pydoctrace 0.3.0

Generate sequence diagrams by tracing Python code execution

pydoctrace uses pre-commit hooks and pre-commit.ci continuous integration to enforce commit messages, code formatting and linting for quality and consistency sake. See the code conventions section if you would like to contribute to the project.

pydoctrace

Generate diagrams by tracing Python code execution.

Here are the diagrams produced by 2 different implementations of the factorial function (n * n-1 * n-2 * ... * 1):

• a sequence diagram for the recursive implementation
• a component diagram for the looping implementation

from pydoctrace.doctrace import trace_to_sequence_puml @trace_to_sequence_puml def factorial_recursive(value: int) -> int: if value <= 1: return value return value * factorial_recursive(value - 1)	from pydoctrace.doctrace import trace_to_component_puml @trace_to_component_puml def factorial_reduce_multiply(value: int) -> int: if value <= 1: return value def multiply(agg: int, value: int) -> int: return agg * value return reduce(multiply, range(1, value + 1), 1)
tests.modules.factorial.factorial_recursive.puml	tests.modules.factorial.factorial_reduce_multiply.puml

The sequence diagram gives more details about the execution but can easily grow large, the component diagram is more concise and helps viewing the architecture behind the calls.

Installation

Use your favorite dependency manager to install pydoctrace:

# using pip
pip install pydoctrace

# using poetry
poetry add pydoctrace

# using pipenv
pipenv install pydoctrace

Usage

Import a decorator and apply it to the functions whom execution you want to trace:

# in my_package/my_module.py
from pydoctrace.doctrace import trace_to_component_puml, trace_to_sequence_puml

def main(parameter):
    validate(parameter)
    do_something(parameter)

@trace_to_component_puml
def validate(parameter):
    validate_is_not_none(parameter)
    validate_is_in_database(parameter)

@trace_to_sequence_puml
def do_something(parameter):
    stuff = load_stuff(parameter)
    save_work(process_job(stuff))

Executing main() will create 2 diagram files in the current working directory:

• validate-component.puml: a PlantUML component diagram showing the interactions between validate() as the start, and validate_is_not_none() and validate_is_in_database() (and all their inner function calls, if any)
• do_something-sequence.puml: a PlantUML sequence diagram showing the interactions between do_something() as the start, and load_stuff(), process_job() and save_work() (and all their inner function calls, if any)

You can then view these files either:

• using an extension for your IDE: jebbs.plantuml for Codium / vsCode, 7017-plantuml-integration for PyCharm
• online at www.plantuml.com/plantuml/uml/

Customize the filename

The export_file_path_tpl attribute of the decorators allows you to define another file path and name:

# in my_package/my_module.py
from pydoctrace.doctrace import trace_to_component_puml, trace_to_sequence_puml

# will create a 'validate-component.puml' file in the 'doc/my_package-my_module' folder
# (pydoctrace will create the subdirectories if necessary)
@trace_to_component_puml(
    export_file_path_tpl='doc/my_package-my_module/validate-component.puml'
)
def validate(parameter):
    ...

# will create a 'my_package.my_module/do_something-2024-02-27_19.16.09_473-sequence.puml' file
@trace_to_sequence_puml(
    export_file_path_tpl='${function_module}/${function_name}-${datetime_millis}-sequence.puml'
)
def do_something(parameter):
    ...

As you can see, you can use the following template tags in the file name:

• ${function_module}: which resolves to do_something.__module__, 'my_package.my_module' in the example
• ${function_name}: which resolves to do_something.__name__, 'do_something' in the example
• ${datetime_millis}: the datetime in ISO-ish format compatible with filenames (Windows does not support ':' in filenames). If the traced function is called several times during the execution, including ${datetime_millis} in the filename template will generate different files that won't overwrite themselves

Filter what is traced

To keep the generated diagrams useful and legible, you probably want to exclude some calls from the tracing (such as calls to print(...) or json.load(...)). pydoctrace comes with ready-to-use presets that you can combine to filter out some calls you do not want to document:

