Contributing Guide

Contributions are welcome, and they are greatly appreciated!

Guide to Submitting Changes

Before you start making changes, please open either a bug report or a feature request. That way, I can tell you if what you are about to do is likely to be accepted before you invest your time in it.

This guide assumes that you have at least a basic knowledge of git, working with command line tools and that you have a GitHub account. If that is not the case then attempting these changes yourself is generally not a good idea. Please do not try to make changes, no matter how trivial, via GitHub’s web UI - it looks tantilisingly simple but invariably ends in carnage…

If you get stuck anywhere throughout the following then please either push your current changes, submit a pull request but mark it as a draft or just ask for help! As a rough rule of thumb, if you’ve gotten nowhere in the last 15 minutes, it’s time to ask for help.

Setting up

The following steps will set you up with an editable installation of hirola and its test dependencies.

  1. Either create and activate a virtual environment or run:

    pip uninstall hirola

    so that your original installation is safely out of the way.

  2. Fork this repo.

  3. Clone and open a terminal inside your local clone.

    git clone https://github.com/your-username/Hirola.git
cd Hirola

  4. Install locally by running:

    pip install -e .[test]

    Note, if you normally use pip3 instead of pip then do so here. The above command tells pip to install your local clone (the .) in editable mode (the -e flag) and additionally install Hirola’s test requirements (the [test]). The first time you run this may take a while if your internet connection is poor.

  5. Compile the C code. (Similarly replace python with python3 if that’s what you normally do.)

    python setup.py -q build

    If you don’t have gcc installed then this will issue a cslug.exceptions.NoGccError() with instructions on how to install gcc. Follow those then retry the above command.

  6. Do a run of the test suite so that you know your starting point works.

    pytest

    All non-skipped tests should pass and you should get a nice green message saying Required test coverage of 100% reached. If that is not the case then please report it.

  7. Create and switch to a new branch to make your changes in.

    git switch -c add-foo-feature

You now have a basic functional development environment. Go ahead and hack away!

Codebase Roadmap

The code base is hopefully small enough that you don’t need a map but here’s a brief one irregardless.

  • The hirola folder is what you ultimately get when you pip install hirola. This is where you will add your features or fix your bugs.

  • The documentation source code lives in docs/source but it also pulls in the contents of the top level README.rst from the root of this repository. Additionally the stub files located in docs/source/reference import the docstrings from hirola’s classes and functions to form the API reference.

  • The tests live in the tests directory.

  • Continuous integration jobs are defined in .github/workflows.

Pretty much everything outside of those directories is just project boilerplate and configuration.

Inside the hirola directory, the main files are:

  • hash_table.h which defines the HashTable() class (or rather structure) used throughout the C code to hold the various components of a hash table. Note that this HashTable() structure is not the hirola.HashTable() class you typically interact with. It’s just a handy C bucket which ultimately becomes the _raw attribute of hirola.HashTable().

  • hash_table.c is where all the work happens. Any functions in there whose names start with HT_ are intended to be HashTable methods but be aware that this is purely just a convention.

  • _hash_table.py provides a user friendly wrapper class (hirola.HashTable()) around the C functions, adding type checking, input normalisation, output checking and whatever else is needed to keep C happy.

The slug = CSlug(...) line near the top of _hash_table.py is the glue between the C and Python code. Any function defined in either of the C source files becomes an attribute of slug.dll (e.g. void foo() becomes slug.dll.foo() in Python). For the most part, types are automatic but see cslug’s arrays and NumPy arrays tutorials for passing arrays in and out of the C code.

Commands Cheatsheet

A list of terminal commands you are likely to need for day to day development. Unless specified otherwise, these should all be ran from the repository’s root.

Compile or recompile the C code

Whenever you modify C code, you’ll have to trigger a recompile for the changes to take effect.

  • From terminal (requires restarting any open Python consoles): python setup.py -q build

  • From Python (no need to restart anything): from hirola._hash_table import slug; slug.make()

  • With hash collision metrics enabled (only needed if you’re writing hash functions):

    • Unix: CC_FLAGS='-D COUNT_COLLISIONS' python setup.py -q build

    • Windows: set CC_FLAGS="-D COUNT_COLLISIONS" && python setup.py -q build

  • With the clang compiler instead of gcc (requires installing clang):

    • Unix: CC=clang python setup.py -q build

    • Windows: set CC=clang && python setup.py -q build

  • Clean (remove all generated files): git clean -Xdf hirola/

Test

To run the test suite, use pytest:

  • Run everything: pytest

  • Run everything including the tests normally skipped: Recompile with COUNT_COLLISIONS enabled (see above) then run pytest as usual.

