Skip to content
Permalink
master
Switch branches/tags
Go to file
 
 
Cannot retrieve contributors at this time

AIRR Standards Development Process

Sub-Projects

The airr-standards repository contains numerous sub-projects within it which are all highly interconnected. Our goal is a simple workflow for development that also insures stable versioning and releases for AIRR data and software associated with publications and database submissions. The current sub-projects include:

  • Minimum information about an Adaptive Immune Receptor Repertoire Sequencing Experiment (MiAIRR).
  • Specification of file formats:
    1. Alignment hit table
    2. Rearrangement annotations
    3. Repertoire metadata
  • OpenAPI schema definitions:
    1. Repertoire metadata
    2. Rearrangement annotations
    3. Ontologies definitions for AIRR data elements.
    4. AIRR Data Commons API
  • Documentation for docs.airr-community.org
  • Python package for reading/writing/validating file formats
  • R package for reading/writing/validation file formats.
  • Software WG defining standardized data sets and metrics.

Some future sub-projects include:

  • Germline Database WG defining criteria, submission protocols, and tools, with associated file formats and specifications.

Git Structure

The master branch is utilized for all development for the ongoing major version of the AIRR Standards. The versioning system, described in separate section below, is designed to allow parallel branches of development and maintenance for multiple AIRR Standards, though currently there is a single version available, MiAIRR V1. When multiple major versions of the AIRR Standards are available, then these instructions will be updated to reflect the development process for older versions.

  • Commits to master can only be performed by the core team, which currently includes Syed Ahmad Chan Bukhari, Christian Busse, Scott Christley, Jason Vander Heiden, and Uri Laserson. Even so, consider using a pull request if the changes affect important MiAIRR or specifications files that may impact others. The core team will be expanded as the number of sub-projects in airr-standards grows.
  • The priority is not to break master for other developers, this includes renaming or deleting fields, changing schema structure, or specification definitions. This type of work requires review by the core team.
  • Members not on the core team should submit all changes as a pull request. This implies creating a feature branch to hold your changes, or forking the repository. Please use a meaningful unique name for your branch; these branches will be removed once merged into master and have stabilized.
  • Our goal is for master never to be broken. Pull requests must pass checks before being merged.
  • Use rebase instead of merge for either bringing your feature branch up-to-date with the latest master or when merging pull requests into master. The goal is for the commit history to be a tree (i.e. a single linear commit history to any specific release), versus a directed acyclic graph (DAG) with multiple commit paths. Using git merge can create these multiple commit paths.
  • Issues do not have to be created for features as discussion can be attached to the pull request. However, if you wish to discuss design/implementations ideas beforehand, please feel free to create an issue.

We have a slack channel setup for GitHub notifications that are more extensive than the normal email notifications. Please contact the core team to be added.

Versioning and Release Process

Versions and associated releases are driven by the recommendations and standards produced by the AIRR Community and their associated Working Groups, with specifications and software tools implementing those recommendations and standards. The various sub-projects within airr-standards do not have independent versioning but are interconnected together through the same version numbers. This way users know exactly which software versions are associated with which standards versions. This has the challenge that normal semantic version numbering (MAJOR.MINOR.PATCH) cannot be applied to the sub-projects, however the goal is that semantic versioning will apply to AIRR Standards as a whole. The design is thus:

  • MAJOR = MiAIRR version
  • MINOR = Implementation specifications
  • PATCH = Software updates

New and updated software tools will generally increment the PATCH number, assuming that the implementation specifications have not changed. The current MiAIRR version is V1 (i.e. MAJOR = 1), and it is expected that this version will rarely change as it required formal adoption by the AIRR Community. Currently, MINOR = 0 as the implementation specifications are in pre-release form going through many changes. Once they are released (i.e. v1.1.0) then MINOR will be incremented whenever a backwards incompatible change is introduced.

The airr-standards repository uses release branches to maintain stable releases separate from ongoing development. Each release branch is for a specific MAJOR.MINOR version, with tags being assigned on that branch for every new PATCH release. When development on master is deemed ready for release, a branch is created from master called release-MAJOR.MINOR where MAJOR.MINOR.0 will be the first associated release tag. The master branch will then be development for a future MAJOR.MINOR release, for example if the current release branch is release-1.1, then the master branch will be development for a future 1.2.0 or 2.0.0 release. All work related to a prior MAJOR.MINOR release, which essentially should only be software updates and bug fixes, should be commited to that release branch. Note: a release may just tag master and not create the actual release branch until later when it is needed.

