pytest-testing-guidelines

name	Pytest Testing Guidelines
description	This skill should be used when the user asks to "write tests", "create tests", "generate tests", "unit tests", "pytest", "test guidelines", "testing rules", "test standards", "how to test", "test this code", "add tests", "validate tests", "check tests", "improve tests", or when working with Python test files. Provides comprehensive guidelines for creating pytest unit tests with 100% coverage following strict naming conventions, mock patterns, and parametrization standards.
version	0.1.0

Pytest Testing Guidelines

Overview

These guidelines define a rigorous testing approach for Python projects using pytest, emphasizing 100% code coverage, strict naming conventions, comprehensive mock verification, and maintainable test structure. Follow these guidelines to create tests that ensure code reliability and prevent regressions.

Core Principles

100% Coverage Requirement: Every line of code must be tested. This ensures that any code change will not have negative impact on untested code.

Comprehensive Mock Verification: All mock interactions must be verified through the mock_calls property. This ensures that all manipulations of mocks are checked, preventing incomplete test coverage of external dependencies.

Consistent Naming: Strict naming conventions improve test readability and maintenance, making it immediately clear what each test covers.

Test Order Matches Source: Tests should appear in the same order as methods in the source file, making it easy to verify complete coverage by visual inspection.

Source Code Requirements

Type Annotations Validation

MANDATORY PRE-CHECK: Before generating tests, improving tests, or validating tests, the commands generate-tests, improve-tests, and validate-tests must first verify that the source code has complete type annotations.

Required type annotations:

• All method/function parameters must have type annotations
• All method/function return values must have type annotations
• Class attributes should have type annotations where applicable

Example of properly typed code:

class DataProcessor:
    def __init__(self, config: dict[str, str]) -> None:
        self.config = config

    def process(self, data: list[str], validate: bool = True) -> dict[str, int]:
        # Implementation
        pass

    def _validate(self, item: str) -> bool:
        # Implementation
        pass

Example of code missing type annotations (will be rejected):

class DataProcessor:
    def __init__(self, config):  # Missing parameter and return types
        self.config = config

    def process(self, data, validate=True):  # Missing all type annotations
        pass

    def _validate(self, item):  # Missing parameter and return types
        pass

Validation process:

1. Parse the source file to identify all functions and methods
2. Check each function/method for:
   • Type annotations on ALL parameters (except self and cls)
   • Return type annotation (including -> None for void functions)
3. If any type annotations are missing:
   • Report which functions/methods are missing annotations
   • List the specific parameters or return types that need annotations
   • STOP - do not proceed with test generation/improvement/validation
4. Only proceed with the requested operation if all type annotations are present

Why this matters:

• Type annotations enable better test generation with correct expected types
• They improve mock configuration accuracy
• They help verify test assertions match the expected return types
• They ensure tests validate the documented contract of the code

Naming Conventions

Test File Names

Mirror the source file structure in the tests/ directory:

src/utils/parser.py → tests/utils/test_parser.py
src/models/user.py → tests/models/test_user.py

Each test file name must start with test_ prefix.

Test Function Names

Follow this pattern: test_<method_name> for single-scenario tests, or test_<method_name>__<case_description> for multiple scenarios.

Basic format:

def test_calculate_total():
# Tests the calculate_total method

Include underscore prefix for private methods:

# For method named _private_helper
def test__private_helper():
# Underscore is included in test name

Multiple test cases:

def test_validate_input__valid_data():


# Test with valid data

def test_validate_input__invalid_format():


# Test with invalid format

def test_validate_input__empty_input():
# Test with empty input

The double underscore __ separates the method name from the case description.

Variable Names

Use these standard names consistently across all tests:

• tested: The class or instance being tested (MUST be assigned FIRST in every test)
• result: The value returned from the method under test (MUST be used for return values)
• expected: The expected value for single assertions (MUST be used when comparing result)
• exp_*: Expected values for multiple assertions (e.g., exp_calls, exp_output, exp_status)

CRITICAL - Variable Naming Requirements:

Every test MUST follow these naming conventions strictly:

1. The tested Variable

• Assign the class or instance being tested to a tested variable FIRST
• Call methods on tested, NEVER directly on the class name

Example for instance methods:

def test_format_name():
    tested = NameFormatter()  # Assign instance to tested
    result = tested.format_name("john", "doe")  # Call on tested
    expected = "John Doe"
    assert result == expected

Example for class/static methods:

def test_parse_arguments():
    tested = SvgToPngConverter  # Assign CLASS to tested (not an instance)
    result = tested.parse_arguments()  # Call on tested
    expected = ConversionInput(...)
    assert result == expected

FORBIDDEN:

# WRONG - calling directly on class name
def test_parse_arguments():
    result = SvgToPngConverter.parse_arguments()  # FORBIDDEN
    ...

2. The result Variable

• ALWAYS store the return value from the method under test in a variable named result
• NEVER use alternative names like output, ret, actual, response, value, etc.