• EXCLUDE_BUILTINS_PRESET: prevents from tracing calls to built-in functions (print, len, all, any, etc.)
• EXCLUDE_STDLIB_PRESET: prevents from tracing calls to all the standard library modules (csv, json, dataclass, etc.)
• EXCLUDE_TESTS_PRESET: prevents from tracing calls to functions belonging to the tests, _pytest, pytest, unittest, doctest modules. This one is particularly useful when you want to generate some documentation from an automated tests
• EXCLUDE_DEPTH_BELOW_5_PRESET: prevents from tracing any calls that involve a call stack of more than five levels after the traced function

By default, EXCLUDE_STDLIB_PRESET and EXCLUDE_TESTS_PRESET are activated. Specify something else at the decorator level to change the call filtering.

from pydoctrace.doctrace import trace_to_component_puml
from pydoctrace.callfilter.presets import (
    EXCLUDE_BUILTINS_PRESET,
    EXCLUDE_CALL_DEPTH_PRESET_FACTORY,
    EXCLUDE_DEPTH_BELOW_5_PRESET,
    EXCLUDE_STDLIB_PRESET,
    EXCLUDE_TESTS_PRESET,
    TRACE_ALL_PRESET,
)

# an empty presets list traces everything
@trace_to_component_puml(filter_presets=[])
def validate(parameter):
    ...
# or you can be explicit
from pydoctrace.callfilter.presets import TRACE_ALL_PRESET
@trace_to_component_puml(filter_presets=[TRACE_ALL_PRESET])
def validate(parameter):
    ...

# filter at a custom callstack level
from pydoctrace.callfilter.presets import EXCLUDE_CALL_DEPTH_PRESET_FACTORY
ABOVE_3_PRESET = EXCLUDE_CALL_DEPTH_PRESET_FACTORY(3)
@trace_to_component_puml(filter_presets=[ABOVE_3_PRESET])
def call():
    call_1()

def call_1():
    call_2()

def call_2():
    call_3()

def call_3(): # will not be traced in the diagram
    ...

# documents the behavior of do_something without the noise of
# the test framework nor the standard libray
from my_package.my_module import do_something
from pydoctrace.callfilter.presets import EXCLUDE_STDLIB_PRESET, EXCLUDE_TESTS_PRESET
@trace_to_component_puml(filter_presets=[EXCLUDE_STDLIB_PRESET, EXCLUDE_TESTS_PRESET])
def test_do_something():
    assert do_something('param') is not None

# for test-generated documentation, you can also use the decorator programmatically
def test_do_something():
    traceable_do_something = trace_to_component_puml(
        export_file_path_tpl='doc/${function_module}/${function_name}-component.puml',
        filter_presets=[EXCLUDE_STDLIB_PRESET, EXCLUDE_TESTS_PRESET, ABOVE_3_PRESET]
    )(do_something)

    assert traceable_do_something('param') is not None

Purposes and mechanisms

The purpose of pydoctrace is to document the execution of some code to illustrate the behavior and the structure of the code base.

• use one of the provided decorators (diagram type and format) to decorate the function whose execution you want to document
• run your code (with a unit test, for instance): when the execution hits the decorated function, the execution is traced and a diagram is drawn to show how the functions are called, how the values are returned and how the errors are handled

pydoctrace is a pure Python tool relying on no other 3rd-party library to work. The project uses only development libraries for testing and documentation purposes.

Doc-tracing

This approach, which I called "doc-tracing" (tracing code execution for documentation purposes), is complementary of other approaches which generate documentation from static code analysis. Static code analysis reads the source code to detect and document data structures (classes, dataclasses, named tuples, enumeration, etc.), functions names and signatures (parameters name and types, returned value types).

