In his last post, Murray writes that GtkTreeView calls gtk_tree_model_iter_next() for every row. This is indeed true, for the simple reason that the internal state keeping of GtkTreeView needs an exact internal representation of what is being displayed. This representation is in the form of an RB-Tree, wherein there is a node for each visible row. The definition of visible that we use here includes all nodes that are outside the current scroll area that is being shown. Visible distinguishes collapsed and expanded nodes from each other. Each node in this RB-Tree contains the properties of that row, such as its height, is it a parent or not, is it selected, etc, etc. In order to build this tree, we need to iterate over all nodes in the model, hence that iter_next() is being called for all these nodes. Even for a million rows this is not a problem, building the RB-Tree is quite a fast operation. This has all been discussed at length recently at gtk-devel-list: see here. The main performance problems have always been the measuring of all rows, which is something you can avoid by using fixed-height-mode. A lot of people have heard about me working on some experimental tree view improvements, this is also true. In my experimental branch fixed-height-mode has been removed, since the need for validating all rows has been removed. It can handle 4 million rows just fine. So, do we really have to put the entire model in the internal representation? This is actually an unanswered question. It of course greatly simplifies the implementation of GtkTreeView, but nobody says that it would not be possible to modify GtkTreeView in such a way that this is not required anymore. I haven't looked at this in depth myself yet, maybe at some point in the future when I have time and finished studies and and and … (I spent a lot of time on University stuff these days, it has been years since I have been so much motivated to work on finishing studies). While Philip and Jürg's ideas on iterator types and GtkTreeModel improvements are certainly interesting, it is not a (complete) solution for this particular problem. It will especially easify the implementation of data sources/tree models and their testability. The solution for large data sources is really in the view part and not (so much) the model part. Time for more Minix hacking now …