For a fun experiment I used Opus 5.5 to fork ccls, clang-based C++ language server, into “edgls” (https://github.com/MaskRay/edgls/) and replaced every use of the clang API with this front end. Definition and references, hover, call and inheritance hierarchies, rename, and diagnostics all work on real projects.
It’s much slower than clang though: Indexing ccls itself (41 TUs, 8 threads): 9.5 (edgls) s vs 6.3 (ccls) s, with 6.1 GB vs 2.5 GB peak memory. Parse-only, EDG is about 1.7× slower than clang -fsyntax-only on that code.
For a fun experiment I used Opus 5.5 to fork ccls, clang-based C++ language server, into “edgls” (https://github.com/MaskRay/edgls/) and replaced every use of the clang API with this front end. Definition and references, hover, call and inheritance hierarchies, rename, and diagnostics all work on real projects.
It’s much slower than clang though: Indexing ccls itself (41 TUs, 8 threads): 9.5 (edgls) s vs 6.3 (ccls) s, with 6.1 GB vs 2.5 GB peak memory. Parse-only, EDG is about 1.7× slower than clang -fsyntax-only on that code.
score 1 · canonical