RAG and large context sizes mitigate this well enough for me. Ingest the library's docs (and maybe a sizable chunk of your codebase) and use that to get better LLM output that isn't out of date or hallucinated.