wetlands 1.0.3

Wetlands is a lightweight library to create conda environment and execute code inside them.

Wetlands

Wetlands is a lightweight Python library for managing Conda environments.

Wetlands can create Conda environments on demand, install dependencies, and execute arbitrary code within them. This makes it easy to build plugin systems or integrate external modules into an application without dependency conflicts, as each environment remains isolated.

For example, if your application needs to use both Stardist and Cellpose, installing them in the same environment may not work due to conflicting dependencies. With Wetlands, you can create a dedicated environment for each library and run them both as needed from your main script.

The name Wetlands comes from the tropical environments where anacondas thrive.

Appose Python is a great alternative to Wetlands. It even provides the ability to run Java environments (see Appose Java) and share memory between the Python world and the Java world. There are other minor differences between the two libraries. For example, Wetlands provides integrated debugging tools to attach VS Code or PyCharm to isolated environments for step-through debugging with breakpoints. See the Debugging guide for more information.

---

Documentation: https://arthursw.github.io/wetlands/latest/

Source Code: https://github.com/arthursw/wetlands/

---

✨ Features

• Automatic Environment Management: Create and configure environments on demand.
• Dependency Isolation: Install dependencies without conflicts.
• Embedded Execution: Run Python functions or scripts inside isolated environments, with both blocking and non-blocking (task-based) APIs.
• Task API: Execute code asynchronously with progress reporting, cancellation, and event-driven callbacks.
• Parallel Execution: Launch multiple worker processes sharing a single Conda environment and distribute work.
• Integrated Debugging: Debug code running in isolated environments using VS Code or PyCharm with breakpoints and step-through execution.
• Pixi & Micromamba: Wetlands uses either a self-contained pixi or micromamba for fast and lightweight Conda environment handling.

📦 Installation

To install Wetlands, simply use pip:

pip install wetlands

🚀 Usage Example

If the user doesn't have pixi or micromamba installed, Wetlands will download and set it up automatically.

from wetlands.environment_manager import EnvironmentManager

# Initialize the environment manager
environment_manager = EnvironmentManager()

# Create and launch an isolated Conda environment named "numpy"
env = environment_manager.create("numpy", {"pip": ["numpy==2.2.4"]})
env.launch()

# Import a module proxy and call functions in the environment
minimal_module = env.import_module("minimal_module.py")
result = minimal_module.sum([1, 2, 3])
print(f"Result: {result}")

# Or use execute() for a direct blocking call
result = env.execute("minimal_module.py", "sum", args=([1, 2, 3],))

# Clean up
env.exit()

with minimal_module.py:

def sum(x):
    import numpy as np
    return int(np.sum(x))

Non-blocking execution with tasks

submit() returns a Task object immediately, letting you monitor progress, cancel, or wait for the result:

# Submit a function for non-blocking execution
task = env.submit("compute.py", "heavy_computation", args=(data,))

# Do other work while the task runs...
print(f"Status: {task.status}")

# Block for the result when ready
task.wait_for()
print(f"Result: {task.result}")

Parallel execution with multiple workers

Launch multiple worker processes sharing the same Conda environment:

env.launch(max_workers=4)

# Distribute work across workers
results = list(env.map("segment.py", "segment", images))

# Or get individual Task objects for full control
tasks = env.map_tasks("segment.py", "segment", images)

Workers that crash or hang are detected and replaced automatically. Set worker_timeout to fail tasks when a worker stops responding:

env.launch(max_workers=4, worker_timeout=300)  # 5-minute inactivity timeout

See the examples/ folder and the documentation for more detailed examples.

🐛 Debugging

Wetlands includes tools to debug code running in isolated environments using VS Code or PyCharm. You can set breakpoints, step through code, and inspect variables in real-time.

Quick Debugging Example

# List all running environments and their debug ports
wetlands list

# Attach VS Code to an environment for debugging
wetlands debug -s /path/to/my/project -n my_env

# Or use PyCharm instead
wetlands debug -s /path/to/my/project -n my_env -ide pycharm

# Kill an environment when done
wetlands kill -n my_env

For detailed debugging instructions and workflows, see the Debugging guide.

🔗 Related Projects

• Conda
• Pixi
• Micromamba

🤖 Development

Use uv to easily manage the project.

Check & Format

Check for code errors with uv run ruff check and format the code with uv run ruff format.

Tests

Test wetlands with uv and ipdb: uv run pytest --pdb --pdbcls=IPython.terminal.debugger:TerminalPdb tests Use --last-failed to only re-run the failures: uv run pytest --pdb --pdbcls=IPython.terminal.debugger:TerminalPdb --last-failed tests

Build and Publish

Build with uv build Publish with uv publish dist/wetlands-VERSION_NAME*

Generate documentation

The Wetlands documentation is generated with mkdocs-material, mkdocstrings, mike and others.

Install the doc tools with uv pip install ".[docs]".

MkDocs includes a live preview server, so you can preview your changes as you write your documentation. The server will automatically rebuild the site upon saving. Start it with: uv run mkdocs serve.

mike is used to generate multiple versions of the docs. To create a new version, mike deploy [version] is used by Github Actions, just update .github/workflows/ci.yml.

The doc is automatically generated by Github Actions (see .github/workflows/ci.yml).

The script scripts/gen_ref_pages.py is used by mkdocs to generate the API reference automatically (see mkdocstrings recipes).

📋 Todo

• Use Pixi features and environment instead of creating one workspace per environment.

📜 License

This project was made by the SAIRPICO team at Inria in Rennes (Centre Inria de l'Université de Rennes) and is licensed under the MIT License.

The logo Wetland was made by Dan Hetteix from Noun Project (CC BY 3.0).