clangd enrichment can trigger clang-tidy crashes and high CPU

[darthghandi]

Summary

Gortex auto-registers clangd for C/C++ repos and currently launches it with --background-index --header-insertion=never. On a large C/C++ repo, semantic enrichment drives a very large LSP workload through clangd. If the repo has a broad .clang-tidy (bugprone-*, readability-*, clang-analyzer-*, performance-*, etc.), clangd 22.1.6 can repeatedly crash while running clang-tidy matchers. Gortex then reconnects and repeats the work, keeping clangd at high CPU.

Evidence

A daemon log showed:

• semantic enrichment starting for lsp-clangd
• clangd argv: --background-index --header-insertion=never
• repeated crash stack dumps in clang-tidy checks:
  • bugprone-pointer-arithmetic-on-polymorphic-object
  • bugprone-not-null-terminated-result
  • performance-unnecessary-copy-initialization
• one completed clangd enrichment run took ~113 minutes and included tens of thousands of LSP requests.

Proposed fix

For gortex semantic enrichment, default clangd should avoid lint diagnostics and background indexing, e.g. --clang-tidy=false --background-index=false --header-insertion=never -j=1. Gortex needs semantic graph signal, not clang-tidy diagnostics, and should avoid reconnect loops amplifying repeated clangd crashes.