Kuba Ober wrote: > I think that the only real solution is that the tree must know of its > iterators and adjust them if tree structure changes. > > I presume that it should call into existence a proper tree base classes > for leafs and nodes. I wonder if they exist in the current design. [No > time to look at the code, sorry :( ] > > Essentially, it should be tree's leaves and nodes which should be able to > know which cursors point at them. Since there will be probably very few > cursors in existence at any given time (say 10 or something of that > order), it's conceiveable that: > a) a tree element, upon each change, will look through the "global" list > of cursors to see which cursors point to it > b) it will adjust them per the pending change > > That also makes cursors invariants between changes to either tree elements > or cursor position, and makes all cursors automatically always valid no > matter what. > > Does it make any sense? Is that the direction you are going? Am I talking > rubbish? I'm just a bystander and overseer, but I would like to know what > pattern are you converging to in the long term.
It has been already discussed in http://marc.theaimsgroup.com/?t=105393027500002&r=1&w=2 (and before, IIRC). The outcome: maybe, but let's try to stick to par "id"s if it is enough. And no, I don't think it is enough for when we have multiple BVs, but that can be introduced later. Regards, Alfredo
