Elixir is a source code cross-referencer inspired by LXR. It’s written in Python and its main purpose is to index every release of a C or C++ project (like the Linux kernel) while keeping a minimal footprint.

It uses Git as a source-code file store and Berkeley DB for cross-reference data. Internally, it indexes Git blobs rather than trees of files to avoid duplicating work and data. It has a straightforward data structure (reminiscent of older LXR releases) to keep queries simple and fast.

You can see it in action on https://elixir.bootlin.com/

• Requirements
• Architecture
  • Database design
• Manual Installation
  • Install Dependences
  • Special dependencies
    • Device Tree and Kconfig Source syntax highlighting
    • Kconfig identifiers support
  • Download Elixir Project
  • Create Directory
  • Set environment variables
  • Clone Kernel source code
  • First Test
  • Create Database
  • Second Test
  • Configure httpd
  • Configure SELinux policy
  • Configure systemd log directory
  • Configure lighthttpd
• REST API usage
  • ident query
• Maintenance and enhancements
  • Using a cache to improve performance
  • Keeping Elixir databases up to date
  • Keeping git repository disk usage under control
• Building Docker images
• Hardware requirements
• Contributing to Elixir
  • Supporting a new project
  • Coding style
  • How to send patches
• Automated testing
  • Licensing of code in t/tree
• License

• Python >= 3.6

• Git >= 1.9

• The Jinja2 and Pygments (>= 2.2) Python libraries

• Berkeley DB (and its Python binding)

• Universal Ctags

• Perl (for non-greedy regexes and automated testing)

• Falcon and mod_wsgi (for the REST API)

Elixir has the following architecture:

The shell script (script.sh) is the lower layer and provides commands to interact with Git and other Unix utilities. The Python commands use the shell script’s services to provide access to the annotated source code and identifier lists (query.py) or to create and update the databases (update.py). Finally, the CGI interface (web.py) and the REST interface (api.py) use the query interface to generate HTML pages and to answer REST queries, respectively.

When installing the system, you should test each layer manually and make sure it works correctly before moving on to the next one.

./update.py stores a bidirectionnal mapping between git object hashes ("blobs") and a sequential key. The goal of indexing such hashes is to reduce their storage footprint (20 bytes for a SHA-1 hash versus 4 bytes for a 32 bit integer).

A detailed diagram of the databases will be provided. Until then, just use the Source, Luke.

For RedHat/CentOS

For Debian

To enable the REST API, follow the installation instructions on mod_wsgi and connect it to the apache installation as detailed in https://github.com/GrahamDumpleton/mod_wsgi#connecting-into-apache-installation.

To know which packages to install, you can also read the Docker files in the docker/ directory to know what packages Elixir needs in your favorite distribution.

Device Tree and Kconfig Source syntax highlighting

sudo pip3 install Pygments-2.6.1.elixir-py3-none-any.whl

Kconfig identifiers support

sudo dpkg -i universal-ctags_0+git20200526-0ubuntu1_amd64.deb
or
sudo rpm -iv universal-ctags-0+git~20e934e3-1.6.x86_64.rpm

sudo apt-mark hold universal-ctags

Two environment variables are used to tell Elixir where to find the project’s local git repository and its databases:

• LXR_REPO_DIR (the git repository directory for your project)

• LXR_DATA_DIR (the database directory for your project)

Now open /etc/profile and append the following content.

And then run source /etc/profile.

First clone the master tree released by Linus Torvalds:

Then, you should also declare a stable remote branch corresponding to the stable tree, to get all release updates:

Then, you can also declare an history remote branch corresponding to the old Linux versions not present in the other repos, to get all the old version still available:

Feel free to add more remote branches in this way, as Elixir will consider tags from all remote branches.

Generating the full database can take a long time: it takes about 15 hours on a Xeon E3-1245 v5 to index 1800 tags in the Linux kernel. For that reason, you may want to tweak the script (for example, by limiting the number of tags with a "head") in order to test the update and query commands. You can even create a new Git repository and just create one tag instead of using the official kernel repository which is very large.

Verify that the queries work:

The CGI interface (web.py) is meant to be called from your web server. Since it includes support for indexing multiple projects, it expects a different variable (LXR_PROJ_DIR) which points to a directory with a specific structure:

• <LXR_PROJ_DIR>

  • <project 1>

    • data

    • repo

  • <project 2>

    • data

    • repo

  • <project 3>

    • data

    • repo

It will then generate the other two variables upon calling the query command.

Now open /etc/httpd/conf.d/elixir.conf and write the following content. Note: If using apache2 (Ubuntu/Debian) instead of httpd (RedHat/Centos), the default config file to edit is: /etc/apache2/sites-enabled/000-default.conf

cgi and rewrite support has been enabled by default in RHEL/CentOS, but you should enable it manually if your distribution is Debian/Ubuntu.

Finally, start the httpd server.

When running systemd with SELinux enabled, httpd server can only visit limited directories. If your /path/elixir-data/ is not one of these allowed directories, you will be responded with 500 status code.

