Nathan Ridge

> Please assign it to me if nobody's working on this yet. I've found it really inconvenient and would like to take a look into this in my spare time....

> I suppose a similar approach can be applied to point 2 as long as the Sema action can see our `HeuristicResolver` -- or we could duplicate part of the...

I think the short answer here is that clangd needs the compile command in `compile_commands.json` to be a **clang-like** command. Specifically, while the name of the driver itself (the argv[0]...

> Which fixes the `loop.cpp` file but if I got to the definition of the `val.h` file it then complains about other emscripten headers included there. Just a guess: is...

Clangd's Modules support is known to be incomplete, with work in progress at https://github.com/clangd/clangd/issues/1293. It's worth trying again once the patch in that bug lands to see if it resolves...

Clangd does load the entire project index into memory, so for a project as large as Chromium I think this is probably expected, yes. Another C++ IDE I've worked with,...

Can you share what LSP messages your client sends to clangd before it crashes? It's fine if it doesn't include paths or contents, just what messages and in what order.

Ok, so the client is opening a large number of files (about 40 in the posted trace) in quick succession on startup. Obviously that shouldn't cause the server to crash...

Regarding the crash, I believe the stack trace suggests that it's taking place on a dedicated thread clangd spins up for adding the contents of the document's preamble to the...

@siegel If you're able to test with clangd 16 and see whether the crash occurs with that release, that would help gain confidence in async preamble indexing being the cuplrit.