I'd suggest making a minor adjustment to the patch to make it so we explicitly stay in the evaluated "domain". Probably not visible change on user level, but semantically it makes sense, is easy, and shows a good example of not mixing evaluated and original IDs.

In D13765#365047, @Sergey Sharybin (sergey) wrote:

I'd suggest making a minor adjustment to the patch to make it so we explicitly stay in the evaluated "domain". Probably not visible change on user level, but semantically it makes sense, is easy, and shows a good example of not mixing evaluated and original IDs.

Hm, afraid this is not going to work [ I either still dont get it -- or failed in trying to explain why in the patch description 🙂 ]
If we use the evaluated object, we cannot isolate a single constraint [the result will always be the result of all constraints]
quote:

For this we want to work on the original object/pchan
(since we need access to its unconstrained original matrix). However the
constraint target should be evaluated (the shrinkwrap constraint needs
to have access to the target's evaluated mesh).

Does this make sense? If mixing domains is somehow a no-no, then I am a bit out of ideas, I am afraid.

Example file (has 3 constraints):

• with my patch: applying shrinkwrap will put the new origin directly on the sphere (where it belongs)
• changing it to use the evaluated object will put the new origin where the result of all 3 constraints were (where it does not belong if I understand correctly)

Fair enough, need to modify some more lines to use matricies from ob_eval. Does P2709 applied on top of this patch work as expected for you?