CORRECT:

def test_calculate_total():
    tested = Calculator()
    result = tested.calculate_total([1, 2, 3])  # CORRECT: use 'result'
    expected = 6
    assert result == expected

FORBIDDEN:

def test_calculate_total():
    tested = Calculator()
    output = tested.calculate_total([1, 2, 3])  # FORBIDDEN: don't use 'output'
    total = tested.calculate_total([1, 2, 3])  # FORBIDDEN: don't use 'total'
    ret = tested.calculate_total([1, 2, 3])  # FORBIDDEN: don't use 'ret'
    ...

3. The expected Variable

• ALWAYS store the expected value for the main assertion in a variable named expected
• NEVER use alternative names like want, exp, correct, answer, etc.
• NEVER inline the expected value in the assertion

CORRECT:

def test_format_date():
    tested = DateFormatter()
    result = tested.format("2024-01-15")
    expected = "January 15, 2024"  # CORRECT: use 'expected'
    assert result == expected

FORBIDDEN:

def test_format_date():
    tested = DateFormatter()
    result = tested.format("2024-01-15")
    assert result == "January 15, 2024"  # FORBIDDEN: inlined expected value

def test_format_date():
    tested = DateFormatter()
    result = tested.format("2024-01-15")
    want = "January 15, 2024"  # FORBIDDEN: don't use 'want'
    assert result == want

4. The exp_* Pattern for Multiple Assertions

• Use exp_ prefix for expected values when a test has multiple assertions
• Common patterns: exp_calls, exp_output, exp_status, exp_error, exp_mock_calls
• NEVER use expected_ prefix (too long) - always use exp_

CORRECT:

@patch('module.api')
def test_fetch_data(mock_api):
    mock_api.get.side_effect = [{"data": "value"}]

    tested = DataFetcher()
    result = tested.fetch("http://example.com")

    expected = {"data": "value"}
    assert result == expected

    exp_api_calls = [call.get("http://example.com")]  # CORRECT: use 'exp_' prefix
    assert mock_api.mock_calls == exp_api_calls

FORBIDDEN:

@patch('module.api')
def test_fetch_data(mock_api):
    ...
    expected_calls = [call.get("http://example.com")]  # FORBIDDEN: don't use 'expected_'
    expected_api_calls = [call.get("http://example.com")]  # FORBIDDEN: too long
    calls = [call.get("http://example.com")]  # FORBIDDEN: not descriptive
    ...

For detailed naming examples, see references/naming-conventions.md.

Assert Complete Outputs, Not Subsets

CRITICAL: When a method produces a collection of outputs (a list, a set of files, written text, created objects), assert on the complete output using equality — not on a subset using membership or individual element checks.

Partial assertions let source-level additions (a new item in a loop, a new line in a file, a new key in a dict) go completely undetected even at 100% line coverage.

FORBIDDEN (subset / membership checks):

# WRONG — only checks two of three symlinks; adding a third goes unnoticed
assert (clinic_home / ".claude").is_symlink()
assert (clinic_home / ".ssh").is_symlink()

# WRONG — only checks partial file content
assert "line1" in content
assert "line2" in content

CORRECT (equality on the complete output):

# CORRECT — equality covers ALL symlinks; adding one fails the test immediately
created = {p.name for p in clinic_home.iterdir() if p.is_symlink()}
expected = {".claude", ".ssh", ".gitconfig"}
assert created == expected

# CORRECT — full file content equality
expected = "line1\nline2\nline3\n"
assert content == expected

Rule of thumb: if the full expected value is deterministic (a constant tuple, a known list of keys, a fixed file content), use ==. Reserve in / is_symlink() / individual element checks only for cases where the output is genuinely partial by design.

This applies everywhere:

• File contents: assert the complete text, not just that certain lines are present
• Collections returned by methods: assert the complete list/set, not just selected items
• Side effects (files created, directories made): verify the complete set of expected side effects

Assertion Style

Singleton Comparisons: Use is (not ==) when comparing to singletons: True, False, None.

Correct:

def test_is_valid():
    tested = Validator()
    result = tested.is_valid("data")
    expected = True
    assert result is expected  # Use 'is' for boolean singletons

Incorrect:

def test_is_valid():
    tested = Validator()
    result = tested.is_valid("data")
    expected = True
    assert result == expected  # Wrong: use 'is' not '=='

When to use is:

• Comparing to True
• Comparing to False
• Comparing to None

When to use ==:

• All other comparisons (strings, numbers, lists, dicts, objects, etc.)

Test Structure and Order

Test Order Matches Source Code

Tests must appear in the same order as methods appear in the source file. This enables:

• Quick visual verification of coverage
• Easy navigation between source and tests
• Clear indication of missing tests

Source file order:

class Calculator:
    def __init__(self):
        pass

    def add(self, a, b):
        pass

    def subtract(self, a, b):
        pass

    def _validate(self, value):
        pass

Test file order:

def test_add():
    pass

def test_subtract():
    pass

def test__validate():
    pass

First Test: Verify Inheritance

MANDATORY: When testing a class that inherits from a base class, the very first test in the test file must verify the inheritance relationship using issubclass().

