mirror of
https://github.com/l0ng-ai/openidea.git
synced 2026-11-08 08:52:12 +00:00
Switching tabs remounts the editor view and the marker cache lived in its plugin; focusing an editor dropped the cache too. So every visit re-ran an implementation search per declaration in view, and on jdtls each one walks the whole workspace (#879: java at ~40% CPU while moving between classes). Answers are now cached per file and dropped only when other files change (edits, or source changes on disk; build output ignored) or a language server settles, through one window-wide subscription whose echo guard survives remounts. Fixes #879