Developing LHCb Software

Learning Objectives

  • Learn how to work with and modify LHCb software projects and packages

  • Learn how to find and search the source code and its documentation

Prerequisites

Before starting, you should have a basic understanding of how to use git, similar to what has been taught during the Starterkit.

In this lesson, we’ll show you a complete workflow for developing the LHCb software using the Git version control system. At LHCb, we use GitLab to manage our Git repositories. Among other features, GitLab allows you to browse the source code for each project, and to review proposed changes (called merge requests) in a convenient web interface. You can find the CERN GitLab instance at https://gitlab.cern.ch.

In principle, there are multiple ways of interacting with the LHCb software repositories:

  1. A vanilla Git workflow using only the standard Git commands. This requires you to clone and compile an entire LHCb project at a time.

  2. An LHCb-specific workflow using a set of lb-* Git subcommands. This allows you to check out individual packages inside a project, and streamlines the modification of a few packages at a time.

Here, we want to focus on the second workflow. The first workflow will be discussed briefly at the bottom ofthis page. Note that, although the git lb-* commands are much easier for small changes to existing packages where recompiling an entire project would be cumbersome, for any serious development the usage of vanilla Git is much more stable. Please consider using it if you can spare the compilation time.

Initial setup

Before jumping in by creating a project in GitLab, you should make sure that your local Git configuration and your settings on GitLab are sufficiently set up.

  • Your name and email address should be set up in your local Git configuration. To ensure that this is the case, run

    git config --global user.name "Your Name"
    git config --global user.email "your.name@cern.ch"
    

    and put in your information.

  • Next, visit https://gitlab.cern.ch/-/user_settings/ssh_keys (logging in with your CERN credentials) and add an SSH key.

  • Finally, run this LHCb-specific configuration command:

    git config --global lb-use.protocol ssh
    

    This makes sure the LHCb commands use the ssh protocol instead of https.

This lesson introduces the commands:

  • lb-dev for setting up a new development environment

  • git lb-use and git lb-checkout for downloading LHCb software packages

If you want to make changes to a software package, you will need to set up a development environment. lb-dev is your friend here:

lb-dev --name DaVinciDev DaVinci/v45r8

The output should look similar to this:

Successfully created the local project DaVinciDev for x86_64-centos7-gcc9-opt in .

To start working:

  > cd ./DaVinciDev
  > git lb-use DaVinci
  > git lb-checkout DaVinci/vXrY MyPackage

then

  > make
  > make test

and optionally (CMake only)

  > make install

To build for another platform call

  > make platform=<platform id>

You can customize the configuration by editing the files 'build.conf' and 'CMakeLists.txt'
(see https://twiki.cern.ch/twiki/bin/view/LHCb/GaudiCMakeConfiguration for details).

lb-dev created local projects are Git repositories

When lb-dev creates the local project directory and creates the initial files there, it also calls git init and commits the first version of the files to the local Git repository (try running git log in there).

You can then use Git to keep track of your development, and share your code with others (for example with a new project in gitlab.cern.ch).

You can even create a new project in GitLab without leaving the terminal! Just define the git remote and a new project will be created as soon as you push.

git remote add ssh://git@gitlab.cern.ch:7999/<username>/<project-name>.git
git push -u origin master

Follow those instructions to compile the software:

cd DaVinciDev
make

Once that’s done, let’s try it out:

./run gaudirun.py

We just successfully ran the simplest possible Gaudi job with gaudirun.py! The ./run script runs the command that follows in the project “runtime” environment where the right varialbes are set. It works similar to lb-run, without the need to specify a package and version.

Obtain a shell with the runtime environment

Sometimes it is convenient to get a shell with the runtime environment so that you don’t need to prefix commands with ./run all the time. To do this, simply run

./run bash -l

While useful, it is generally advised to stick to prefixing commands with ./run (or lb-run). This way the command starts in the same reproducible environment every time and there’s less risk of getting to an inconsistent state.

Your new development environment won’t be very useful without any software to modify and build. So let’s check out one of the existing LHCb packages! These are stored in the LHCb Git repositories.

In order to obtain the source code of the package you want to work on, we’ll use the method described at the Git4LHCb page. These are a set of subcommands, starting with git lb-, that are designed to make developing LHCb software easier. For example, if you want to modify a TupleTool (which you discovered in the TupleTools and branches lesson), start by executing the following in the DaVinciDev directory:

git lb-use Analysis
git lb-checkout Analysis/run2-patches Phys/DecayTreeTuple

What if git asks for my password?

Make sure you succesfully completed the instructions under initial setup.

