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
- Manual Installation
- Install Dependencies
- Download Elixir Project
- Create a virtualenv for Elixir
- Create directories for project data
- Set environment variables
- Clone Kernel source code
- First Test
- Create Database
- Second Test
- Configure httpd
- Configure SELinux policy
- Configure systemd log directory
- Configuration for other servers
- REST API usage
- Maintenance and enhancements
- Building Docker images
- Hardware requirements
- Contributing to Elixir
- Automated testing
- License
-
Python >= 3.8
-
Git >= 1.9
-
The Jinja2 and Pygments (>= 2.7) 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)
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 web interface (web.py) and
uses 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.
For Debian
sudo apt install python3-pip python3-venv libdb-dev python3-dev build-essential universal-ctags perl git apache2 libapache2-mod-wsgi-py3 libjansson4
python -m venv /usr/local/elixir/venv . /usr/local/elixir/venv/bin/activate pip install -r /usr/local/elixir/requirements.txt ---
mkdir -p /path/elixir-data/linux/repo mkdir -p /path/elixir-data/linux/data
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.
export LXR_REPO_DIR=/path/elixir-data/linux/repo export LXR_DATA_DIR=/path/elixir-data/linux/data
And then run source /etc/profile.
First clone the master tree released by Linus Torvalds:
cd /path/elixir-data/linux git clone --bare https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git repo
Then, you should also declare a stable remote branch corresponding to the stable tree, to get all release updates:
cd repo git remote add stable git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git git fetch stable
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:
cd repo git remote add history https://github.com/bootlin/linux-history.git git fetch history --tags
Feel free to add more remote branches in this way, as Elixir will consider tags from all remote branches.
. ./venv/bin/activate ./update.py <number of threads>
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:
$ ./elixir/query.py v4.10 ident raw_spin_unlock_irq C $ ./elixir/query.py v4.10 file /kernel/sched/clock.c
|
Note
|
v4.10 can be replaced with any other tag.
NOTE: Don’t forget to activate the virtual environment!
|
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 replace /etc/apache2/sites-enabled/000-default.conf with docker/000-default.conf.
Note: If using httpd (RedHat/Centos) instead of apache2 (Ubuntu/Debian),
the default config file to edit is /etc/httpd/conf.d/elixir.conf.
Finally, start the httpd server.
systemctl restart apache2
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:
chcon -R -t httpd_sys_rw_content_t /path/elixir-data/
To check if it takes effect, run the following codes:
ls -Z /path/elixir-data/
In case you want to check SELinux log related with httpd, run the following codes:
audit2why -a | grep httpd | less
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:
PrivateTmp=false
Other HTTP servers (like nginx or lighthttpd) may not support WSGI and may require a separate WSGI server, like uWSGI.
Information about how to configure uWSGI with Lighthttpd can be found here: https://redmine.lighttpd.net/projects/lighttpd/wiki/HowToPythonWSGI#Python-WSGI-apps-via-uwsgi-SCGI-FastCGI-or-HTTP-using-the-uWSGI-server
Pull requests with example uWSGI configuration for Elixir are welcome.
After configuring httpd, you can test the API usage:
Send a get request to /api/ident/<Project>/<Ident>?version=<version>&family=<family>.
For example:
curl http://127.0.0.1/api/ident/barebox/cdev?version=latest&family=C
The response body is of the following structure:
{
"definitions":
[{"path": "commands/loadb.c", "line": 71, "type": "variable"}, ...],
"references":
[{"path": "arch/arm/boards/cm-fx6/board.c", "line": "64,64,71,72,75", "type": null}, ...]
}
At Bootlin, we’re using the Varnish http cache as a front-end to reduce the load on the server running the Elixir code.
.-------------. .---------------. .-----------------------. | Http client | --------> | Varnish cache | --------> | Apache running Elixir | '-------------' '---------------' '-----------------------'
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 $ELIXIR_THREADS if you want to change the number of threads used by
update.py for indexing (by default the number of CPUs on your system).
As you keep updating your git repositories, you may notice that some can become
considerably bigger than they originally were. This seems to happen when a gc.log
file appears in a big repository, apparently causing git’s garbage collector (git gc)
to fail, and therefore causing the repository to consume disk space at a fast
pace every time new objects are fetched.
When this happens, you can save disk space by packing git directories as follows:
cd <bare-repo> git prune rm gc.log git gc --aggressive
Actually, a second pass with the above commands will save even more space.
To process multiple git repositories in a loop, you may use the
utils/pack-repositories that we are providing, run from the directory
where all repositories are found.
Dockerfiles are provided in the docker/ directory.
To build the image, run the following commands:
# git clone https://github.com/bootlin/elixir.git ./elixir # docker build -t elixir --build-arg ELIXIR_VERSION=`git rev-parse --short HEAD` ./elixir/Dockerfile ./elixir
ELIXIR_VER build argument is optional. Since .git directory is not copied into Docker image by default, the option is used to pass a version string to Elixir.
You can then run the image using docker run.
Here we mount a host directory as Elixir data:
# mkdir ./elixir-data # docker run -v ./elixir-data/:/srv/elixir-data -d --name elixir-container elixir
The Docker image does not contain any repositories.
To index a repository, you can use the index-repository script.
For example, to add the musl repository, run:
# docker exec -it -e PYTHONUNBUFFERED=1 elixir-container \ /bin/bash -c 'export "PATH=/usr/local/elixir/venv/bin:$PATH" ; \ /usr/local/elixir/utils/index-repository \ musl https://git.musl-libc.org/git/musl'
Without PYTHONUNBUFFERED environment variable, update logs may show up with a delay.
Or, to run indexing in a separate container:
# docker run -e PYTHONUNBUFFERED=1 -v ./elixir-data/:/srv/elixir-data \ --entrypoint /bin/bash elixir -c \ 'export "PATH=/usr/local/elixir/venv/bin:$PATH" ; \ /usr/local/elixir/utils/index-repository \ musl https://git.musl-libc.org/git/musl'
You can also use utils/index-all-repositories to start indexing all officially supported repositories.
After indexing is done, Elixir should be available under the following URL on your host: http://172.17.0.2/musl/latest/source
If 172.17.0.2 does not answer, you can check the IP address of the container by running:
# docker inspect elixir-container | grep IPAddress
The Docker image does not automatically update repositories by itself.
You can, for example, start utils/update-elixir-data in the container (or in a separate container, with Elixir data volume/directory mounted)
from cron on the host to periodically update repositories.
You can easily use the Docker image as a development server by following the steps above, but mounting Elixir source directory from the host
into /usr/local/elixir/ in the container when running docker run elixir.
Changes in the code made on the host should be automatically reflected in the container.
You can use apache2ctl to restart Apache.
Error logs are available in /var/log/apache2/error.log within the container.
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:
cd /srv/git git clone --bare https://github.com/zephyrproject-rtos/zephyr
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:
cd $LXR_PROJ_DIR mkdir -p zephyr/data ln -s /srv/git/zephyr.git repo export LXR_DATA_DIR=$LXR_PROJ_DIR/data export LXR_REPO_DIR=$LXR_PROJ_DIR/repo
Now, go back to the Elixir sources and test that tags are correctly extracted:
./script.sh list-tags
Depending on how you want to show the available versions on the Elixir pages,
you may have to apply substitutions to each tag string, for example to add
a v prefix if missing, for consistency with how other project versions are
shown. You may also decide to ignore specific tags. All this can be done
by redefining the default list_tags() function in a new projects/<projectname>.sh
file. Here’s an example (projects/zephyr.sh file):
list_tags()
{
echo "$tags" |
grep -v '^zephyr-v'
}
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:
v4 v4.16 v4.16 v4 v4.16 v4.16-rc7 v4 v4.16 v4.16-rc6 v4 v4.16 v4.16-rc5 v4 v4.16 v4.16-rc4 v4 v4.16 v4.16-rc3 v4 v4.16 v4.16-rc2 v4 v4.16 v4.16-rc1 ...
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:
./script.sh get-latest
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:
./update.py <number of threads>
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:
prove
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.)
Elixir is copyright (c) 2017—2020 its contributors. It is licensed AGPLv3.
See the COPYING file included with Elixir for details.