This test ensures that:

• The class is properly inheriting from the expected base class
• The inheritance hierarchy is correct
• Any refactoring that breaks inheritance is immediately caught

Example:

# Source: scripts/user_input_logger.py
class UserInputsLogger(BaseLogger):
    def process(self, data):
        pass


# Test file: tests/scripts/test_user_input_logger.py
def test_inheritance():
    """Verify UserInputsLogger inherits from BaseLogger."""
    assert issubclass(UserInputsLogger, BaseLogger)


def test_process():
    # Remaining tests follow...
    pass

Pattern:

def test_inheritance():
    """Verify <ClassName> inherits from <BaseClassName>."""
    assert issubclass(ClassName, BaseClassName)

This test should:

• Always be named test_inheritance
• Always be the first test in the file
• Include a docstring describing the inheritance relationship
• Use a simple assert issubclass(DerivedClass, BaseClass) statement

Testing NamedTuple Classes

MANDATORY: When testing a NamedTuple class, the first test must verify the NamedTuple structure using the is_namedtuple() helper function.

The test must:

1. Be named test_class (not test_inheritance)
2. Assign the class itself to the tested variable (not an instance)
3. Define a fields dictionary with field names and their types
4. Use the is_namedtuple() helper function to verify the structure

Setup - Add helper to conftest.py:

Add this function to your tests/conftest.py file (create if it doesn't exist). This makes the helper available to all test files automatically:

# tests/conftest.py
from typing import get_type_hints


def is_namedtuple(cls, fields: dict) -> bool:
    """
    Verify that a class is a NamedTuple with the expected fields and types.

    Args:
        cls: The class to check
        fields: Dictionary mapping field names to their expected types

    Returns:
        bool: True if cls is a NamedTuple with exactly the specified fields and types
    """
    return (
            issubclass(cls, tuple)
            and hasattr(cls, "_fields")
            and isinstance(cls._fields, tuple)
            and len([field for field in cls._fields if field in fields]) == len(fields.keys())
            and get_type_hints(cls) == fields
    )

Example usage in tests:

# Source: models/validation_result.py
from typing import NamedTuple

class ValidationResult(NamedTuple):
    has_errors: bool
    errors: list[str]

# Test file: tests/models/test_validation_result.py
from validation_result import ValidationResult

# Note: is_namedtuple is automatically available from conftest.py

def test_class():
    """Verify ValidationResult is a NamedTuple with correct fields."""
    tested = ValidationResult
    fields = {"has_errors": bool, "errors": list[str]}
    assert is_namedtuple(tested, fields)

Why use is_namedtuple() instead of issubclass():

• Verifies it's a tuple subclass
• Checks the _fields attribute exists and is correct
• Validates all expected fields are present
• Ensures field types match expectations using get_type_hints()
• Catches any changes to the NamedTuple structure

For a complete NamedTuple testing example, see examples/namedtuple-test-file.py.

Testing Dataclass Classes

MANDATORY: When testing a dataclass, the first test must verify the dataclass structure using the is_dataclass() helper function.

The test must:

1. Be named test_class (not test_inheritance)
2. Assign the class itself to the tested variable (not an instance)
3. Define a fields dictionary with field names and their type strings
4. Use the is_dataclass() helper function to verify the structure

Setup - Add helper to conftest.py:

Add this function to your tests/conftest.py file alongside is_namedtuple():

# tests/conftest.py
from dataclasses import fields as dataclass_fields, is_dataclass as dataclass_is_dataclass


def is_dataclass(cls, fields: dict) -> bool:
    """
    Verify that a class is a dataclass with the expected fields and types.

    Args:
        cls: The class to check
        fields: Dictionary mapping field names to their expected types (can be type objects or strings)

    Returns:
        bool: True if cls is a dataclass with exactly the specified fields and types
    """
    if not dataclass_is_dataclass(cls):
        return False
    actual_fields = dataclass_fields(cls)
    if len([field for field in actual_fields if field.name in fields]) != len(fields.keys()):
        return False
    for field in actual_fields:
        expected_type = fields[field.name]
        actual_type = field.type
        if expected_type != actual_type:
            return False
    return True

Example usage in tests:

# Source: models/transcript_segment.py
from dataclasses import dataclass

@dataclass
class TranscriptSegment:
    speaker: str
    text: str
    chunk: int
    start: float
    end: float

# Test file: tests/models/test_transcript_segment.py
from transcript_segment import TranscriptSegment

# Note: is_dataclass is automatically available from conftest.py

def test_class():
    """Verify TranscriptSegment is a dataclass with correct fields."""
    tested = TranscriptSegment
    fields = {
        "speaker": "str",
        "text": "str",
        "chunk": "int",
        "start": "float",
        "end": "float",
    }
    assert is_dataclass(tested, fields)

Why use is_dataclass() helper:

• Verifies the class is decorated with @dataclass
• Checks all expected fields are present
• Validates field types match expectations
• Catches any changes to the dataclass structure
• More comprehensive than checking __dataclass_fields__ manually

For a complete dataclass testing example, see examples/dataclass-test-file.py.

