Is anyone else seeing massive indexing latency with multi-file context in VS Code forks?

Hey everyone,

I’m hoping to validate an issue I’ve been hitting over the last few weeks regarding AI extension indexing inside large repository architectures. We are currently scaling up our internal codebase, which heavily relies on deep module interdependencies.

When running standard code completion engines like GitHub Copilot inside standard VS Code setups, the autocomplete triggers within an acceptable 200ms window. However, the moment we attempt to refactor across multiple files simultaneously, the context mapping breaks down completely. The engine fails to track variables across un-indexed modules, leading to hallucinated dependencies.

To debug this, we isolated the environment and ran tests shifting the entire stack over to native forks designed specifically for deeper codebase comprehension. The difference in how the underlying systems handle background semantic indexing is night and day. If you are struggling with similar developer productivity bottlenecks, I highly recommend checking out this benchmark analysis on https://www.smashingapps.com/cursor-ai-vs-github-copilot-2026/84142046819000 to see how both indexing engines handle deep-context repositories under heavy loads.

The core issue seems to stem from how the language server protocol handles multi-root workspaces versus single-directory indexing loops. Has anyone found a specific workspace setting or custom .gitignore optimization to force standard extensions to index cross-file boundaries faster without choking memory allocation?

Appreciate any insights or configuration tweaks you guys might have!