дзеркало
https://github.com/l0ng-ai/openidea.git
синхронізовано 2026-11-08 19:02:17 +02:00
l0ng-ai
fc6a27d4ba
jdtls ran on the parallel collector, which never shrinks the heap while idle, so each server kept its import/build peak committed for as long as it lived. It now runs on G1 with a periodic collection every minute and a 10–30% free-heap band, so memory goes back to the OS once the work is done (a SkyWalking checkout: 1.9 GB → 1.25–1.3 GB idle). A collector set in the JVM options still replaces ours with its tuning; setting either heap-free ratio drops both of ours. The Equinox launcher gets -nosplash: without it its native library loads AppKit on macOS and every jdtls shows as Not Responding in Activity Monitor. Part of #1008.