Re: [gtkmm] Custom TreeStore Problems

[Christof Petig <christof petig-baender de>]

Christer Palm schrieb:
> Just curios - how do you perform a binary search when there's only a 
> forward_iterator? Isn't that in fact pretty much the root to the 
> TreeModel performance problems in general?

I guess you are right here: I suspect std::lower_bound determines that 
TreeStore::iterator is a sequential access iterator and emulates binary 
search by skipping elements (if increment is much cheaper than 
comparison this is a reasonable move). Of course this assertion does not 
hold for trees (which is why they usually implement lower_bound as a 
more efficient member function). I did not try whether linear search 
might be faster (my custom model is _much_ (10x at 10k) faster).
I still have no clue why this did not show up more cleanly in profiling 
(the call numbers were high but time seemed to be spent elsewhere).

-----
So, to Murray: Do you see that there is Value to a custom TreeModel in 
this case? Another example might be a model where column contents 
interact with each other (e.g. Color is dependant on text content).
-----

But ... this does not yet explain the problem mentioned on the 
gtk-app-devel list (see my previous post for the source). He did only 
append.

   Christof

PS: My custom TreeModel example at
http://cvs.berlios.de/cgi-bin/viewcvs.cgi/*checkout*/manuproc/Komponenten/examples/owntreemodel.cc?rev=HEAD&content-type=text/plain
performs well while calculating about 4G rows. (does not depend on 
anything else but gtkmm)