Entry 330

"scene does not sleep the event task: the pass runs on in its frame, disabled"

2026-10-01scriptenginedecomp

Filescrates/piney-event/src/vm/exec.rscrates/piney-event/src/host.rscrates/piney-game/src/area_host.rscrates/piney-game/src/field_host.rscrates/piney-game/src/world.rstools/test_event_vm.pycrates/piney-event/tests/vm_fixture.txtdocs/engine/event-vm.mddocs/engine/field-walk.md

74 read from the thread code that the event task, put to sleep inside scene, wakes on the next frame. It left that unchecked against the game. It does not sleep at all. All addresses are INF SLUS_202.67.

Why

ChangeScene (0x00167380) ends in ChangeRequest(6, 7) (0x001671e0). That calls ccDisableThEvent, then ccSleepNoSleepThread(1, 1) (0x0015a200), which does two things:

• it raises the sleep count of every thread whose flags (+0x10) lack bit 0;
• it puts the calling thread to sleep (flag 0x20, SleepThread) only if its flags lack bit 1 (andi 0x2 at 0x0015a27c).

ccThEvent sets both bits on itself as it starts (flags |= 3, 0x001b5a8c). So ChangeRequest returns into Execute, whose scene case (0x001b01d8) goes straight back to the loop head. Nothing there or in eventSub (0x001b5ef0, all 62 blocks) tests the phase. The rest of the block and the rest of the pass run in the same frame, with eventMng.phase at -1. Blocks with a phase condition then fail CheckOpen; blocks with none still run.

mode was already modelled this way (ChangeRequest(num, 7), then the next instruction). Every instruction that follows a scene in any volume's scripts is bookkeeping, with no wait: end_event, gate_unmark, mail, bbs_post, news_add, menu_clear, set_block.

The port

The interpreter had scene wait (Wait::ChangeRequest) until the next mode's host answered no. The rest of the block then ran a frame later, in the next area. It also left the phase at 4 until the frame's end, when the host disabled it.

Now scene calls Host::change_scene and disables the VM, and the pass goes on. Wait::ChangeRequest and the hosts' answers to it are gone.

Two tests show the difference:

• event_2_plays_to_the_scene_change (world.rs) now ends with event 2 closed (CLOSED | 0b111), no block playing, and the play pass done, all in frame 2295, the scene's frame.
• event_21_elk_walks_up_in_mac_anu first failed with only the disable missing. Event 21's block 1 ends in scene -2 ... (the same town again). With the phase still 4, block 4 (phase >= 4) opened in the old town and walked Elk in the new town's set-up, before his entry ("not done"). With the disable, block 4 waits for the new town's play, as in the game.

The checks' harness

tools/test_event_vm.py stubbed ChangeScene out entirely, so the game side never disabled the task. It is now hooked to do what its ChangeRequest does to the event manager (phase -1), without the scene. vm_fixture.txt is regenerated from it.