pydistill 0.2.0

Extract Python models and dependencies into standalone packages

pydistill

Extract Python models and their transitive dependencies into standalone, self-contained packages.

Why?

You have a large Python project and want to share some Pydantic models (or any Python classes) with another service without:

• Publishing your entire codebase
• Manually copying files and fixing imports
• Maintaining a separate package by hand

PyDistill automates this: point it at your entry points, and it extracts exactly what's needed into an importable package.

Installation

# With uv
uv add pydistill

# With pip
pip install pydistill

Quick Start

# Extract a model and its dependencies
pydistill \
    --entry myapp.models:User \
    --entry myapp.models:Order \
    --base-package myapp \
    --output-package extracted_models \
    --output-dir ./dist/extracted_models

This will:

1. Discover all local imports transitively from User and Order
2. Copy the relevant source files to ./dist/extracted_models/
3. Rewrite imports (myapp.* → extracted_models.*)
4. Generate __init__.py files so the package is importable
5. Generate pyproject.toml so the output is installable with pip

Install the extracted output directly:

pip install ./dist/extracted_models

Configuration

CLI Options

pydistill [OPTIONS]

Options:
  -e, --entry MODULE:NAME     Entry point (can be repeated)
  -b, --base-package PACKAGE  Base package to extract from
  -p, --output-package PACKAGE Output package name
  -o, --output-dir DIR        Output directory
  -s, --source-root DIR       Additional source roots (can be repeated)
  -c, --config FILE           Path to config file
  -n, --dry-run               Show what would be extracted
  --clean                     Remove output directory first
  -q, --quiet                 Suppress output
  -f, --filesystem-only       Skip importlib, use only filesystem resolution
  --dist-name NAME            Distribution name in generated pyproject.toml
  --dist-version VERSION      Distribution version in generated pyproject.toml
  --dependency SPEC           Dependency specifier (can be repeated)
  --format                    Format extracted files (default: ruff format)
  --formatter CMD             Custom formatter command (implies --format)
  --version                   Show version
  -h, --help                  Show help

Configuration File

Create a pydistill.toml in your project root:

[pydistill]
entries = [
    "myapp.users.models:User",
    "myapp.orders.models:Order",
]
base_package = "myapp"
output_package = "extracted_models"
output_dir = "./dist/extracted_models"
clean = true
dist_name = "extracted-models"
dist_version = "0.1.0"
dependencies = ["pydantic>=2.0", "email-validator>=2.0"]
# filesystem_only = true  # Enable for uninstallable projects
# format = true           # Format extracted files
# formatter = "ruff format"  # Custom formatter command

Then just run:

pydistill

PyDistill automatically searches for pydistill.toml in the current directory and parent directories.

CLI arguments override config file values.

Example

Given this project structure:

myapp/
├── __init__.py
├── common/
│   ├── __init__.py
│   └── types.py          # Status enum, Address model
├── users/
│   ├── __init__.py
│   └── models.py         # User model (imports common.types)
└── orders/
    ├── __init__.py
    └── models.py         # Order model (imports users.models, common.types)

Running:

pydistill -e myapp.orders.models:Order -b myapp -p extracted -o ./dist/extracted

Produces:

dist/extracted/
├── pyproject.toml
├── __init__.py
├── common/
│   ├── __init__.py
│   └── types.py          # Status, Address (imports rewritten)
├── users/
│   ├── __init__.py
│   └── models.py         # User (imports rewritten)
└── orders/
    ├── __init__.py
    └── models.py         # Order (imports rewritten)

All imports like from myapp.common.types import Status become from extracted.common.types import Status.

Extracting from Uninstallable Projects

Need to extract models from a project that can't be installed in your current environment (e.g., has Windows-only dependencies on macOS)? Use --filesystem-only:

pydistill \
    -e their_app.models:SomeModel \
    -b their_app \
    -p extracted \
    -o ./dist/extracted \
    -s /path/to/their/project \
    --filesystem-only

This skips Python's importlib and resolves modules purely via filesystem search in the specified source roots.

Formatting Extracted Code

The AST-based import rewriting can produce code that's not perfectly formatted. Use --format to automatically format extracted files:

# Use ruff (default)
pydistill -e myapp:Model -b myapp -p out -o ./dist --format

# Use black instead
pydistill ... --format --formatter black

# Custom ruff config
pydistill ... --format --formatter "ruff format --line-length 120"

Formatting failures are non-fatal—extraction will succeed even if the formatter is unavailable.

How It Works

1. Parse entry points - Resolve module:ClassName to source files
2. Discover imports - Use Python's ast module to find all import and from ... import statements
3. BFS traversal - Follow imports transitively within the base package (ignores third-party like pydantic, datetime, etc.)
4. Rewrite imports - Use ast.NodeTransformer to rewrite module paths
5. Generate package - Copy files preserving structure, create __init__.py files, and write pyproject.toml

Use Cases

• Microservices: Share domain models between services without a monorepo
• API clients: Generate a lightweight client package with just the request/response models
• CI/CD: Automatically generate model packages when source changes
• Cross-platform extraction: Extract from projects with platform-specific dependencies using --filesystem-only

Limitations

• Only follows imports within --base-package (by design)
• Extracts entire modules, not individual classes (a module containing User and Admin will include both)
• Relative imports are preserved as-is (they work in the output package since structure is maintained)
• Comments are not preserved: The AST-based rewriting uses ast.parse() and ast.unparse(), which strips comments from source files. This is acceptable for artifact/tarball use cases but means the extracted code loses inline documentation. Use --format to ensure consistent formatting of the output.
• Dynamic imports are not traced: Imports using importlib.import_module() or __import__() cannot be statically analyzed
• Star imports (from x import *): The module path is rewritten correctly, but transitive dependencies imported via __all__ re-exports may not be discovered

Development

# Clone and install
git clone https://github.com/yourorg/pydistill
cd pydistill
uv sync

# Run tests
uv run pytest

# Run pydistill locally
uv run pydistill --help

License

MIT