Public Releases

Public releases of the airr-standards API packages, schema, and documentation require the following steps to be performed:

  • Update the release notes contained in the NEWS files within the standards documentation (docs/standards/news.rst), python package (lang/python/NEWS.rst), and R package (lang/R/NEWS.md).
  • Update the date and version in lang/R/DESCRIPTION.
  • Run R CMD check on the R package using the options: --as-cran --run-donttest.
  • Rebuild the R package documentation for the ReadTheDocs site using the lang/R/docs/build.R script.
  • Build the R source package and test it on http://win-builder.r-project.org before submitting to CRAN.
  • Test the upload/download of the python package on https://test.pypi.org.
  • Submit the R package to CRAN and wait for approval. Repeat the R package steps as needed until approved by CRAN.
  • Tag the repository with the version number.
  • Generate the release on GitHub.
  • Verify that ReadTheDocs automatically builds the documentation site correctly.
  • Upload the python package to PyPI.

Note, the order of these steps must be more or less as described above. Some deviation is okay, but the following are essential:

  • The release should only be tagged after all release notes and other documentation have been finalized. The default documentation that users see on ReadTheDocs will be built from the newest numerical version tag and will not included changes made after the version tag.
  • The release should not be tagged until after CRAN has accepted the R package. CRAN has a tendency not to accept an initial submission and require changes.
  • The repository must be tagged prior to uploading the python package to PyPI. Versioneer is used for annotating the version number of the python package and it uses the tag list of the git repository to do so.

Field Deprecation

Deprecated schema fields should remain in the schema until at least the next MAJOR version number. Preferably, deprecated schema fields should be retained indefinitely, unless there is a clear need to remove them.

Deprecated fields which appeared in a previous official release must be labeled with the appropriate x-airr tags that denote deprecation, explain the rationale, and specify the replacement fields (if any). For example:

organism:
  description: Binomial designation of subject's species
    x-airr:
      deprecated: true
      deprecated-description: Field was renamed to species for clarity.
      deprecated-replaced-by: [species]

Development Setup

Local development and testing can be performed either directly in your machine environment, or you can use a docker container which avoids conflicts with other software. For all cases, you will want a local copy of the github repository.

git clone https://github.com/airr-community/airr-standards.git

If you will be working in a docker container, you can pull down the airr-standards image, which has all of the prerequisites installed, or you can use your own image. The airr-standards image provides a python3 environment.

docker pull airrc/airr-standards

You will want to mount your local copy of the github repository into the container. This example command puts your local copy at the /work directory. Note that a copy of the repository exists in the image from when it is built, but don't edit that copy as your changes will get lost when you exit the docker container. It is suggested that you edit files outside of the docker container to prevent permissions issues. Also, avoid doing git commands inside the docker container. Certains commands that write data like git add or git commit can change permissions and make your local copy of the repository unusable.

cd airr-standards
docker run -v $PWD:/work -it airrc/airr-standards bash

Python Library

Normal users would install the python library using pip which pulls the package from the internet. For development, you want to install from your local copy. The commands are similar whether in docker or directly in your machine environment. Starting at the top level repository directory:

cd lang/python
python setup.py install

If you are working directly in your machine environment, you may want to install in your user site packages instead of the system site packages, which can be done by adding the --user option to the install.

python setup.py install --user

You can run the python test suite from the same directory where you do the install command.

python -m unittest discover

R Library

Users can install the latest release from CRAN in the usual way via install.packages("airr"). To build the package from a local source copy first install the build dependencies:

install.packages(c("devtools", "knitr", "rmarkdown", "testthat"))

Then run the following R commands from the package root lang/R:

library(devtools)
install_deps(dep=T)
document()
build()
install()

Tests can be run from the same directory as follows:

library(devtools)
test()

Documentation

The documentation at docs.airr-community.org is built using sphinx in a python3 environment. From the top level airr-standards directory, run this command to build a local version of the website.

sphinx-build -a -E -b html docs docs/_build/html

The documentation can then be viewed in your browser by opening the file docs/_build/html/index.html.

Code Style

General Guidelines

  • Do not commit .gitignore files or IDE project files.

Python

R

  • Follow the Bioconductor style, except use snake_case for variable names and a single # for comments.
  • Use roxygen2 for inline documentation and namespace management.