Development workflow#
A substantial change to the public API is described in an issue before a pull request is opened; small fixes and documentation corrections go directly to a pull request. Work on a topic branch of a fork, one coherent change per branch.
Change procedure#
Read
AGENTS.md, and the implementation and tests of the code being changed.Change the code and the documentation of the public behaviour it affects. Generated files are regenerated, not edited.
Run the checks:
Change
Command
Python
./scripts/lint.sh: ruff’s format and lint checks and codespell;--fixapplies ruff’s formatting and fixesAny code
./scripts/run_tests.sh, or a subset of the suite during developmentC, CUDA or the ABI
The full suite;
python scripts/check_device.pyon a machine with a CUDA deviceDocumentation
./scripts/build_docs.sh(warnings are errors), and./scripts/build_docs.sh --executefor changed examplesPDF manual
./scripts/build_docs_pdf.sh, afterpython -m playwright install chromiumCommit, push to the fork, and open a pull request (Pull requests).
A topic branch takes upstream changes by rebasing, which rewrites its history:
git fetch upstream
git rebase upstream/main
git submodule update --init --recursive
Pull requests#
A pull request is opened against pulserver/bartorch:main from a topic branch,
as a draft while work remains. Its description follows the repository’s
template:
Section |
Contents |
|---|---|
What changed |
A concise summary |
Why |
The problem, the issue it addresses and, for a new method, the paper |
Validation |
The commands run and their results; for a numerical change, the reference and the tolerance; hardware or optional dependencies that were not available |
Public API and documentation |
Changes to public behaviour and the documentation updated for them |
CI runs on every pull request:
Workflow |
Checks |
|---|---|
Tests |
The suite on Linux with GCC 14, and on macOS and Windows with clang, each on Python 3.10 and 3.14; a CUDA build, tested on the host because the runner has no device |
Lint, Typos |
|
Docs |
|
Publish to PyPI |
Every wheel built, repaired and tested as a release would be, when the pull request changes the build ( |
A pull request is merged when CI passes and a maintainer has approved it.
Generated files#
File |
Generator |
Test holding it |
|---|---|---|
|
|
|
|
|
|
Logo, mark, architecture and encoding figures in |
|
None |
Plots in |
|
None |
run_tests.sh regenerates both Python files before the suite. The logo’s
sources and their terms are listed in License and third-party notices. The
documentation’s build products, docs/generated/, docs/api_objects.rst,
docs/auto_examples/ and the PDF manual, are not tracked;
./scripts/build_docs.sh --clean removes them.
Published documentation#
The docs workflow publishes into the gh-pages branch, one directory per
version: latest from main, and vX.Y.Z from a release tag, which
scripts/publish_docs.py also copies to stable when it is the newest
release. The site root redirects to stable, or to latest before the first
release, and the version switcher reads versions.json beside the
directories.
BART submodule updates#
external/bart pins a revision of the downstream fork pulserver/bart
(BART fork). Moving it is a change of its own: sync the fork, update the
submodule pointer, run python scripts/gen_catalogue.py, and review the
resulting API differences. tests/test_tools.py requires every BART command
to be wrapped by hand, generated into bartorch.tools, or listed as
private with a reason in src/bartorch/_coverage.py.
FINUFFT submodule updates#
external/finufft is upstream FINUFFT at one commit, unmodified. Moving it is
a change of its own: choose a commit whose upstream CI passed, update the
submodule pointer, record the commit and the reason in the pull request, and
run the suite on every platform. tests/test_finufft.py holds the transform
against explicit discrete Fourier sums and against BART’s gridder. A change
FINUFFT needs goes upstream, not into the checkout.
Tests#
A numerical test compares BART with a reference outside BART: an independent FFT or DFT, an explicit sum, a closed-form signal, a finite difference or an adjoint identity. Comparing two entry points into the same BART implementation establishes nothing about its correctness. Tests are pytest functions whose names state the property tested.
Code of conduct#
Our pledge#
We pledge to make participation in this project a harassment-free experience for everyone, regardless of age, body size, visible or invisible disability, ethnicity, sex characteristics, gender identity and expression, level of experience, education, socio-economic status, nationality, personal appearance, race, religion, or sexual identity and orientation.
Our standards#
Behaviour that helps:
showing empathy and kindness toward other people
respecting differing opinions, viewpoints, and experiences
giving and gracefully accepting constructive feedback
accepting responsibility, apologising to those affected by our mistakes, and learning from the experience
Behaviour that is not acceptable:
sexualised language or imagery, and sexual attention or advances of any kind
trolling, insulting or derogatory comments, and personal or political attacks
public or private harassment
publishing others’ private information, such as a physical or email address, without their explicit permission
Enforcement#
Report abusive, harassing, or otherwise unacceptable behaviour to matteo.cencini@gmail.com. All complaints will be reviewed and investigated promptly and fairly. The maintainers will respect the privacy and security of the reporter of any incident.
Maintainers who do not follow or enforce this code in good faith may face temporary or permanent repercussions as determined by other maintainers.
Attribution#
Adapted from the Contributor Covenant, version 2.1.