed without full type-inference, so the renamer conservatively excludes them.
• Keyword argument names are not renamed. fn(key=val) call-site keyword strings are bare AST strings, not Name nodes, and are not updated when a parameter is renamed.
• No scope-aware renaming. The same identifier used in two independent function scopes maps to the same obfuscated name (which is semantically correct but less obfuscated than it could be).
• No control-flow obfuscation. Opaque predicates, bogus branches, and integer encoding are not implemented.

Running the test suite

After a dev install (from source):

pytest

Coverage is enforced at ≥ 95% on every CI run.

# With coverage report
coverage run -m pytest && coverage report

# E2E tests with benchmark output
pytest tests/e2e/ -v -s

With Poetry, run the same commands through the project environment, for example poetry run pytest, poetry run coverage run -m pytest, and poetry run pytest tests/e2e/ -v -s.

Authors

David Teather — davidteather

License

MIT — see LICENSE.