One Test Per Method Minimum

Every method must have at least one test. If a method requires multiple test scenarios, prefer parametrization (see below) or create multiple test functions with case suffixes.

Exclude __main__ Blocks from Tests

IMPORTANT: Do NOT write tests for if __name__ == "__main__": blocks.

These blocks are:

1. Excluded from coverage by default (configured in pyproject.toml via exclude_lines)
2. Entry points that simply call other methods which should already be tested
3. Difficult to test properly without hacky approaches like exec(compile(...))

FORBIDDEN:

class TestMainBlock:
    """Tests for the __main__ block execution."""

    def test_main_runs(self) -> None:
        # FORBIDDEN: Don't test __main__ blocks
        exec(compile('if __name__ == "__main__": ...', "<string>", "exec"), ...)

Why this matters:

• The __main__ block typically just calls ClassName.run() which should already have tests
• Testing it requires hacky approaches that are fragile and hard to maintain
• Coverage tools are configured to exclude these lines, so testing them adds no value
• The effort is better spent on testing the actual logic in the methods being called

Mock Usage

Use side_effect for Return Values

MANDATORY: When configuring mock return values, use side_effect instead of return_value. This is a strict requirement and must never be violated. For call chains, only the final call in the chain must use side_effect (for example: mock.return_value.get.return_value.add.side_effect = [...]).

Correct:

@patch('module.api_client')
def test_fetch_data(mock_client):
    mock_client.return_value.get.side_effect = [{"status": "success"}]

Incorrect - FORBIDDEN:

@patch('module.api_client')
def test_fetch_data(mock_client):
    mock_client.return_value.get.return_value = {"status": "success"}  # NEVER use return_value for the final call

For complex return objects (like HTTP responses), use SimpleNamespace instead of MagicMock:

CRITICAL: When a mock returns an object with attributes or methods, use SimpleNamespace NOT MagicMock. This avoids the need to verify mock_calls on the returned object.

CORRECT - Use SimpleNamespace for response objects:

from types import SimpleNamespace

@patch('module.requests.post')
def test_api_call(mock_post):
    # Use SimpleNamespace - no need to verify mock_calls on response
    mock_post.side_effect = [
        SimpleNamespace(
            status_code=200,
            text="response text",
            json=lambda: {"data": "value"}  # Lambda for methods that return values
        )
    ]

    tested = ApiClient()
    result = tested.call_api()

    expected = {"data": "value"}
    assert result == expected

    # Only need to verify mock_post, NOT the response object
    exp_post_calls = [call("https://api.example.com")]
    assert mock_post.mock_calls == exp_post_calls

FORBIDDEN - Using MagicMock for response objects:

@patch('module.requests.post')
def test_api_call(mock_post):
    # FORBIDDEN: Using MagicMock for response requires mock_calls verification
    mock_response = MagicMock()
    mock_response.status_code = 200
    mock_response.json.side_effect = [{"data": "value"}]
    mock_post.side_effect = [mock_response]

    result = tested.call_api()

    # If you use MagicMock, you MUST verify its mock_calls - easy to forget!
    exp_response_calls = [call.json()]
    assert mock_response.mock_calls == exp_response_calls  # Often forgotten!

Why SimpleNamespace is preferred:

• No mock_calls verification needed - it's just a data container
• Simpler test code with less boilerplate
• No risk of forgetting to verify mock interactions
• Lambdas work for methods: json=lambda: {...}, read=lambda: b"data"

Use capsys Instead of Mocking print

CRITICAL: When testing code that uses print(), use pytest's built-in capsys fixture instead of mocking print.

FORBIDDEN - Mocking print:

@patch("module.print")
def test_output(mock_print):
    tested = MyClass()
    tested.display_message()

    # Complex extraction of print calls - error prone!
    print_calls = [c[0][0] for c in mock_print.call_args_list if c[0]]
    assert "Expected message" in print_calls

    # Must also verify mock_calls - easy to forget!
    exp_print_calls = [call("Expected message")]
    assert mock_print.mock_calls == exp_print_calls

CORRECT - Using capsys fixture:

def test_output(self, capsys) -> None:
    tested = MyClass()
    tested.display_message()

    captured = capsys.readouterr()

    assert "Expected message" in captured.out
    # Or for exact match:
    expected = "Expected message\n"
    assert captured.out == expected

Why capsys is preferred:

