CodeGraph is a tool that gives coding agents direct access to the depths of a codebase. It’s orders of magnitude more efficient (both in time and token use) than plain text search.
Unfortunately, on an Ada codebase, it currently returns “0 files indexed” – it doesn’t know how to parse Ada. I’ve requested to implement Ada support and added helpful detail information.
The project maintainer uses GitHub reactions to gauge interest in a language and decide on implementation priority.
Regardless if you’re into LLM-assisted coding or not, I’m convinced it can only help Ada if there’s support for it in modern developer tooling. That’s why I’m asking you kindly to leave a thumbs-up on my inclusion request.
I wasn’t familiar with codegraph, thanks for the reference. The tool looks sophisticated and efficient.
On my side, I had noticed with Mistral that when I referenced a variable, it would perform a very “primitive” search, relying on grep / ls / find to locate variables or identify the relevant files.
I figured then that in an Ada skill, it would be worth telling the LLM how to use gnatfind or gnatxref (having it read the .ali files directly seems less efficient) to navigate the source code more effectively.
This would be a simpler “low-tech” solution to reduce the tokens wasted navigating the sources, with the downside of requiring compilable sources, but with the advantage of doing without yet another MCP server, and with no risk of misinterpreting the source code.