Modules & Imports

1. What Is It?

A module is a file (or package) of related code you can pull into another file by name. Imports are the mechanism for doing that pulling. Without modules, every program is one giant source file, nothing is reusable, and naming collisions are inevitable.

The concept solves three problems at once: reuse (one math library, used by everyone), namespacing (your parse and mine don't collide), and encapsulation (exposing a deliberate API while hiding the rest).

2. How It Works

At its core, importing does two things:

1. Find and load the module — locate the file on disk, compile/parse it, execute any top-level code.
2. Bind names into the importing scope so you can refer to the module's contents.

# Python — file math_utils.py
PI = 3.14159
def area(r): return PI * r * r

# Python — in another file
import math_utils                            # load, bind 'math_utils'
print(math_utils.area(2))

from math_utils import area, PI              # bind specific names
from math_utils import area as compute_area  # rename

import math_utils as mu                      # whole-module alias

// Java — elsewhere
import geom.Circle;            // make Circle visible by short name
double a = Circle.area(2);

import static geom.Circle.area; // bring the static method itself into scope
double b = area(2);

#include is not import

C++'s #include is a preprocessor directive that pastes the header file's text into the including file. The compiler then compiles the combined result. This is why you need include guards (#pragma once or #ifndef ... #define ...) — without them, a header included twice defines everything twice.

Modern C++ (modules introduced in C++20; import std; standardized in C++23) provides real modules that behave more like Python/Java — but #include remains standard in most codebases today.

Python import mechanics

When Python sees import foo:

1. Check sys.modules — already loaded? Return the cached module.
2. Search sys.path (a list of directories) for foo.py or foo/ package.
3. Add the new module to sys.modules, then execute its top-level code once.
4. Bind foo in the current namespace.

Top-level code runs at import time, not just when you call a function inside. Print statements, side effects, network calls — they all fire the first time anyone imports the module.

3. What You Actually Need to Know

• Import once, used everywhere. Python caches modules in sys.modules. A second import foo is a dict lookup, not a re-execution. This is also why editing a module isn't seen by a running interpreter — you need to restart or use importlib.reload.
• Avoid from module import *. It dumps every public name into your scope, causing collisions and making it impossible to trace where a name came from. Prefer explicit imports.
• Circular imports happen when a.py imports b.py and b.py imports a.py. Python partially loads one, hits the second, and you get a half-initialized module with some names missing. The fix is usually restructuring (move shared code into a third module) or deferring the import to inside a function.
• Package vs. module. A module is a single file. A package is a directory containing an __init__.py (Python) or a namespace (Java package, C++ namespace). Packages let you group related modules and nest them.
• Java's package = directory. Package com.example.geom must live in com/example/geom/. The class loader and build tools rely on this layout (javac itself does not reject a source file in another directory).
• C++ header/implementation split. Headers declare what exists (signatures, class definitions). .cpp files implement them. Headers are for everyone who uses your code; implementations get compiled separately and linked.
• Debugging clues:
  • ModuleNotFoundError / ImportError (Python) → wrong path, missing __init__.py, or name collision with a local file.
  • ClassNotFoundException (Java runtime) → class wasn't on the classpath.
  • C++ linker error undefined reference → you declared something in a header but never compiled/linked its implementation.
  • Surprising side effects at startup → module's top-level code runs on import; move heavy work into functions.

4. Language Differences

Python's "execute at import" model is the most surprising of the three — a module isn't just declarations; it's a script that runs the first time it's imported.

5. Tradeoffs & Decisions

• Explicit imports vs. star imports. Explicit is almost always better — grep-ability, no collisions, IDE autocomplete. Star imports are OK for a small, well-known utility module or inside a REPL.
• Aliasing. import numpy as np is a convention that saves typing and makes code portable across projects. Aliasing is good when the full name is long or clashes.
• Import location: top of file vs. inside function. Top-of-file is standard — it's clear what the file depends on. Deferred (inside-function) imports are justified only for resolving circular imports or avoiding expensive loads at startup.
• Package boundaries as API contracts. What's exported vs. internal matters. Python has __all__ and leading underscores as conventions. Java has public/package-private/private. C++ separates .h (public) from .cpp (internal) and uses anonymous namespaces for file-local functions.
• Dependency depth. A module that transitively imports thousands of lines takes longer to load and is a maintenance liability. Keep imports shallow where possible.

If you see a circular import, it usually means two modules should be one, or a third module should hold their shared code. You'd choose deferred imports only when restructuring isn't possible.

6. Interview Cheat Sheet

• A module is a reusable unit of code (file or package); imports bind names from that module into your current scope.
• Python imports execute the module's top-level code once and cache the result in sys.modules — subsequent imports are cheap.
• C++ #include pastes text; it's a preprocessor operation, not a true import. Hence header guards and the header/.cpp split.
• Circular imports happen when two modules need each other at load time; fix by restructuring or deferring.
• Avoid from m import * — it pollutes the namespace and hides the source of names.
• Package boundaries are your API contract — be deliberate about what's public vs. internal.

Follow-ups:

• "What does import actually do in Python?" — Locate the module on sys.path, add it to sys.modules, execute its top-level code once, and bind the module name in the caller's namespace.
• "Why do C++ headers need include guards?" — #include pastes the header text; without guards, including a header transitively twice would redefine everything and cause a compile error.
• "How do circular imports break?" — The first module is partially loaded when the second tries to use it, so some names it needs don't yet exist.