/home/jacques/blender-git/blender/source/blender/makesrna/intern/rna_attribute.c: In function ‘RNA_def_attribute’:
/home/jacques/blender-git/blender/source/blender/makesrna/intern/rna_attribute.c:604:3: error: implicit declaration of function ‘rna_def_vertex_color_group’ [-Werror=implicit-function-declaration]
  604 |   rna_def_vertex_color_group(brna);
      |   ^~~~~~~~~~~~~~~~~~~~~~~~~~

Guess, I can just remove this line for testing.

In rna_attribute.c it says, that the PRIMITIVE domain in a point cloud are the points. However, in attribute.c, the VERTEX domain is used for points in a point cloud. It's probably best to change the description in rna_attribute.c.

I wonder if it would be good to splitup ATT_DOMAIN_PRIMITIVE into ATT_DOMAIN_POLYGON and ATT_DOMAIN_HAIR.

Also I'm not sure about the ATT prefix. Not sure if I've ever seen that as abbreviation for "Attribute". ATTR might work a bit better.

I consider this ready for review now.

The only issue that has not been solved is unique naming across all domains. This is technically quite difficult to implement since our CustomData manipulation all happens per domain throughout mesh modelling tools and the modifier stack. Enforcing this reliably would probably require refactoring custom data storage as a whole. Some other applications (e.g. Houdini) also don't enforce this and have a priority order for domains.

So, I don't think this is something we have to solve now, if at all.

I posted 3 follow up patches as well, for the UI, storage of position and radius, and a spreadsheet view. Besides that still important to make this complete would be to support for rendering in Cycles and Eevee with these attributes, as well as Workbench visualization. And of course particle nodes integration.

LGTM, just some small comments. This can be merged without another round of review from me.

In D8200#210464, @Brecht Van Lommel (brecht) wrote:

The only issue that has not been solved is unique naming across all domains. This is technically quite difficult to implement since our CustomData manipulation all happens per domain throughout mesh modelling tools and the modifier stack. Enforcing this reliably would probably require refactoring custom data storage as a whole. Some other applications (e.g. Houdini) also don't enforce this and have a priority order for domains.

Ok, I think I can live with that. When we define a priority order for domains, an attribute name alone should still be able to uniquely define an attribute in a given context.