Autocomplete LRU Cache Design Choice

[Ayyanaruto]

Hey Continue team 👋

I was going through the Autocomplete implementation and noticed that the LRU cache uses SQLite with queries. I was wondering why not use an in-memory cache instead?

For example, using a simple Map in Node.js could make get/put operations faster by skipping DB queries. Or maybe a hybrid approach could work keeping recent entries in memory and syncing them to SQLite periodically in batches if persistence is needed.

Just curious about the thought process behind this design choice :👀

[RomneyDa]

I think the original concern was with memory bloat in larger repos, but that seems like a non-issue with good design. Seems like an interesting potential improvement to me

[Ayyanaruto]

Yeah, with a proper LRU size limit and cleanup strategy, the memory side should stay pretty manageable.
Should I open an issue for this? Would be super excited to work on it! 😄

What do you think would work best in our case a purely in-memory cache or an in-memory cache that syncs to the DB? Since in-memory would be volatile but definitely the fastest option.