• Built-in pytest fixture - no imports or patches needed
• Simpler assertion syntax - just check captured.out or captured.err
• No mock_calls verification required
• Works with any code that writes to stdout/stderr
• Captures actual output, not mock interactions
• Less boilerplate and setup code

capsys usage patterns:

def test_stdout(self, capsys) -> None:
    print("Hello")
    captured = capsys.readouterr()
    assert captured.out == "Hello\n"

def test_stderr(self, capsys) -> None:
    import sys
    print("Error", file=sys.stderr)
    captured = capsys.readouterr()
    assert captured.err == "Error\n"

def test_multiple_prints(self, capsys) -> None:
    print("Line 1")
    print("Line 2")
    captured = capsys.readouterr()
    assert "Line 1" in captured.out
    assert "Line 2" in captured.out

Verify All Mocks with mock_calls

MANDATORY: After each test, ALL mock objects must be verified through the mock_calls property. This includes:

• The main patched mock (e.g., mock_db)
• Any MagicMock() objects you create (e.g., mock_response)
• Any .return_value mock instances (e.g., mock_db_class.return_value)

CRITICAL - Parametrized Tests Are NOT Exempt: Parametrized tests MUST verify all mocks just like regular tests. If a test has mock parameters, each mock MUST have a corresponding assert mock.mock_calls == exp_* statement.

FORBIDDEN - Parametrized test without mock verification:

@pytest.mark.parametrize("input_val,expected", [...])
@patch("module.api")
def test_fetch(mock_api, input_val, expected):
    mock_api.get.side_effect = [{"data": input_val}]
    tested = Fetcher()
    result = tested.fetch(input_val)
    assert result == expected
    # FORBIDDEN: mock_api is NEVER verified with mock_calls!

CORRECT - Parametrized test with mock verification:

@pytest.mark.parametrize("input_val,expected", [...])
@patch("module.api")
def test_fetch(mock_api, input_val, expected):
    mock_api.get.side_effect = [{"data": input_val}]
    tested = Fetcher()
    result = tested.fetch(input_val)
    expected_result = expected
    assert result == expected_result

    exp_api_calls = [call.get(input_val)]  # MUST verify mock
    assert mock_api.mock_calls == exp_api_calls

When to avoid parametrization with mocks: If mock verification differs significantly between test cases, use separate non-parametrized tests instead of trying to parametrize with complex mock verification logic.

CRITICAL: Always verify at the mock object level, NOT at individual method level:

• CORRECT: assert mock_response.mock_calls == exp_calls (object level)
• FORBIDDEN: assert mock_response.read.mock_calls == exp_calls (method level)

CRITICAL: Use single assertion with hard-coded values:

• CORRECT: assert mock.mock_calls == exp_calls (single assertion)
• FORBIDDEN: assert len(mock.mock_calls) == 3 then checking individual items
• FORBIDDEN: Using variables in expected values: call(f"URL: {tested_url}")
• CORRECT: Use hard-coded literals: call("URL: http://example.com")

Correct verification:

Using SimpleNamespace:

@patch('module.requests.get')
def test_fetch_data(mock_get):
    # Create mock response object
    mock_get.side_effect = [SimpleNamespace(status_code=200, json=lambda: {"data": "value"})]

    tested = APIClient()
    result = tested.fetch_data("http://api.example.com/data")

    expected = {"data": "value"}
    assert result == expected

    # CRITICAL: Verify with single assertion and HARD-CODED values
    exp_get_calls = [call("http://api.example.com/data")]
    assert mock_get.mock_calls == exp_get_calls

Using embedded mocks:

@patch('module.requests.get')
def test_fetch_data(mock_get):
    # Create mock response object
    mock_response = MagicMock()
    mock_response.json.side_effect = [{"data": "value"}]
    mock_get.side_effect = [mock_response]

    tested = APIClient()
    result = tested.fetch_data("http://api.example.com/data")

    expected = {"data": "value"}
    assert result == expected

    # CRITICAL: Verify with single assertion and HARD-CODED values
    exp_get_calls = [call("http://api.example.com/data")]
    assert mock_get.mock_calls == exp_get_calls

    # CRITICAL: Verify response mock at OBJECT level with hard-coded values
    exp_response_calls = [call.json()]
    assert mock_response.mock_calls == exp_response_calls  # Single assertion!

Common mistakes:

1. Forgetting to verify mock objects:

# WRONG - mock_response is not verified!
@patch('module.requests.get')
def test_fetch_data(mock_get):
    mock_response = MagicMock()
    mock_response.json.side_effect = [{"data": "value"}]
    mock_get.side_effect = [mock_response]

    result = tested.fetch_data("url")

    # FORBIDDEN: Only verifying mock_get, missing mock_response!
    assert mock_get.mock_calls == [call("url")]

2. Verifying at wrong level:

# WRONG - verifying individual methods instead of object!
@patch('module.urlopen')
def test_fetch(mock_urlopen):
    mock_response = MagicMock()
    mock_response.read.side_effect = [b"data"]
    mock_urlopen.side_effect = [mock_response]

    result = fetch("url")

    # FORBIDDEN: Verifying at method level!
    assert mock_response.read.mock_calls == [call()]  # WRONG LEVEL

    # CORRECT: Verify at object level
    exp_response_calls = [call.read()]
    assert mock_response.mock_calls == exp_response_calls  # Object level!

