I think both the modifier and particle system are part of the same object datablock, so I'm not sure this is strictly against the Blender design?

Copying the particle system modifier should also copy the particle system, and remapping the pointer would be done along with that. For dynamic paint it's not obvious what it should do, if there is no particle system with the same name on the target object I guess any particle system pointer should be cleared.

Or copying these modifier could just not be supported for now.

If it helps, the new operator I made to copy modifiers to selected objects (D9537) also copies the particle system to the new object if a particle system with the same settings doesn't exist already.

This should not be handled at operator level, as copying of modifiers happens every time you copy and Object ID (including CoW copying e.g.).

@Brecht Van Lommel (brecht) the problem is that modifier copy code has no way to tell for which object it is copying data, and it should most definitively never add dependency to new IDs (like what would happen if creating a new particle system using a new particle settings).

That is why I don't think it should hold pointers to data that do not belong to itself, because it should not and cannot manage that data properly. So a lose reference (by names) should be used instead, just like what we do with e.g. vertex groups.
[EDIT] And then higher-level code (operators) can actually duplicate or add other data as needed.

Not supporting duplication of those modifiers is not an option, again because it would break depsgraph/CoW handling.

I agree that the design is not great here, and I hope that with a new particle system we simply store all settings in the modifier / node rather than in a separate list somewhere.

However I don't think that a loose coupling based on name really improves the situation. You'd still need all the same code to make sure the particle system is copied along with the modifier, and new code will be needed to keep the link valid when the particle system is renamed. It makes things safer against crashes, but does not really improve the architecture in my opinion.

I suggest putting the code from D9537: Copy a single modifier to all selected objects. into a function that can be reused for the outliner, and ensure we get the design right when a new particle system is added.

CoW should work correctly by the way, since BKE_object_copy_particlesystems updates the pointers.

In T82945#1095666, @Brecht Van Lommel (brecht) wrote:

CoW should work correctly by the way, since BKE_object_copy_particlesystems updates the pointers.

Huuuu yes indeed.... this is soooo fragile. Will at least add proper comments in code about this exception handling :(

In T82945#1095665, @Brecht Van Lommel (brecht) wrote:

I suggest putting the code from D9537: Copy a single modifier to all selected objects. into a function that can be reused for the outliner, and ensure we get the design right when a new particle system is added.

OK, sounds like a reasonable workaround/temp solution.