Useful as it is, static code analysis does not help much to understand how the code pieces work together; doc-tracing attempts to complement this approach by producing documentation while the code runs. Some use cases:

• you start working on a legacy codebase: study the behavior of a particular function by temporarily tracing its executions
• you finished a user story, document its implementation by tracing the execution of some integration tests ("given ... when ... then ...") to illustrate how right cases and errors are handled.
• generally, the sequence diagrams illustrate how the modules and functions interacts; and as such, they help discussing architecture
• tracing code execution can also be useful when teaching computer programming to illustrate how algorithms work

How is the code execution traced?

When a function decorated by pydoctrace is called:

1. a context manager is created
2. during which a tracing function is passed to sys.settrace, which is called when different events happen in the execution stack:
   • when functions are called
   • when values are returned
   • when exceptions are raised or handled
3. the sequence diagram is drawn and exported in a stream fashion (when possible) so that its memory footprint is minimal
4. once the decorated function stops its execution, the tracing function is removed from the code execution

⚠️ Caveat: pydoctrace uses the sys.settrace API, which is meant to be used by debuggers. Therefore, a warning is emitted when pydoctrace is used in a debug mode (and does not trace the decorated function anymore).

Tests

# directly with poetry
poetry run pytest -v

# in an activated virtual environment
pytest -v

Code coverage (with missed branch statements):

poetry run pytest -v --cov=pydoctrace --cov-branch --cov-report term-missing --cov-fail-under 82

Changelog

• 0.3.0: added configurable filters to exclude function calls from the tracing process (do not trace calls to the standard library modules and functions by default), templating for output file names. Thank you Philippe Tony for suggesting filtering out calls above a certain calls depth from the traced function. Started integration tests
• 0.2.0: PlantUML exporter for component diagrams, added unit tests
• 0.1.2: added Github actions for the automated tests and interaction with pre-commit.ci for the code linting
• 0.1.1: [deleted release: wrong contents] added Github actions for the automated tests and interaction with pre-commit.ci for the code linting
• 0.1.0: ✨ first release, PlantUML exporter for sequence diagrams; diagram files are saved in the current working directory

Licence

Unless stated otherwise, all works are licensed under the MIT license, a copy of which is included here.

Contributions

• Luc Sorel-Giffo

Pull requests

Pull-requests are welcome and will be processed on a best-effort basis.

Pull requests must follow the guidelines enforced by the pre-commit hooks:

• commit messages must follow the Angular conventions enforced by the commitlint hook
• code formatting must follow the conventions enforced by the isort and ruff-formatter hooks
• code linting should not detect code smells in your contributions, this is checked by the ruff hook

When requesting a feature, please consider involving yourself in the review process once it is developed.

Code conventions

The code conventions are described and enforced by pre-commit hooks to maintain style and quality consistency across the code base. The hooks are declared in the .pre-commit-config.yaml file.

When you contribute, set the git hooks (pre-commit and commit-msg types) on your development environment:

poetry run pre-commit install

Before committing, you can check your changes manually with:

# put all your changes in the git staging area (or add the changes manually and skip this)
git add -A

# run all hooks
poetry run pre-commit run --all-files

# run a specific hook
poetry run pre-commit run ruff --all-files

Commit messages

Please, follow the conventions of the Angular team for commit messages. When merging your pull-request, the new version of the project will be derived from the messages.

I use the redjue.git-commit-plugin codium/vsCode extension to help me write commit messages.

Code formatting

This project uses isort and ruff to format the code. The guidelines are expressed in their respective sections in the pyproject.toml file.

Best practices

This project uses the ruff linter, which is configured in its section in the pyproject.toml file.

Similar tools and bibliography

• https://stackoverflow.com/questions/45238329/it-is-possible-to-generate-sequence-diagram-from-python-code
• https://github.com/gaogaotiantian/viztracer
• https://9to5answer.com/how-to-generate-a-sequence-diagram-from-java-source-code
• https://medium.com/javarevisited/how-to-generate-sequence-diagrams-in-intellij-e2bb7cec2b0b