Contributing Code

We’d love to accept your patches and contributions to this project. This page contains a number of instructions and guidelines that you may want to follow so your PR can get merged in a timely manner.

Setup

While for using LibreLane, we recommend any of the installation methods, for development we really recommend docs/source/installation/nix_installation/index. Nix allows you to demo changes with LibreLane and/or tools quite easily.

This guide will assume you have LibreLane installed via Nix.

When developing LibreLane, you want to run nix develop .#dev. This will allow you to run YOUR current edits to LibreLane using python3 -m librelane <args>, except it will not attempt to build LibreLane itself as part of the Nix environment.

Branching

For various reasons, it’s recommended to call working branches, even in your forks, something else other than master, main, or dev as these branch names do have some special behavior associated with them.

Note

The main branch is the stable branch for LibreLane, i.e., this branch is updated less frequently and only accepts bugfixes.

Feature contributions should be directed towards the dev branch.

Testing

Before you submit your changes, it’s prudent to perform some kind of smoke test. python3 -m librelane --smoke-test tests a simple spm design to ensure nothing has gone horribly wrong.

LibreLane also runs two sets of tests per PR, namely, a set of design tests and a set of unit tests. Unit tests are further broken down into infrastructure tests and step implementation tests.

We do require a successful run of the unit tests to merge contributions. To do so, in the nix develop .#dev environment, run:

git submodule update --init ./test/steps/all
# To run all tests:
pytest -n auto -m all
# To run just the infrastructure tests:
#   * We don't bother passing (-n auto) to this test because the time taken to
#     allocate workers exceeds the time taken to run the tests.
pytest
# To run just step implementation tests:
pytest -n auto -m step_impl_test

Designs

As stated, designs are automatically run by the CI. We really don’t recommend you run them yourself and there’s no script to really do that[1].

Designs aren’t supposed to simply just pass or fail: LibreLane compares a number of performance and area metrics as well. We don’t want to get metrics to get noticeably worse; it could indicate bad methodology changes or a regression in the relevant tools.

The CI harvests a set of said metrics and uploads it as GitHub Actions “artifact”. You can then use the metrics comparison script to compare your PR with the base commit it is targeting. You have to wait for the “Merge Metrics” job for your Pull Request’s CI to conclude, but, there are two ways to do this:

  1. (Recommended) Write a comment starting with !metrics. GitHub Actions will run the comparison and comment the result.

  2. Run it manually in the nix develop .#dev as follows:

    python3 .github/scripts/compare_pr_metrics.py <PR NUMBER> \
  --github-token <A GITHUB PERSONAL ACCESS TOKEN> \
  --metrics-cache-repo librelane/librelane-metrics \
  --repo librelane/librelane > comparison.md

    Tip

    If you have the gh CLI installed and authenticated, you may choose to authenticate by simply passing --github-token $(gh auth token).

Dealing with failures

Infrastructure unit tests must be fixed. Collaborate with maintainers if you’re not sure why something is failing.

Step implementations unit tests, as you may have surmised, are not stored in this repo (to save on clone times), and are stored in a submodule. This complicates pull requests, as you have to open two pull requests across two repos.

If the issue is simple (an error code needs to be updated or similar), you may elect to exclude the test from running by adding it to test/steps/xfails.

The same goes for design tests: if the failure is simple enough to fix, you may simply comment out the relevant design in .github/test_sets/test_sets.yml.

Language Standards

Python

Python code should be written for Python 3.10+, and be typed. i.e., we require explicit type annotations for all major API functions.

You will need to ensure that your Python code passes linting with our three chosen tools (and one optional tool):

Tool

Kind

Command

Description

black

Formatter

black .

Ensures indentation and whitespace follow a strict standard without having you lift a finger.

flake8

Linter

flake8 .

Finds a number of common programming pitfalls.

mypy

Type-Checker

mypy .

Ensures that you’re using compatible types, i.e., you are not passing a string to a function that accepts an int, or passing None to a non-optional variable, and such.

ruff (optional)

Linter

ruff check .

Our pyproject.toml uses ruff as a simple parsing checker, i.e., makes sure your code can still parse under Python 3.10 as it is entirely too easy to write code that by accident only works on later versions of Python. We presently do not use other features of ruff.

Do all arithmetic either in integers or using the Python decimal library. All (numerous) existing uses of IEEE-754 are bugs we are interested in fixing.

Tcl

Only use Tcl to interface with tools that only have a Tcl interface (or have an immature Python interface)- i.e., Yosys, OpenROAD and Magic.

1TBS-indented, four spaces, lower_snake_case for local/global variables and UPPER_SNAKE_CASE for environment variables. Unfortunately it is impossible to add any other guidelines or standards to the Tcl code considering it is Tcl code. Please exercise your best judgment.

Yosys, OpenROAD and Magic Scripts

There are some special guidelines for scripts in scripts/yosys, scripts/openroad, and scripts/magic:

  • The scripts for each tool are a self-contained ecosystem: do not source scripts from outside their directories.

    • You may duplicate functionality if you deem it necessary.

  • Do not reference the following environment variables anywhere in order to avoid causing recursion when generating issue reproducibles:

    • $PWD

    • $RUN_DIR

    • $DESIGN_DIR

Submissions

Make your changes and then submit them as a pull requests to the:

  • main branch: For bugfixes.

  • dev branch: For new features.

Consult GitHub Help for more information on using pull requests.

You need to understand what code you are changing, what the change does, and justify that change in the commit messages and PR.

All code contributions must follow the Large Language Model Contribution Policy.

The Approval Process

For a PR to be merged, there are two requirements:

  • There are two automated checks, one for linting and the other for functionality. Both must pass.

  • An LibreLane team member must inspect and approve the PR.