remove_from_group error when quitting game

[volzhs]

Godot version:

3.2.4 6d5daac

OS/device including version:

not related

Issue description:

Error appears when quitting game.
More precisely, this error occurred when WorldEnvironment node is being removed.

ERROR: remove_from_group: Condition "!data.grouped.has(p_identifier)" is true.
   At: scene/main/node.cpp:1698.

it's a regression from 4954a86.
before this commit, this does not happen.

Steps to reproduce:

1. Add WorldEnvironment node in scene
2. Run game
3. Quit

Node::remove_from_group(Node * const this, const StringName & p_identifier) (scene/main/node.cpp:1697)
WorldEnvironment::_notification(WorldEnvironment * const this, int p_what) (scene/3d/world_environment.cpp:50)
WorldEnvironment::_notificationv(WorldEnvironment * const this, int p_notification, bool p_reversed) (scene/3d/world_environment.h:38)
Object::notification(Object * const this, int p_notification, bool p_reversed) (core/object.cpp:929)
Node::_propagate_exit_tree(Node * const this) (scene/main/node.cpp:305)
Node::_propagate_exit_tree(Node * const this) (scene/main/node.cpp:294)
Node::_propagate_exit_tree(Node * const this) (scene/main/node.cpp:294)
Node::_propagate_exit_tree(Node * const this) (scene/main/node.cpp:294)
Node::_propagate_exit_tree(Node * const this) (scene/main/node.cpp:294)
Node::_propagate_exit_tree(Node * const this) (scene/main/node.cpp:294)
Node::_set_tree(Node * const this, SceneTree * p_tree) (scene/main/node.cpp:2546)
SceneTree::finish(SceneTree * const this) (scene/main/scene_tree.cpp:619)
OS_X11::run(OS_X11 * const this) (platform/x11/os_x11.cpp:3611)
main(int argc, char ** argv) (platform/x11/godot_x11.cpp:56)

Minimal reproduction project:

fxaa_transparent_bg.zip

[volzhs]

cc @lawnjelly

[lawnjelly]

Is probably an order of update thing, as the collision pairing is now being updated from draw, so it's possible some things are never getting unpaired during shutdown. Will take a look. 👍

It actually occurs with a single world_environment node as the only thing in the scene, which is making debugging easier!

It turns out to be to do with accessing the scenarios in the new update that was put in specifically. It doesn't need to call the BVH to trigger the bug, all it has to do is loop through the scenario_owner scenarios and call get to get a pointer from the RID. So it could be that these are best not updated at this particular point or I need a guard. I'll see if I can fix but if not I'll ask reduz as this is something he will know more about.

[lawnjelly]

So it turns out the function I was calling to loop through the scenarios in visual_server seems to be debug only. So I've made a PR to revert the problematic bit temporarily, until I've consulted with reduz on the proper way to get access to the scenarios. Am too tired to address this fully tonight.

In the meantime you can turn the BVH off in rendering->quality->spatial_partitioning->use_bvh and it should work as normal.

Function that causes the problem is RID_Owner::get_owned_list. It could be it is not thread safe as well as being debug only.

[akien-mga]

Fixed by #45335.