Under the hood, git lb-use will add and fetch the Analysis repository as a remote in Git – check this with git remote -v! git lb-checkout will then perform a partial checkout of the run2-patches branch of the Analysis repository, only adding the files under the Phys/DecayTreeTuple. directory.

Which project to use in git lb-use?

The project name to pass to git lb-use depends on the directories you want to check out and work on, and not on the project name you passed to lb-dev. Moreover you can call git lb-use several times for different remote projects in the same local project:

lb-dev --name DaVinciDev DaVinci/v45r8
cd DaVinciDev
git lb-use Analysis
git lb-use Stripping
git lb-use DaVinci

Note that in order for this to work, projects you specify in lb-use must be dependencies of the project you specify in lb-dev. In other words, the top-level project should be at the top of the dependency chain.

After checking out new packages, always remember to make them known to your local project with make configure (sometimes it might be necessary to fully clean up the build with make purge):

make configure

You can now modify the Phys/DecayTreeTuple package and run make to build it with your changes. You can test your changes with the ./run script.

./run gaudirun.py ntuple_options.py

If you have made changes that you’d like to be integrated into the official LHCb repositories, you can use git lb-push to push it to a new branch in the central Git repository. But please read the instructions page first.

Note that no one has permission to push directly to the official protected branches (e.g. master or run2-patches) of LHCb projects. To get your changes merged there from the branch to which you lb-push-ed, you need to create a merge request (MR), so the project maintainer can check your code. This can be done on the project repository web page, for example.

When pushing to a branch in a project in GitLab you will see a message like:

remote:
remote: Create merge request for my-branch:
remote:   https://gitlab.cern.ch/lhcb/Analysis/merge_requests/new?merge_request%5Bsource_branch%5D=my-branch
remote:

You can use the URL in the message to quickly create a merge request for the changes you just pushed.

When your merge request is approved and merged (which can be after some additional commits on your part), your changes are part of an official branch of the respective project, and your contributions are part of the LHCb software stack. Congratulations!

Nightlies

Typically before being approved every merge request will be tested in the so-called nightly builds. The nightly build infrastructure allows for building and testing the entire stack of LHCb physics projects, either overnight (in standard slots such as lhcb-run2-patches) or on demand (using /ci-test in GitLab discussions). The relevant maintainers will help with the exact workflow (e.g. who schedules the testing), which varies a bit depending on the project.

The automatic builds and tests can show compilation problems (e.g. new warnings) and unintended consequences of the proposed changes. To achieve the latter, they run some reference jobs and compare the output of the new build with a reference file.

If the aim of the commit was to change the ouput, e.g. because you increased the track reconstruction efficiency by a factor of two, mention it in the merge request description, such that the maintainers of the affected project can update the reference file.

Often it is useful to work directly from nightly builds (e.g. to debug an issue on another platform). You can use a nightly build version of a project with:

lb-dev -c x86_64-centos7-gcc9-opt --nightly lhcb-run2-patches DaVinci/run2-patches

A more detailed description of the command is found at SoftwareEnvTools.

If you want to take a look the source code, without checking it out, you can easily access the repository through the GitLab web interface. This website also provides search functionality, but the output is not always easy to read, especially if it returns many hits. To search a project much quicker, you can use lb-glimpse. It allows you to search for a given string in the source code of a particular LHCb project (and all its dependencies).

lb-glimpse "PVRefitter" DaVinci/v45r8

This works with every LHCb project and released version. Since it’s a shell command, you can easily process the output using less, grep, and other tools.

To get an idea of how a certain component of the LHCb software works, you can also access the doxygen documentation. One set of doxygen web pages is generated for several related projects, and is linked from the projects web sites, like for DaVinci. For example, the doxygen documentation for DaVinci v45r8 is here. See also the LHCb Computing web page for a list of projects.

Working with a full project checkout

The git lb-* commands are not strictly necessary, but they’re very convenient if you just want to quickly edit one package. Otherwise you’d have to build the entire project in which the package is residing, instead of using the precompiled version. However, if you develop across multiple packages, or want to use more sophisticated git commands, nothing prevents you from checking out an entire project – just don’t be surprised if it takes hours to compile on lxplus! Note that repeated compilation will be much faster and using a dedicated multi-core machine can speed things up a lot.

To check out a project, run the following:

git clone https://:@gitlab.cern.ch:8443/lhcb/DaVinci.git

replacing DaVinci with the project name of your choice. Next, initialise and compile it:

lb-project-init
make

optionally followed by make test to run the tests and/or make install to install it to the InstallArea directory. That’s all! You now have a vanilla Git repository containing all the source files of the project. You can find out more about working with full projects at the Git4LHCb page.