  • Run everything but stop on the first failure: pytest -x

  • Run a single test file (ignore the FAIL Required test coverage of 100% not reached error it issues): pytest tests/test_hash_table.py

  • Run a single test function by name: pytest -k test_automatic_resize

  • Run a single test function by its full path: pytest tests/test_hash_table.py::test_automatic_resize

New tests can be added by defining functions whose names starts with test_ in python files whose name also starts with test_ inside the tests folder. Tests should be ordered so that low level tests happen before high level tests so that the first test to fail (as given by pytest -x) indicates exactly where the break is rather than indicating that a more complex function is broken as a side effect of the lower level function’s being broken. The per-file running order is determined by pytestmark = pytest.mark.order() calls at the top of each file and within each file, tests are ordered simply by their line numbers.

Run coverage

Coverage tells us which lines of code were never ran when running the test suite. The test suite automatically collects coverage statistics.

  • Do a full run of the test suite: pytest

  • Generate an HTML report: coverage html

  • View said report:

    • Linux: xdg-open htmlcov/index.html

    • macOS or FreeBSD: open htmlcov/index.html

    • Windows: start htmlcov/index.html

Run automatic code formatter

  • Install with: pip install -r requirements-dev.txt

  • Run on all Python files: yapf -rip .

  • Run on one file: yapf -i hirola/_hash_table.py

Build documentation

All documentation commands should be ran inside the docs folder.

  • Install docs requirements: pip install -r requirements.txt

  • Build: make html

  • View:

    • Linux: xdg-open build/html/index.html

    • macOS or FreeBSD: open build/html/index.html

    • Windows: start build/html/index.html

Trigger continuous integration

Continuous integration allows us to quickly test all platforms and Python versions. First push your changes to GitHub then either:

  • Trigger from the web UI:

    1. Go to your fork’s GitHub page.

    2. Select the Actions tab.

    3. Say yes if it prompts you to enable actions.

    4. On the left hand side, select Test.

    5. Press Run Workflow.

    6. Select the branch you are working on from the drop down menu.

    7. Press the green Run Workflow button.

    8. Wait a few seconds then refresh the page. Your new job should appear below.

  • Trigger using GitHub’s CLI:

    gh workflow run --ref=your-branch-name test.yml

    Then see it running by checking the Actions tab on your fork.

Benchmarking

There is a really crude benchmark script which compares the speed of hirola against Python’s set(). The number it emits is how many times faster hirola is (i.e. big number is better). For historical reasons, it is invoked via:

python tests/benchmarks.py benchmark

I am in the process of replacing this script.

Please note that the binaries on PyPI are compiled with clang instead of gcc. Clang produces binaries which are about 20% faster so unless you also compile with clang, it is not meaningful to compare to a hirola downloaded from PyPI.

Before Submission

Before you submit a pull request, here is a checklist of things that I am likely to moan about if your changes don’t meet the criteria below.

  1. Make sure that there are no nonfunctional or stylistic changes to existing code. I don’t give a wet-slap about PEP8 - I care very much about the signal to noise ratio of git log -S "new code", git log -- filename.py and git diff as well as the ability to merge, cherry-pick and rebase without merge conflicts.

  2. The test suite passes with 100% coverage (see Test and Run coverage). If your adding code then this means that you will also have to add tests to keep coverage happy.

  3. If adding functionality, the docs need to be updated. Add Google style docstrings to new classes or functions, ensure that they appear somewhere in the API reference section of the docs and, unless the feature is quite niche, find a good place to add it to the README.rst.

  4. There is a clear distinction between public and private API. Anything that is intended to be used by end users should be documented. Any functions that aren’t intended for use should have an underscore prefixed name or be defined in an underscore prefixed submodule to serve as a signal both to users and IDE code completions not to use them. Hidden, undocumented or barely documented functionality leads to nightmares where users don’t know what they can safely use without fear of their code breaking after upgrading hirola and hirola developers can’t change anything for fear of breaking someone else’s downstream project.

  5. Python source code should be formatted by yapf (see Run automatic code formatter).

  6. Proper grammar is used for anything textual. This means capital letters, full stops and no skipping the joining words - this goes for comments, documentation, docstrings and commit messages.

That’s you, go ahead and submit…

If you wish, append By your name / username / email / URL / some other piece of information you wish to be identified by to a commit message and I will add it to the credits section of the README.