mirror of
https://github.com/l0ng-ai/openidea.git
synced 2026-11-19 00:10:00 +00:00
Searching files, folders or symbols while a streaming scan indexed a project of more than 2,048 files could freeze the app for good. Queries hold IndexData::read while scanning on the global Rayon pool; the scan parsed on that same pool, took IndexData::read inside parse_found and published (IndexData::write, writer-preferring) mid-pass, so every worker could block on the lock or on the bounded channel while the query waited for a worker. - parse_found no longer locks: the previous content hash is captured during classification, and unchanged-content metadata is refreshed inside publish. - Background indexing and snapshot hydration run on a dedicated INDEX_POOL. - find_files, find_dirs and find_symbols are async commands on the blocking pool, off the main thread. Adds a test with concurrent queries across a 4,600-file scan and an unchanged-content rescan. Co-authored-by: dengliu <5108773+dengliu@users.noreply.github.com>