Indexing a C++ repository with LSIF

This guide walks through setting up LSIF generation for a C++ codebase using lsif-clang. These instructions should apply to any C++ project that is buildable with clang or clang++. (This should also cover most projects built with gcc or g++.)

Local dev setup

  1. Copy the files in the lsif-docker directory of sourcegraph/tesseract to a local lsif-docker directory in your C++ repository (the one you wish to index).

  2. Replace the contents of lsif-docker/ with commands that install any requisite build dependencies of the project. These should be dependencies that do not vary from revision to revision.

  3. Modify lsif-docker/ to clone your repository to the /source directory in the Docker container's filesystem.

  4. Modify lsif-docker/ to generate a compilation database (compile_commands.json).

    1. If you use autotools to build your project (./ && ./configure && make), you can probably keep the existing contents.

    2. If you build your project using CMake, you can use cmake -DCMAKE_EXPORT_COMPILE_COMMANDS=ON ..

    3. If you use Bazel, you can use bazel-compilation-database:

      git clone --depth=10 /bazel-compilation-database
    4. If you use another build system or if any of the above steps break, consult this very helpful guide to generating compilation databases for various build systems. It may be helpful to docker build your container and docker run -it $IMAGE to get an interactive shell into the container, so you can ensure the build environment is correct. We recommend getting the project to build normally first (e.g., emit a binary) and then following the aforementioned guide to modify the regular build steps to emit a compilation database.

      1. Most often, the compile_commands.json file will be emitted in the root directory of the repository. If this is not the case, you'll also need to modify lsif-docker/ to cd into the directory containing it and then run lsif-clang --project-root=/source compile_commands.json. If you're unsure of where compile_commands.json will be emitted, just continue to the next step for now.
  5. Run docker build lsif-docker to build the Docker image.

  6. Generate a Sourcegraph access token from your Sourcegraph instance (Settings > Access tokens). Give it sudo permission.

  7. Run the following command to generate and upload LSIF data to Sourcegraph:


with the following substitutions:

  • SRC_ACCESS_TOKEN=: the Sourcegraph access token you just created
  • SRC_ENDPOINT=: the URL to your Sourcegraph instance
  • PROJECT_REV=: the revision of you repository to be indexed
  • $IMAGE_ID: the ID of the Docker image you just built

If successful, you should see the upload visible in the repository settings page like this.

For reference, some examples of Dockerized C++ LSIF generation are:

Without Docker

It can sometimes be difficult to replicate the build environment inside a separate Docker container. If this situation applies to you, you'll need to install lsif-clang directly to your local dev environment.

  1. Install lsif-clang in your environment using the instructions in the lsif-clang repository.

  2. Install the src CLI.

  3. You'll need a way to generate a compilation database (i.e., a compile_commands.json file). There are different methods of doing so depending on your build tool, and we recommend reading these excellent notes. If there isn't an explicit way to generate one with your build tool, we recommend using Bear, which should be generic enough to handle any C++ build (but might be less efficient than explicit generation methods).

  4. Generate the compile_commands.json file in the root directory of the repository.

  5. Run lsif-clang compile_commands.json from the root directory. This should emit a dump.lsif file.

  6. Run src lsif upload from the root directory. You may first have to authenticate to your Sourcegraph instance.

If you run into issues along the way, a useful reference is one of the Dockerfiles currently used for LSIF generation for an open-source repository.

CI setup

Incorporating LSIF generation and uploading in CI will allow precise code navigation to remain up-to-date without any human intervention.

If you created a Dockerfile that encapsulates LSIF generation, you can use the same one in your CI pipeline.

If you installed lsif-clang directly into your host machine in development, you'll need to incorporate those steps into your build scripts.


With Docker

If the docker run command fails, you likely have an error in one of the lsif-docker/*.sh files. The general rule is if you can get your project to build normally (i.e., generate an executable), you can get the LSIF indexer to generate LSIF. So we recommend the following approach if things don't work on the first try:

  1. Build the Docker image: docker build lsif-docker
  2. Run the container with an interactive shell: docker run -it $IMAGE_ID bash
  3. In the container shell, cd /source and figure out what steps are needed to build the project.
  4. Once the build successfully completes, figure out which steps are needed to generate the compile_commands.json file. We have found this guide to be a useful resource.
  5. Once you've successfully generated compile_commands.json, cd into the directory containing compile_commands.json and run lsif-clang --project-root=/source compile_commands.json. This should generate a dump.lsif file in the same directory. This dump.lsif should contain JSON describing all the symbols and references in the codebase (it should be rather large).
  6. Once the dump.lsif file is generated correctly, set the environment variables SRC_ACCESS_TOKEN and SRC_ENDPOINT to the appropriate values in your shell. Then run src lsif upload from the directory containing the lsif.dump file. This should successfully upload the LSIF dump to Sourcegraph.
  7. After you've successfully done all of the above in the container's interactive shell, incorporate these steps into the lsif-docker/*.sh files. Then re-build the Docker container and try running docker run again.