3. Checking length and using variables:

# WRONG - checking length then individual calls with variables!
@patch('builtins.print')
def test_process(mock_print):
    tested_url = "http://example.com"
    process(tested_url)

    # FORBIDDEN: Checking length and individual items
    assert len(mock_print.mock_calls) == 2
    assert mock_print.mock_calls[0] == call(f"Processing: {tested_url}")
    assert mock_print.mock_calls[1] == call("Done")

    # CORRECT: Single assertion with hard-coded values
    exp_print_calls = [
        call("Processing: http://example.com"),  # Hard-coded!
        call("Done")
    ]
    assert mock_print.mock_calls == exp_print_calls

Forbidden Mock Assertions

NEVER USE these assertion helper methods:

• mock.assert_called()
• mock.assert_called_once()
• mock.assert_called_with(...)
• mock.assert_called_once_with(...)
• mock.assert_not_called()
• mock.assert_any_call(...)
• mock.assert_has_calls(...)
• mock.call_count
• mock.call_args
• mock.call_args_list

NEVER USE ANY from unittest.mock in mock_calls assertions or expected values:

• ANY is a lazy matcher that defeats the purpose of precise mock verification
• Instead of call.method(ANY), construct the exact expected argument
• If the argument is hard to predict (e.g., a timestamp), mock the source of non-determinism and construct the exact expected object

FORBIDDEN:

from unittest.mock import ANY

# FORBIDDEN: ANY hides what the actual argument should be
assert mock_client.mock_calls == [call.send(ANY)]

# FORBIDDEN: ANY in expected values
assert result == ANY

CORRECT — mock the non-deterministic source and construct exact expected values:

@patch("module.Email.now")  # Mock the timestamp source
@patch("module.EmailClient")
def test_send(mock_client_cls, mock_email_now):
    mock_email_now.side_effect = ["2026-01-01T00:00:00Z"]
    # ... setup ...

    expected_email = Email(
        subject="Hello",
        send_at="2026-01-01T00:00:00Z",  # Exact value from mocked source
    )
    assert mock_client.mock_calls == [call.send(expected_email)]  # Exact match!

NEVER verify at method or attribute level:

• mock_response.read.mock_calls (FORBIDDEN - method level)
• mock_response.read.return_value.decode.mock_calls (FORBIDDEN - nested attribute level)

NEVER check length or index mock_calls:

• assert len(mock.mock_calls) == 3 (FORBIDDEN - checking length)
• assert mock.mock_calls[0] == call(...) (FORBIDDEN - indexing)
• assert call(...) == mock.mock_calls[1] (FORBIDDEN - indexing with reversed order)

NEVER use variables in expected values:

• call(f"URL: {tested_url}") (FORBIDDEN - using variable)
• call(tested_value) (FORBIDDEN - using variable)
• Always use hard-coded literals: call("URL: http://example.com") (CORRECT)

ALWAYS use:

• mock.mock_calls property at the object level for verification
• Pattern: assert mock_object.mock_calls == exp_calls (single assertion)
• Hard-coded literal values in all expected calls

For comprehensive mock patterns and examples, see references/mock-patterns.md.

Multiple Scenarios

When testing multiple scenarios for the same method, use these approaches in order of preference:

1. Parametrize (Preferred)

Use @pytest.mark.parametrize with pytest.param to define each scenario:

@pytest.mark.parametrize("input_value,expected", [
    pytest.param(5, 10, id="positive"),
    pytest.param(-5, 0, id="negative"),
    pytest.param(0, 5, id="zero"),
])
def test_add_five(input_value, expected):
    tested = Calculator()
    result = tested.add_five(input_value)
    assert result == expected

Benefits:

• Compact representation
• Clear scenario identification
• Easy to add new cases
• Better test output with id parameter

2. Loop Within Test

For simple scenarios with similar setup:

def test_validate_email():
    tested = Validator()

    test_cases = [
        ("[e-mail]", True),
        ("invalid.email", False),
        ("@example.com", False),
    ]

    for email, expected in test_cases:
        result = tested.validate_email(email)
        assert result == expected

3. Multiple Test Functions

For complex scenarios requiring different setup or mocks:

def test_process_data__valid_input():
    # Complex setup for valid input
    pass


def test_process_data__invalid_format():
    # Different setup for invalid format
    pass


def test_process_data__network_error():
    # Mock network error scenario
    pass

For detailed parametrization patterns, see references/parametrize-examples.md.

