From: Janek Kozicki (janek_listy_at_[hidden])
Date: 2007-07-18 12:00:32
Matias Capeletto said: (by the date of Tue, 17 Jul 2007 14:28:28 -0300)
> It was accepted in march :)
> But your feedback will be great! Really...
Hi Matias, thanks for your faith in... how good my comments might be :)
Currently I'm not going to go straight to the C++ coding (aka.
testing Bimap "live"). I am preparing grounds for huge refactoring.
And I'm looking at how to do it the best way.
My codebase of 15k lines already works quite good. Now in fact I want
to reorganize everything inside so it will be clean again. Because
currently more and more hacks appear to get something done.
I plan to stop using my own plugin-loading and serialization library
and switch to boost::extension and ::serialization. Propose my own
Multimethods lirbary for boost inclusion (after boostifying it ;).
And stop using my own containers, but use boost instead.
So... in fact I have another question about containers. But I'm
afraid that bimap is not an answer here... So I keep [multi_index] in
the subject line :-)
Imagine that in my scientific simulation software I have bodies that
can interact with each other. Any number of bodies can interact. As
you already know, bodies are idenfied by id numbers. But
interactions are identified by std::set<int> - a group of bodies that
are interacting. So I have a class Interaction and it's indexed by
sets of numbers like:
1,2,3 class Interaction between boedies 1,2,3
2,4,6 some other Interaction
This is 'groups' in fact (the one with which I started this thread).
To make stuff easier to manage, I'll use negative numbers for
group_id and positive for body_id. So basically I could rewrite above
-1 class Interaction between boedies 1,2,3
-4 some other Interaction
What I need here is a *cooperation* between containers (because
information about group components is stored in other container). So
with the Interaction container above I need to iterate FAST on following
- all interactions from first to last (easy)
- all bodies that take part in interaction -3, answer is: 3,4
- all interactions in which body 3 is involved, answer is: -1,-2,-3
This is convoluted :) And even worse... becasue I need interactions
to have groups too! We can call it a group of groups, like this:
identifier group interaction
-1 -5 class Interaction between boedies 1,2,3
-2 -5 Interaction
-3 -6 Interaction
-4 -6 some other Interaction
You can see that group -5 has two components: -1,-2, and group -6 has -3 and -4.
now I need to
- iterate over all interactions from group -5, answer is: -1 -2
- iterate over all bodies from group -5, answer is: 1,2,3
- all interactions involving body 3 that are in group -5, answer: -1 -2
- all interactions in which group of bodies 2 and 3 interact, answer: -1 -2 (may be slower)
I'm stil not sure if that's all i need. Still working on it ;-)
The hardest(??) part is that I will need separate containers (cannot
make one huge container for everything). So imagine, I have a class
Model, that holds the context of my computations.
To simplify a little:
I have two classes that describe a single body. Those are BodyState
(position, velocity) and BodyShape (vertex positions). They reside in
So, I say, that body_states and body_shapes describe the same
body with body_id=5. And If I need to know to which groups it belongs I look in
groups. To see its name (a body could have name too) I see into names.
typedef int id;
typedef id body_id;
typedef id group_id;
bimap< tagged< int, body_id >, multiset_of< tagged< int, group_id > > >
groups; // a bimap between positive body_id <-> negative group_id
std::map<body_id , BodyState*>
std::map<body_id , BodyShape*>
Now I want to throw interactions into the basket. They can (I'm still
not sure) use the 'groups' container to index the content. Easiest is to say:
But here, group_id used in 'interactions' is a "group of groups".
And (of course) I want to iterate over all interactions involved (or
bodies involved) as described above.
So I guess that I need multi_index here, since bimap won't be enough? :)
PS: actually I have more classes and containers here, this is simplified a bit.
This is not the end.... ;)
But for the next part I'm almost sure that I'll need a custom container.
I'll not describe it now, just to keep you excited a bit.
-- Janek Kozicki |
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk