Development Procedures and Guidelines

We encourage anyone to help us develop the PypeIt code base to better suit your needs and to improve its algorithms. If you do so, please follow this list of procedures and guidelines. In particular, please note our Code of Conduct.

Installation

If you plan to develop/alter the PypeIt code directly, you should install the software via git, instead of pip or conda. See Developer Installation.

This isn’t required, but to simplify some of the commands below, I assume that you have an environmental variable that points to your installation of the PypeIt repository. E.g., add something like the following to your ~/.zshrc or ~/.bash_profile:

export PYPEIT_DIR=$HOME/Work/packages/PypeIt

Branches

PypeIt maintains two persistent branches:

release: This is the primary stable version of the code. Modulo any recent hotfixes, this is the version of the code that is installed when using pip and its the version that is used to produce the latest documentation. Pull requests to this branch are only done before tagging a new release of the code or to perform critical bug hotfixes. The release schedule is irregular and decided by the core developers.

develop: This is the main development version of the code. It should be stable enough to use, but it may contain experimental, unsupported code that is work in progress.

When editing the code, please create a new branch stemming from the develop branch. You should also pull and merge in the most recent version of the release branch to make sure your new branch includes any very recent hot-fixes. On the command line, you can do this as follows:

cd $PYPEIT_DIR
git checkout release
git pull
git checkout develop
git pull
git checkout -b my_new_feature
git merge --no-ff release

Note

In terms of the merge with the release branch, beware that you may need to start a new section of the CHANGES.rst file to reflect a jump in the version number. This should only be necessary if your branch is the first one after a new tag is released.

Development Principles and Communication

The main thing to keep in mind when developing for PypeIt is that its primary use is as an end-to-end reduction pipeline. This has a few implications:

By default, the execution of run_pypeit should continue either until a critical error is raised or the reduction is complete. No direct interaction with the code should be required at any point. PypeIt does have some interactive components, but these are executed via separate scripts.

Any input needed from the user for your feature should be provided by User-level Parameters (preferred) or as a command-line argument.

When developing and debugging, you may need to interact with the code using pdb or IPython.embed; however, these instances should be removed before performing a pull request.

The success or failure of any given procedure must be assessed via automatically generated quality-assessment figures (preferred) or via scripts that interact with the primary output files.

See here for guidance on adding a new spectrograph to the list of spectrograph data that PypeIt can reduce.

If your development includes adding a new executable script, see advice at Developing New PypeIt Scripts.

Feature development in PypeIt is unlikely to be fully independent of other development activities. Your feature will likely depend on or influence the outcome of other modules during the data-reduction process. This leads to a few important guidelines:

Make sure that your branch is always up-to-date with the develop and release branches. E.g.:
cd $PYPEIT_DIR
git checkout release
git pull
git checkout develop
git pull
git checkout my_new_feature
git merge --no-ff release
git merge --no-ff develop
Consider the effects of simultaneous development efforts on your work and vice versa. For example, if you’re working on a specific module of the code that depends on the result/datamodel of the wavelength-calibration module, you should communicate this and find out if someone else is developing that module and how/if they’re changing it. Depending on the scale of those changes, development priorities may need to be worked out to minimize merge conflicts and the need to immediately rework/refactor new code.

When you’re ready to, you can submit a PR at any time, but the core development team will need to discuss the merge order to ensure a smooth process.

Our primary means of communication for development is the PypeIt developers Slack and a biweekly telecon. Contact X Prochaska for Slack access and/or the relevant Zoom link.

Testing the Code

PypeIt performs extensive testing using the PypeIt Development Suite; follow that link for more details on executing the tests. What follows describes how to add new tests.

Development Suite

To add new tests to the development suite

Add the new data to shared Google Drive under RAW_DATA. The tests are organized into setup directories under a directory named for the instrument.

Add a new PypeIt Reduction File specific to this data to the PypeIt Development Suite repo under pypeit_files. The file name must be lower case and named after the instrument and setup, for example: keck_deimos_1200g_m_7750.pypeit.

If desired, add any files for pypeit_sensfunc, pypeit_flux_calib, pypeit_coadd_1dspec, pypeit_coadd_2dspec to the PypeIt Development Suite repo under sensfunc_files, fluxing_files, coadd1d_files, coadd2d_files, respectively.

Edit test_setups.py in the PypeIt Development Suite under test_scripts. Follow the instructions at the top of that file.

Run the full development test suite to completion. Once all tests pass, the test_priority_file will be updated with the new test. This file tells the test scripts what order to run the tests in for optimum CPU utilization. Commit test_priority_list and any other files added to the dev-suite repository and submit a pull request.

Unit Tests

Unit tests are located in the $PYPEIT_DIR/pypeit/tests directory. To run them, make sure you have pytest installed and then:

cd $PYPEIT_DIR
pytest

If some tests fail, you can run an individual test, e.g. test_wvcalib.py with

cd $PYPEIT_DIR
pytest -s pypeit/tests/test_wvcalib.py

Note that the “-s” option allows you to insert interactive debugging commands into the test, here test_wvcalib.py, to help determine why the test is failing.

Warning

Running these tests generates some files that should be ignored. Please do not add these test files to the repository. We try to include clean-up as part of the tests, but these are not always caught.

Note also that the use of pytest requires the test dependencies to be installed, e.g. via pip install -e .[test]. It is also possible, and often preferable, to run tests within their own isolated environments using tox. This provides the capability to easily run tests against different versions of the various dependencies, including different python versions. The available tox environments are defined in $PYPEIT_DIR/tox.ini and can be listed by running tox -a. To run tests against the default dependencies using the default python, do:

cd $PYPEIT_DIR
tox -e test

To specify a python version, do something like:

cd $PYPEIT_DIR
tox -e py38-test

To test against, for example, the main branch for astropy on GitHub, you can do:

cd $PYPEIT_DIR
tox -e py38-test-astropydev

Similar dev dependencies are configured for numpy, ginga, and linetools, as well.

Unit tests included in the main PypeIt repo should not require large data files. Some files are kept in the repo for this purpose (see the pypeit/tests/files directory), but they should be minimized to keep the size of the package distribution manageable. Unit tests that require input data files should instead be added to the PypeIt Development Suite.

Workflow

A typical PypeIt development workflow is as follows:

Create a new branch stemming from the develop branch (hot-fixes should instead branch from release):
cd $PYPEIT_DIR
git checkout release
git pull
git checkout develop
git pull
git checkout -b my_new_feature
git merge --no-ff release
Develop and debug the feature
Run the unit tests, fix any failures, add new tests that test your new feature(s), and/or modify the tests to accommodate your new feature:
cd $PYPEIT_DIR
pytest
or preferably:
cd $PYPEIT_DIR
tox -e test
Run the Development Suite and fix any failures:
cd $PYPEIT_DEV
./pypeit_test develop
Warning

The Development Suite is extensive and takes significant computing resources and time. The PypeIt development team consistently executes these tests using cloud computing. We recommend you ensure that your pypeit branch successfully runs on either a specific instrument of interest or shane_kast_blue first, and then someone on the PypeIt development team can execute the tests in the cloud. From the top-level directory of the Development Suite, you can run all tests for shane_kast_blue as follows:
./pypeit_test all -i shane_kast_blue
Edit $PYPEIT_DIR/CHANGES.rst to reflect your key developments and update the documentation. You can compile the docs using the update_docs script (see below), which is just a simple convenience script for executing make clean ; make html in the doc directory.
cd $PYPEIT_DIR
./update_docs
Any warnings in the sphinx build of the docs must be fixed. If you’re having difficulty getting the right sphinx/rst incantation, ping the documentation channel in the PypeIt Developers Slack.
Make sure all your edits are committed and pushed to the remote repository:
cd $PYPEIT_DIR
git add -u
git commit -m 'final prep for PR'
git push
Submit a Pull Request (PR). Unless otherwise requested, all PRs should be submitted to the develop branch.

Note

The addition of new commits causes setuptools_scm to automatically increment the version based on the last tag that was pushed. This will be of the form {next_version}.dev{distance}+{scm letter}{revision hash}. See the setuptools_scm documentation for more details.

Pull Request Acceptance Requirements

Once you’ve submitted a pull request, we’ll review your PR and provide comments on the code. The minimum requirements for acceptance of a PR are as follows:

If your PR introduces a new instrument (see New Spectrograph) that PypeIt is to support for the long term, this instrument must be added to the Development Suite. That means raw data should be added to the Google Drive (see here) and relevant tests should be added to the $PYPEIT_DEV/pypeit_test script (via a PR to the PypeIt Development Suite) such that the new instrument is included in list of instruments tested by executing ./pypeit_test develop.

The CI tests run by GitHub (see the Checks tab of the PR) on the remote repository must pass.

You (or someone running the tests on your behalf) must post a successful report resulting from your execution of the Development Suite, which should look like this:

Example posting of successful tests.

For hotfixes, these tests can be circumvented at the discretion of the core developers in the cases where the hotfix is obviously correct.

All new methods and classes must be at least minimally documented. “Minimally documented” means that each method has a docstring that gives at least:

a one sentence description of the purpose of the method,

a complete list of the required and optional arguments and their meaning,

a description of the returned objects, if there are any.

Documentation is expected to adhere to Sphinx syntax; i.e., the docstrings should be reStructuredText. We accept both Google-format docstrings and Numpy-format docstrings, but the former is preferred.

The docstrings for any changes to existing methods that were altered must have been modified so that they are up-to-date and accurate.

Spurious commented code used for debugging or testing is fine, but please let us know if you want it to be kept by adding a relevant comment, something like # TODO: Keep this around for now, at the beginning of the commented block. Otherwise, we’re likely to remove the commented code when we come across it.

“Unsupported code,” that is code that is experimental and still work in progress, should be minimized as much as is reasonable. The relevant code block should be clearly marked as experimental or WIP, and it should not be executed by the main PypeIt executable, run_pypeit.

At least two reviewers must accept the code.

Tagging Protocol

The core development team will regularly tag “release” versions of the repository. Tagging a release version of the code is triggered anytime the development branch of the code is merged into the release branch. The tagging process is as follows:

At biweekly PypeIt telecons or over the PypeIt developers Slack, the core development team will decide to merge the develop branch into release.

A branch is created off of develop (typically called staged) and then a PR is issued to merge staged into release. This release...staged PR must meet the same Pull Request Acceptance Requirements when merging new branches into develop. Code review is expected to be limited (because all code changes will have been reviewed before pulling into develop), but the result of the dev-suite tests must be shown and approved. The reason for creating the new branch instead of a direct release...develop PR is to allow for the following updates to staged before merging (develop is a protected branch and cannot be directly edited):

Update the documentation by executing cd doc ; make clean ; make html, add any updated files, and correct any issued errors/warnings.

Fix any test failures. As necessary, an accompanying PypeIt Development Suite PR may be issued that includes test fixes required code changes. If no code changes are required, a PypeIt Development Suite PR should be issued that merges its develop branch directly into its main branch.

Make any final updates to CHANGES.rst and reset the relevant version header to be the intended tag number.

Add a new release doc to the doc/releases directory, parsing changes from the CHANGES.rst file, and add include it in the whatsnew.rst doc.
Once the release branch and the PypeIt Development Suite main branch are updated, the dev-suite tests are re-run using these two branches. These tests must pass before tagging. Once they pass, the code is tagged as follows:
# Create a tag of the form X.Y.Z (using 1.14.0 here as an example).
# The current autogenerated version is found in pypeit/version.py.
cd $PYPEIT_DIR
git checkout release
git pull
git tag 1.14.0

# Push the new tag
git push --tags
The tag is released for pip installation.
# Make sure you have the most recent version of twine installed
pip install twine --upgrade
# Construct the pip distribution
python setup.py sdist bdist_wheel
# Test the upload
twine upload --repository pypeit-test dist/*
# Upload, this time it's for keeps
twine upload --repository pypeit dist/*
For the uploading, you need a ~/.pypirc file that looks like this:
[distutils]
index-servers =
    pypeit
    pypeit-test

[pypeit]
repository: https://upload.pypi.org/legacy/
username = pypeit
password = [ask for this]

[pypeit-test]
repository: https://test.pypi.org/legacy/
username = pypeit
password = [ask for this]
After a new version is uploaded to pip, a new PR is automatically generated for the conda-forge/pypeit-feedstock repository. Follow the commit checklist there before merging that PR, which will trigger the release of a new pypeit package on conda-forge. For more information on how to manually update the conda-forge pypeit package, see How to manually update the PypeIt conda-forge feedstock (what to do if the bots are broken).

DOI

If we wish to generate a new DOI for the code, it is as simple as

Generate a new release on GitHub.

Update the DOI in the README.rst

This document was developed and mutually agreed upon by: Kyle Westfall, J. Xavier Prochaska, Joseph Hennawi.

Last Modified: 07 Sep 2023

Additional Developer Links

Here are some developer-specific docs: