Using a remote index
Usually clangd will build an index of your project in the background. For large projects, this can be slow.
If someone has set up a clangd index server for your project, then you can query this central index instead of building your own.
Obtaining a suitable version of clangd
You need a remote-index-enabled version of clangd, such as the clangd 12 release on github.
The official LLVM releases and Debian etc packages of clangd 12 are not remote-index-enabled!
(If building clangd from source, you need
-DCLANGD_ENABLE_REMOTE=On in CMake).
Configuring clangd to use the remote index
You’ll need the address of the server you want to connect to, and the local source code it matches.
In your user configuration, add a section like:
If: PathMatch: /path/to/code/.* Index: External: Server: someserver:5900 MountPoint: /path/to/code/
What to expect
After restarting clangd, it should no longer want to index your whole project, and you should get complete results for find-references etc.
Results may not be entirely up-to-date. Index servers usually scan the project once a day, and if you’re working on a branch you may get further skew. Files that you’ve had open will still be indexed locally and will be up-to-date.
Public index servers
Some open-source projects have public servers:
Running up an index server
This is a little more involved, and you’ll want to understand the design of the remote index. In short:
- you should periodically check out and index your project with a command like
clangd-indexer --executor=all-TUs /proj/compile_commands.json > proj.idx
- run the server as
clangd-index-server proj.idx /proj. It listens on port 50051 by default, and reloads the index file when it is overwritten.
clangd-index-server tools can be found in
clangd_indexing_tools.zip on the
clangd/llvm-remote-index is an example of a production-ready instance, with the indexing step running on GitHub Actions and the server running on Google Compute Engine.