Fix emission_shape_changed signal error when using ShaderMaterial with GPUParticles3D

[paddy-exe]

Fixes #100113 (comment)

• Bugsquad edit, fixes: GPUParticles3D cannot set ShaderMaterial in process_material #101679

[akien-mga]

This won't work, since the process_material.is_valid() condition is true, so it will never evaluate the else if.
It should be:

Suggested change

	if (process_material.is_valid()) {
	material_rid = process_material->get_rid();
	} else if (cast_to<ParticleProcessMaterial>(*process_material)) {
	process_material->connect("emission_shape_changed", callable_mp((Node3D *)this, &GPUParticles3D::update_gizmos));
	}
	if (process_material.is_valid()) {
	material_rid = process_material->get_rid();
	if (Ref<ParticleProcessMaterial>(process_material).is_valid()) {
	process_material->connect("emission_shape_changed", callable_mp((Node3D *)this, &GPUParticles3D::update_gizmos));
	}
	}

[akien-mga]

There are some more issues with #100113 and its follow-up #101537:

• Fix regression from already disconnected emission shape changed signal #101537 removed the signal disconnect, but not the previous two lines which you just added to be able to disconnect the signal. These should be removed too. But I wonder: when is the signal disconnected now? What happens if you set the same ParticleProcessMaterial twice in a GPUParticles3D? Won't it error that the signal is already connected? The way we handle this usually is that the set_process_material setter should first check if the current material needs the signal disconnected, before assigning the new material and connecting the signal.
• The signal is emitted only in TOOLS_ENABLED, but this isn't documented, and there's still code like here that connects it in non-TOOLS_ENABLED builds.
• Finally there's [4.4-beta1] Loading compute shader on Linux crashes the editor #101611 which is why I started looking into Add visualization of 3D particle emission shapes #100113, but the above nitpicks don't seem to be what's causing the crash. I haven't had time to investigate further yet.

[paddy-exe]

Alright, I think I have solved the problem with ShaderMaterial and the signal connect and disconnect. Also added a documentation note in the class reference. In the CurveEdit class at least when setting a new curve you disconnect the signals and then re-add them to the new assigned one (see here) so I took inspiration from that.

I tested the various problem cases removing and re-adding the same material, adding a ShaderMaterial instead, using the same material on several GPUParticles3D Nodes and haven't encountered any errors or crashes (together with #101636 that is).

[AThousandShips]

Makes sense

[Repiteo]

Will be good to get this field-tested early

[Repiteo]

Thanks!