Currently, erlide parses the code as AST and traverses that whenever it needs information. It would be more efficient if a flat database of symbols can be built and searched, something like ctags but with more details.
This would be useful even for non-LSP erlide.