如何实现类似NumPy的Python包导入方式——直接导入包文件夹名即可访问内部函数

Great question! The key to making your package behave like NumPy (where import mypackage gives direct access to your functions) lies in leveraging the __init__.py file in your package directory. This file acts as the "entry point" for your package and defines what gets exposed to users when they import it.

Here's a step-by-step guide to achieve this:

1. Keep Your Folder Structure #

First, maintain your existing package structure—it's already on the right track:

mypackage/
    __init__.py
    package.py

2. Define Your Functions in package.py #

Let's say package.py has some example functions:

# package.py
def calculate(x):
    return x * 2

def greet(name):
    return f"Hello, {name}!"

3. Modify __init__.py to Expose Functions #

The __init__.py file controls what's accessible from the top-level mypackage namespace. You have a few options here, depending on how you want users to interact with your package:

Option A: Explicitly Import Specific Functions (Recommended) #

This is the cleanest approach, as it makes it clear exactly what's available at the top level. Add this to __init__.py:

# __init__.py
from .package import calculate, greet

Now, when users run import mypackage, they can call your functions directly:

import mypackage

print(mypackage.calculate(10))  # Output: 20
print(mypackage.greet("Alice")) # Output: Hello, Alice!

Option B: Import All Functions (With Guardrails) #

If you want to expose all functions from package.py without listing them individually, use from .package import *—but pair it with a __all__ list in package.py to avoid exporting unintended names.

First, add __all__ to package.py:

# package.py
__all__ = ["calculate", "greet"]  # List exactly what should be exported

def calculate(x):
    return x * 2

def greet(name):
    return f"Hello, {name}!"

def _internal_helper():  # This won't be exported because it's not in __all__
    pass

Then update __init__.py:

# __init__.py
from .package import *

This achieves the same result as Option A, but keeps your __init__.py concise if you have many functions.

Option C: Expose the Submodule (Like NumPy's Submodules) #

If you want users to also access mypackage.package.calculate while still having top-level access, import the submodule into __init__.py:

# __init__.py
from . import package
from .package import calculate, greet

Now users can choose either approach:

import mypackage

mypackage.calculate(10)          # Works
mypackage.package.calculate(10)  # Also works

4. Bonus: Install Your Package System-Wide #

To make your package accessible anywhere on your system (just like NumPy), create a pyproject.toml file in the parent directory of mypackage:

# pyproject.toml
[build-system]
requires = ["setuptools>=61.0"]
build-backend = "setuptools.build_meta"

[project]
name = "mypackage"
version = "0.1.0"

Then install it using pip:

pip install .

Now you can import mypackage from any script on your machine.

Note: If you wanted import package to work directly, that would require package.py to be in your script's directory or Python path—but that's not a standard package structure. The NumPy-style approach is to import the package folder name (mypackage) and access functions from there, which is what we've set up.

内容的提问来源于stack exchange，提问作者zltn.guba