With this, the changes from #61966 might not be necessary anymore.

With this, the changes from #61966 might not be necessary anymore.

I prefer to separate this fix from reverting the other one. I'm not sure if there is a use case where an internal node has an owner set. I'm afraid that reverting #61966 may (re)introduce some issues. Preserving both for a while will be safer for stability.

Maybe just pass in the corresponding base (child) nodes and find their owners, if the number of their child nodes is the same (Will it be different?).

Thanks for suggestions 👍 , I'll look at them and improve the code as soon as possible.

@marcinn1 @marcinn (not sure which account you're actively using?)

Just bumping to remind you about unresolved review comments. This can't make it into 4.2 at this point, but if you could find time to address the notes above, this should help to move this PR along.

Thanks. I'd look at this at evening.

I've rebased the branch, added suggested changes (excl r_ prefix), tested with MRP. Seems to be ok.

There are a couple of unresolved minor comments, but otherwise looks ok.

Should I add these changes before merging?

Yep.

Yep.

OK, this evening.

Updated.

Ok it is necessary, but in this situations we normally use const_cast<>.

Thanks for reminding me why I abandoned c++ ~20yrs ago ;)

Thanks!