2. Installation and Setup

The Belle II Software is meant to work an any recent 64 bit Linux system but it is only tested and provided in binary form for a few select distributions

  • Scientific Linux 6 or CentOS 6 (sl6)

  • Enterprise Linux 7 or CentOS 7 (el7)

  • Ubuntu 16.04 (ubuntu1604)

  • Ubuntu 18.04 (ubuntu1804)

If you run on one of these distributions the most convenient way to obtain the Belle II Software is to use it via CVMFS which is readily available on KEKCC and many HEP specific software resources. It can also easily be installed on your local machine following the CVMFS Client Quick Start guide.


In the following it is assumed that you have configured your access to the code repository

If you want to install the Belle II Software without CVMFS please have a look at the following documents, otherwise if you have CVMFS available please continue with the Setup of the Belle II Software

Alternatively you can use Jan’s docker container to get the Belle II software, for example on Mac or Windows machines.

2.3. Setup of the Belle II Software

The command

$ source /cvmfs/belle.cern.ch/tools/b2setup release-XX-YY-ZZ

sets up the Belle II Software Tools and the Belle II software version release-XX-YY-ZZ.


If you use the software without CVMFS please replace /cvmfs/belle.cern.ch/tools by the path where you installed the Belle II Software tools.


There are some extra environment variables which can be set to customize the setup

After that all the Command Line Tools will be setup correctly and ready to use.

When the tools are already set up one can set up a Belle II software version directly with

$ b2setup release-XX-YY-ZZ

To only set up the tools without a release use

$ source /cvmfs/belle.cern.ch/tools/b2setup


The setup of tools and releases has to be done in every shell you plan on using the Belle II Software.

2.3.1. Choosing a release

You can see available releases of basf2:

$ b2setup --help

after the tools have been set up.

Some releases are old, but left available for legacy reasons. If you try to setup an old/unsupported release, you will get a message advising you to move to a newer (supported) release. You can check if a given release is supported with:

$ b2help-releases your_release_version

Releases come in four kinds:

  1. A full release, name of the form release-AA-BB-CC.

  2. A light release, name of the form light-YYMM-CODENAME.

  3. A monthly build, name of the form build-YYYY-MM-DD.

  4. A prerelease, name of the form prerelease-AA-00-00abc.

Monthly builds are made from the development version of the code, they are a CI tool. Pre-releases are for testing and validation before a full release.


Neither monthly builds nor prereleases are supported for use in analysis. They are not recommended for anything other than software validation. They can disappear without warning.

Full release

A full release (release-AA-BB-CC) is a complete version of basf2 that has undergone extensive testing and valiation. Real data production and Monte Carlo simulation campaigns are always based on a full release. With a full release you are able to perform the full reconstruction chain, including analysis.

The numbering of releases follows semantic versioning, AA is the major release number, BB the minor, and CC the patch. A major version of the software may contain non-backward-compatible changes to the user interface.


If you do not know what release you need to use, then the newest supported full release should be your default choice. The command b2help-releases with no arguments, shows this.


We support the most recent patch to the last two major releases.

Light releases

A light release (light-YYMM-CODENAME) is a release made from only the analysis, skim, mdst, mva, b2ii, and framework packages.

They are suitable for doing high-level analysis tasks which do not require the generation or reconstruction of data. If you are running over some MC or data that already exists (e.g. was produced by the data production group) and want access to newer features, you should consider using a light release.


By construction, a light release can only be used to process files in the mdst format. Reading any other file format (e.g., cdst) will cause a crash.


Unfortunately from light release to light release the syntax may change and you might have to update your analysis scripts.


We support the last two light releases.

For more technical information about light releases, see BELLE2-NOTE-TE-2018-013.

2.3.2. Physics Analysis Setup

If you want to develop your analysis you can setup your own analysis project with

$ b2analysis-create analysis_name release_version

where you should replace the analysis_name with a meaningful name for your analysis. This will be the directory name for your project as well as the name of the git repository on the server. release_version should be replaced with the release version your analysis will be based on. After this you can

$ cd analysis_name
$ b2setup

to setup your analysis project. You can add your own basf2 Module to this analysis by running

$ b2code-module ModuleName

where ModuleName is the name of the module you want to create. The command will ask you a few questions that should be more or less self-explanatory. The requested information includes your name, module parameters, input and output objects, methods, and descriptions for doxygen comments. If unsure you can usually just hit enter. The b2code-module command will create a skeleton header and source file of your module and include them in the files known to git.