Mock Scenarios

When to Use Mocks

Use mocks for:

1. External systems: Database queries, HTTP requests, file I/O
2. Non-deterministic behavior: datetime.now(), random.random(), uuid.uuid4()
3. Complex setup: When creating the real scenario would be overly complicated
4. Internal class dependencies: When a method depends on other methods of the same class, mock those methods to avoid test duplication

Example with internal method dependency:

class DataProcessor:
    def validate(self, data):
        # Complex validation logic tested separately
        ...

    def process(self, data):
        if not self.validate(data):
            raise ValueError("Invalid data")
        # Processing logic
        return transformed_data

# GOOD - Mock validate() to focus on process() logic only
@patch.object(DataProcessor, "validate")
def test_process(mock_validate):
    mock_validate.side_effect = [True]

    tested = DataProcessor()
    result = tested.process({"key": "value"})

    expected = {"key": "transformed"}
    assert result == expected

    exp_validate_calls = [call({"key": "value"})]
    assert mock_validate.mock_calls == exp_validate_calls

# BAD - Testing validate() behavior again inside process() tests
def test_process__duplicates_validation_tests():
    tested = DataProcessor()
    # This test ends up re-testing validate() logic
    result = tested.process({"key": "value"})
    ...

Example with datetime:

@patch('module.datetime')
def test_create_timestamp(mock_datetime):
    mock_datetime.now.side_effect = [datetime(2024, 1, 1, 12, 0, 0)]

    tested = TimestampGenerator()
    result = tested.create_timestamp()

    expected = "2024-01-01 12:00:00"
    assert result == expected

    exp_calls = [call.now()]
    assert mock_datetime.mock_calls == exp_calls

When NOT to Use Mocks

Prefer real instances over mocks for simple data objects. Mocks should isolate code from external dependencies, not replace simple data structures.

Use real instances when:

1. Simple data containers: NamedTuples, dataclasses, TypedDicts, or plain classes that only hold data
2. No side effects: Objects that don't perform I/O, network calls, or modify external state
3. Trivial construction: When creating a real instance requires no more effort than configuring a mock and checking its calls

IMPORTANT for NamedTuples and Dataclasses:

• NEVER mock NamedTuple classes or instances. NamedTuples are immutable data containers with no side effects - always use real instances created with keyword arguments.
• NEVER mock dataclass classes or instances. Dataclasses are data containers - always use real instances created with keyword arguments. Only mock external dependencies or internal methods when testing other methods on the same class.

Example - Prefer real instance:

from hook_information import HookInformation


# GOOD - HookInformation is a simple NamedTuple, use real instance
def test_session_directory():
    hook_info = HookInformation(
        session_id="test123",
        exit_reason="user_exit",
        transcript_path=Path("/path/to/transcript.jsonl"),
        workspace_dir=Path("/home/user/project"),
        working_directory=Path("/home/user/project/subdir"),
    )

    result = UserInputsLogger.session_directory(hook_info)

    expected = Path("/home/user/project/.artifacts/user_inputs")
    assert result == expected

Example - Avoid unnecessary mocks:

# BAD - Using Mock for a simple data object adds indirection without benefit
def test_session_directory():
    mock_hook_info = Mock()
    mock_hook_info.workspace_dir = Path("/home/user/project")

    result = UserInputsLogger.session_directory(mock_hook_info)
    # ...

Why prefer real instances:

• Type safety: Real objects catch attribute typos and type errors at test time
• Better documentation: Tests using real objects show the actual interface requirements
• Detect breaking changes: Mocks can hide breaking changes (e.g., renamed attributes still "work" on mocks)
• Simpler tests: No need to configure mock attributes when a real instance is just as easy to create

Rule of thumb: If the object is a data container with no behavior to mock (no methods that perform I/O, no side effects), use a real instance.

Running Tests

Tests must pass these commands without errors:

Run All Tests

uv run pytest tests/

Run with Coverage

uv run pytest -v tests/ --cov=.

Coverage must reach 100% for all source files.

Run Specific Test File

uv run pytest tests/utils/test_parser.py -v

Complete Examples

Regular Classes

See examples/complete-test-file.py for a full working example that demonstrates:

• Proper test file structure with inheritance testing
• Naming conventions for all scenarios
• Mock usage with side_effect and mock_calls
• Parametrization for multiple scenarios
• 100% coverage of source file

NamedTuple Classes

See examples/namedtuple-test-file.py for a full working example that demonstrates:

• NamedTuple structure verification with is_namedtuple() helper
• Testing NamedTuple methods and immutability
• Creating test instances with keyword arguments
• Parametrization for NamedTuple scenarios
• 100% coverage of NamedTuple source file

Dataclass Classes

See examples/dataclass-test-file.py for a full working example that demonstrates:

• Dataclass structure verification with is_dataclass() helper
• Testing dataclass methods and mutability
• Testing default values and default_factory fields
• Mocking internal methods to avoid test duplication
• Creating test instances with keyword arguments
• Parametrization for dataclass scenarios
• 100% coverage of dataclass source file

Additional Resources

Reference Files

For detailed guidance on specific topics:

• references/naming-conventions.md - Comprehensive naming examples for all scenarios
• references/mock-patterns.md - Mock patterns with side_effect and mock_calls verification
• references/parametrize-examples.md - Advanced parametrization techniques