To allow httpd server to visit /path/elixir-data/, run following codes:

To check if it takes effect, run the following codes:

In case you want to check SELinux log releated with httpd, run the following codes:

By default, the error log of elixir will be put in /tmp/elixir-errors. However, systemd enables PrivateTmp by default. And, the final error directory will be like /tmp/systemd-private-xxxxx-httpd.service-xxxx/tmp/elixir-errors. If you want to disable it, configure httpd.service with the following attribute:

Here’s a sample configuration for lighthttpd:

Send a get request to /api/ident/<Project>/<Ident>?version=<version>&family=<family>. For example:

The response body is of the following structure:

At Bootlin, we’re using the Varnish http cache as a front-end to reduce the load on the server running the Elixir code.

To keep your Elixir databases up to date and index new versions that are released, we’re proposing to use a script like utils/update-elixir-data which is called through a daily cron job.

You can set $ELIXIR_THREADS if you want to change the number of threads used by update.py for indexing (default is 10 and minimum is 5).

As you keep updating your git repositories, you may notice that some can become considerably bigger than they originally were. This seems to happen when a gc.log file appears in a big repository, apparently causing git’s garbage collector (git gc) to fail, and therefore causing the repository to consume disk space at a fast pace every time new objects are fetched.

When this happens, you can save disk space by packing git directories as follows:

Actually, a second pass with the above commands will save even more space.

To process multiple git repositories in a loop, you may use the utils/pack-repositories that we are providing, run from the directory where all repositories are found.

Docker files are provided in the docker/ directory. To generate your own Docker image for indexing the sources of a project (for example for the Musl project which is much faster to index that Linux), download the Dockerfile file for your target distribution and run:

Then you can use your new container as follows (you get the container id from the output of docker build):

You can the open the below URL in a browser on your host: http://172.17.0.2/musl/latest/source (change the container IP address if you don’t get the default one)

Performance requirements depend mostly on the amount of traffic that you get on your Elixir service. However, a fast server also helps for the initial indexing of the projects.

SSD storage is strongly recommended because of the frequent access to git repositories.

At Bootlin, here are a few details about the server we’re using:

• As of July 2019, our Elixir service consumes 17 GB of data (supporting all projects), or for the Linux kernel alone (version 5.2 being the latest), 12 GB for indexing data, and 2 GB for the git repository.

• We’re using an LXD instance with 8 GB of RAM on a cloud server with 8 CPU cores running at 3.1 GHz.

Elixir has a very simple modular architecture that allows to support new source code projects by just adding a new file to the Elixir sources.

Elixir’s assumptions:

• Project sources have to be available in a git repository

• All project releases are associated to a given git tag. Elixir only considers such tags.

First make an installation of Elixir by following the above instructions. See the projects subdirectory for projects that are already supported.

Once Elixir works for at least one project, it’s time to clone the git repository for the project you want to support:

After doing this, you may also reference and fetch remote branches for this project, for example corresponding to the stable tree for the Linux kernel (see the instructions for Linux earlier in this document).

Now, in your LXR_PROJ_DIR directory, create a new directory for the new project:

Now, go back to the Elixir sources and test that tags are correctly extracted:

Depending on how you want to show the available versions on the Elixir pages, you may have to apply substitutions to each tag string, for example to add a v prefix if missing, for consistency with how other project versions are shown. You may also decide to ignore specific tags. All this can be done by redefining the default list_tags() function in a new projects/<projectname>.sh file. Here’s an example (projects/zephyr.sh file):

Note that <project_name> must match the name of the directory that you created under LXR_PROJ_DIR.

The next step is to make sure that versions are classified as you wish in the version menu. This classification work is done through the list_tags_h() function which generates the output of the ./scripts.sh list-tags -h command. Here’s what you get for the Linux project:

The first column is the top level menu entry for versions. The second one is the next level menu entry, and the third one is the actual version that can be selected by the menu. Note that this third entry must correspond to the exact name of the tag in git.

If the default behavior is not what you want, you will have to customize the list_tags_h function.

You should also make sure that Elixir properly identifies the most recent versions:

If needed, customize the get_latest() function.

If you want to enable support for compatible properties in Devicetree files, add dts_comp_support=1 at the beginning of projects/<projectname>.sh.

You are now ready to generate Elixir’s database for your new project:

You can then check that Elixir works through your http server.

If you wish to contribute to Elixir’s Python code, please follow the official coding style for Python.

The best way to share your contributions with us is to file a pull request on GitHub.

Elixir includes a simple test suite in t/. To run it, from the top-level Elixir directory, run:

The test suite uses code extracted from Linux v5.4 in t/tree.

The copied code is licensed as described in the COPYING file included with Linux. All the files copied carry SPDX license identifiers of GPL-2.0+ or GPL-2.0-or-later. Per GNU’s compatibility table, GPL 2.0+ code can be used under GPLv3 provided the combination is under GPLv3. Moreover, GNU’s overview of AGPLv3 indicates that its terms "effectively consist of the terms of GPLv3" plus the network-use paragraph. Therefore, the developers have a good-faith belief that licensing these files under AGPLv3 is authorized. (See also this issue comment for another example of a similar situation.)