Design of clangd
A language server’s main responsibilities are:
- keeping track of open files and unsaved changes
- providing the editor with up-to-date diagnostics (warnings/errors)
- responding to requests from the editor (such as go-to-definition)
Most of the design documentation focuses on the concepts, if you’re interested in the implementation we try to keep the code well-documented.
Request handling
clangd is based on the clang compiler, and at its core runs the clang parser
in a loop. The parser produces diagnostics as it encounters problems, and the
end result is a clang AST. The AST is saved to answer queries like “what kind
of symbol is under the cursor”.
There is one such loop for each open file. The TUScheduler class manages a
collection of ASTWorkers, each running on its own thread. Most operations run
on these threads, though code completion is a notable exception.
Index
C/C++/Objective-C are designed so that you can parse one source file at a time, without needing to see the whole program. This means that the resulting AST lacks certain information:
- if you call a function, we can find a declaration for it, but we may not have seen its definition
- there’s no way to find all the references to a struct you defined
- code completion can suggest variables that are in scope, but not those
where you’d need to add an
#include
To solve this, clangd maintains a database of symbols found anywhere in the program, called the index. This is extracted from each open file as it is parsed, and also by parsing the whole project in the background.
Remote index
The remote index system allows a codebase index to be built offline and shared across many clangd instances with an RPC server.