Example Files

Working examples demonstrating guidelines:

• examples/conftest.py - Example conftest.py with is_namedtuple() and is_dataclass() helper functions
• examples/complete-test-file.py - Complete test file with source code showing all patterns for regular classes
• examples/namedtuple-test-file.py - Complete test file demonstrating NamedTuple testing patterns
• examples/dataclass-test-file.py - Complete test file demonstrating dataclass testing patterns

Quick Reference

Test structure:

• For classes with inheritance: First test must be test_inheritance() verifying issubclass(DerivedClass, BaseClass)
• For NamedTuple classes: First test must be test_class() using is_namedtuple(tested, fields) helper (add helper to tests/conftest.py)
• For dataclass classes: First test must be test_class() using is_dataclass(tested, fields) helper with string type values (add helper to tests/conftest.py)

Test naming: test_method_name or test_method_name__case

Variable naming (MANDATORY - use these exact names):

• tested - the class or instance being tested (assign FIRST, call methods on it)
• result - the return value from the method under test (NEVER use output, ret, actual)
• expected - the expected value for single assertions (NEVER inline in assert)
• exp_* - expected values for multiple assertions (NEVER use expected_* prefix)

Assertions:

• Use is for singletons: assert result is True, assert result is None
• Use == for everything else: assert result == "value"

Mock setup: Always use side_effect - mock.method.side_effect = [return_value]

Mock return objects: Use SimpleNamespace NOT MagicMock for return objects (like HTTP responses). Use lambdas for methods: SimpleNamespace(status_code=200, json=lambda: {"data": "value"})

Capturing print output: Use pytest's capsys fixture - NEVER mock print. Pattern: captured = capsys.readouterr() then assert "message" in captured.out

Mock scope: Mock external systems, I/O, non-deterministic behavior, and internal class method dependencies. Use real instances for simple data objects (NamedTuples, dataclasses). For dataclasses, you may mock internal methods when testing other methods to avoid duplication.

Mock verification: Verify ALL mock objects at object level with single assertion and hard-coded values

• Main patched mocks must be verified
• Created MagicMock() objects must be verified
• .return_value instances must be verified
• CRITICAL: Parametrized tests MUST verify mocks too - they are NOT exempt!
• Pattern: assert mock_obj.mock_calls == exp_calls (single assertion)
• NEVER verify at method level: assert mock_obj.method.mock_calls == exp_calls
• NEVER check length: assert len(mock.mock_calls) == 3
• NEVER index: assert mock.mock_calls[0] == call(...)
• NEVER use variables in expected values: call(f"URL: {url_var}")
• ALWAYS use hard-coded literals: call("URL: http://example.com")

Parametrize: @pytest.mark.parametrize("arg,expected", [pytest.param(..., id="case")])

Coverage: uv run pytest tests/ --cov=. -v must show 100%

Excluded from tests: if __name__ == "__main__": blocks - do NOT write tests for these (excluded from coverage)

Critical Rules:

1. ALWAYS start with test_inheritance() as the first test when testing a class that inherits from a base class, OR test_class() using is_namedtuple() helper for NamedTuple classes, OR test_class() using is_dataclass() helper for dataclass classes
2. ALWAYS assign the class or instance to a tested variable FIRST, then call methods on tested - NEVER call methods directly on the class name
3. ALWAYS store return values in a variable named result - NEVER use output, ret, actual, response, etc.
4. ALWAYS store expected values in a variable named expected - NEVER inline expected values in assertions
5. ALWAYS use exp_* prefix for multiple expected values (e.g., exp_calls) - NEVER use expected_* prefix
6. NEVER use return_value to set what a mock returns - always use side_effect
7. ALWAYS verify ALL mock objects with mock_calls in EVERY test - missing verification is forbidden
8. Parametrized tests are NOT exempt - every mock MUST be verified with mock_calls even in parametrized tests
9. ALWAYS verify at OBJECT level (mock.mock_calls), NEVER at method level (mock.method.mock_calls)
10. ALWAYS use single assertion format: assert mock.mock_calls == exp_calls
11. NEVER check length or index: assert len(mock.mock_calls) == 3 or mock.mock_calls[0]
12. ALWAYS use hard-coded literal values in expected calls - NEVER use variables
13. NEVER use assert helpers like assert_called_with() - only use mock_calls
14. ALWAYS use is for True/False/None comparisons - never use == for singletons
15. PREFER real instances over mocks for simple data objects (NamedTuples, dataclasses, TypedDicts)
16. USE SimpleNamespace NOT MagicMock for return objects (HTTP responses, etc.) - avoids forgetting mock_calls verification
17. USE pytest's capsys fixture to capture print output - NEVER mock print
18. NEVER write tests for if __name__ == "__main__": blocks - they are excluded from coverage

Apply these guidelines consistently to create maintainable, comprehensive test suites that ensure code reliability and prevent regressions.