Geometry Boost List: [ggl] rtree index

Subject: [ggl] rtree index
From: Barend Gehrels (Barend.Gehrels)
Date: 2009-06-20 04:33:31

Hi Federico,

>> What I would like, but I don't know if possible, is that the node/leaf do
>> not have virtual methods and no dynamic memory (other than std containers).
>> But  first we have to be sure that this is really possible. I still didn't
>> have a close look. It might mean that a leaf and a node are the same object
>> (so the only difference is that they do / don't have children, but that is
>> dynamic).
>>
>> This will be a major change, but if possible, a major increase in
>> performance.
>>     
>
> I really don't know if it's possible (I mean, surely it's possible but
> if it's not very hard to do). I barely remember that we tried to
> replace it quickly with Hartmut in the SoC but we failed. But maybe
> I'm confused with the virtual functions in tree index generic
> interface (that it's not present in the GGL version).
>
> I think that the node and leaf are not only different on that, they
> are different also in the way the elements/boxes are inserted and some
> other subtle differences. So we should have different versions of the
> methods, but I don't know now why it couldn't be done with templates.
>
> I'll begin removing the shared_ptrs and then I'll see if we can do
> something with that... Suggestion/ideas are welcome of course! :D
>   
OK, I understand. But you have to replace the shared pointers with 
something... they point to different things (either node or leaf). If it 
is another pointer, I don't know if you will gain so much...

Regards, Barend

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.osgeo.org/pipermail/ggl/attachments/20090620/d03ab219/attachment-0001.html