To compile your code simply type

$ scons

in your analysis working directory,

An advantage of having the analysis code in git is that you can check it out at any other location and continue your work there. The git repository takes care of synchronizing the multiple local version of the code. To get the code of an existing analysis with a certain name type

$ b2analysis-get <analysis name>

Again, changes can be submitted to the git repository with git commit followed by git push. To get the changes made in a different local version and committed to the central repository to your current local analysis working directory, use the command

$ git pull --rebase


After any update to your analysis code you need recompile it by running scons again.

Keeping your analysis up-to-date

Periodically you should update the release version of the software which your analysis is based upon. You will want to keep on top of improvements and bug-fixes. At the very least, you should update before your current release becomes unsupported. See Choosing a release for detailed explanation.

You can update your analysis project to a newer release using

$ b2analysis-update newer_release_version

after setting up your analysis (b2setup from your analysis directory). If no newer_release_version is specified the currently recommended one is taken.


After updating the release version you may have to adjust your analysis code to the new release.

A newer major release, or newer light release may contain non-backward-compatible changes to the user interface.

2.3.3. Development Setup

If you plan on developing code you should consider checking out the development version locally instead of using a pre compiled release:

$ b2code-create development

This will obtain the latest version from git. Once this is done you can setup this version using

$ cd development
$ b2setup

If you want to do the development from a certain version or branch just use git checkout to obtain it and run b2setup to make sure the correct externals version is set up, e.g.

$ git checkout release-XX-YY-ZZ
$ b2setup

After creating a development setup or switching to a different release you have to compile it.


after any update to your analysis code you need recompile it by running scons again.

2.3.4. Compiling your Code

To compile the code we use the SCons build system. Usually you can simply compile the code with running

$ scons


You have to recompile your code every time you modify, add or remove a file. If in doubt just run scons to be safe.

You can tell the build system how many CPUs to use in parallel by using the -j parameter:

$ scons -j 4

will compile the code with four jobs in parallel.


By default, scons will use as many parallel jobs as there are CPUs available on the system. This is fine on a local system but might not be desirable on a shared system like KEKCC or DESY NAF. Please adjust the amount of jobs to not block the whole machine if there are others using it.

To find out how many CPUs are available you can run

$ nproc

To find out what other options you can use please run

$ scons --help

Some of the parameters you can use are


Show a list of all available options.

-j N

Allow N jobs at once.


search of the directory tree for the SConstruct file. Use this to run scons from a sub directory of your code


be more quiet which will omit some status messages


show full commands passed to the compiler


change the color of the log messages. Possible values are: off, light, dark


build a light release. Useful to speed up compilation if you are developing high level analysis tools.


also build the sphinx documentation in the build/html sub directory.


if given all libraries will be checked for dependencies in the SConscript which are not actually needed


if given all libraries will be checked for missing direct dependencies after build


if you change any of the arguments --light, --extra-libpath or --extra-ccflags scons will recompile most of the code. So best keep the arguments consistent to avoid lengthy recompiles.

You can also supply a package name to only build the given package. For example if you know you only modified a file in the pxd package you can run scons pxd to only compile the pxd package. This is faster but will ignore some dependencies.


Always run a full scons before committing anything

2.3.5. Opening a pull request

To make your development part of the official software, you have to open a pull request. Before you can do this, you have to create a branch prefixed bugfix/ or feature/. Ideally, you should have created a JIRA issue for your development. Then, you can directly create a branch from there. If you already have local changes, execute the following sequence of git commands:

  1. git stash

  2. git pull

  3. git checkout <branchname>

  4. git stash pop

  5. git add

  6. git commit

  7. git push

Usually, it’s best to open the pull request only after you think that all the work has been completed and it is ready for review. However, there might be situations where you would like to get input from others. In that case, you might already open a pull request in an earlier stage. But please state in the description or in the title of the pull request that this is still work in progress (WIP) and which type of feedback you would like to receive from the reviewers.

Speaking about reviewers, the librarians of all packages that you touched have to be included as reviewers. You can find a list of the current librarians here. Alternatively, you can look directly into the .librarians files of the corresponding packages or, after opening the pull request, click on the grayed out Merge button in the top right corner, which should provide you with the necessary information as well.

Before the pull request can be merged, all reviewers must have approved and the build has to be successful. After you opened a pull request, each time you push new commits to your branch, a new build is initiated. Here is a list of best practices to make the review as smooth as possible:

  • Split changes of different issues into different commits.

  • Provide meaningful commit messages so that the reviewers know what was intended with those changes.

  • It should go without saying that the commit message must not contain inappropriate or even offensive language.

  • Make sure that your code compiles before pushing it.

  • Run at least the unit-tests of the packages that you touched (see Testing Tools)

  • Once you have opened a pull request, try not to push commits individually. Instead, make commits locally and push them at the end of the day or when you have finished all of your work.