Switching levels at runtime (eval 'load_level \"world2\"' while a 617-unit world was loaded) hard-wedged Enu. The window froze and never recovered; two sample captures ~30s apart show the worker thread parked on the same condition variable the entire time.
Worker thread (100% of samples, both captures):
worker_thread
run_deferred
change_loaded_level (serializers)
load_level → load_units
code= → change_code → load_script → ctx.run (VM)
rawExecute → new_instance (host_bridge)
Build.clone → Build.init → init_unit
defaults (EdValue init)
subscriptions.send → send_or_buffer → channelSend
pthread_cond_wait <-- parked forever
It's mid-load of the new world, running a spawner script; cloning an instance sends the clone's Ed field defaults to subscribers, and the receiving context's channel is full. Nothing drains it, so the load never completes.
Main thread (~90% of samples): Main::iteration → VisualServerWrapMT::sync → cond_wait, with the render thread busy in RasterizerCanvasGLES3 polygon drawing (UI/canvas redraw). Frames are progressing, but whatever should be ticking the receiving Ed context during a live level switch isn't running — the worker stayed blocked on the identical channelSend across both captures, so this is a wedge, not slow drain.
Repro context:
- world1 loaded (58 root builds + 556 spawner instances = 617 units)
eval 'load_level "world2"' (world2: 81 scripts, several spawners)- Enu froze during the switch; force-quit required
Likely the same failure family as the take1 demo collapse (scripts erroring, extreme slowdown under heavy load until restart): Ed channel backpressure with a stalled consumer. A bounded channel send from the worker with no guaranteed drainer during change_loaded_level deadlocks; options include ticking the main/receiving ctx during level switch, making the send fail/buffer instead of blocking, or draining before running spawner scripts.
Full samples captured (main + worker + render stacks) — available on request.
Switching levels at runtime (
eval 'load_level \"world2\"'while a 617-unit world was loaded) hard-wedged Enu. The window froze and never recovered; twosamplecaptures ~30s apart show the worker thread parked on the same condition variable the entire time.Worker thread (100% of samples, both captures):
It's mid-load of the new world, running a spawner script; cloning an instance sends the clone's Ed field defaults to subscribers, and the receiving context's channel is full. Nothing drains it, so the load never completes.
Main thread (~90% of samples):
Main::iteration → VisualServerWrapMT::sync → cond_wait, with the render thread busy inRasterizerCanvasGLES3polygon drawing (UI/canvas redraw). Frames are progressing, but whatever should be ticking the receiving Ed context during a live level switch isn't running — the worker stayed blocked on the identicalchannelSendacross both captures, so this is a wedge, not slow drain.Repro context:
eval 'load_level "world2"'(world2: 81 scripts, several spawners)Likely the same failure family as the take1 demo collapse (scripts erroring, extreme slowdown under heavy load until restart): Ed channel backpressure with a stalled consumer. A bounded channel send from the worker with no guaranteed drainer during
change_loaded_leveldeadlocks; options include ticking the main/receiving ctx during level switch, making the send fail/buffer instead of blocking, or draining before running spawner scripts.Full samples captured (main + worker + render stacks) — available on request.