Re: [Boost-users] [Graph] need suggestion

Subject: Re: [Boost-users] [Graph] need suggestion
From: Cedric Laczny (cedric.laczny_at_[hidden])
Date: 2012-01-03 00:49:01

On Monday, January 02, 2012 11:18 PM, srinivas boppu wrote:
> Hello All,
> 

Hi

> I need a graph structure as follows. (Here my input is a Graph which I call
> it as " Control Flow Graph".)
> 
> On this "Control Flow Graph", I run a algorithm it creates another graph,
> and on that I will run again
> the algorithm, it will create another "Graph"  and it continues  until it
> become a  "Single Node".
> 

Out of the box, I would recommend you to have a look at filtered_graph<>, that 
can be applied to an adjacency_list<>. It can be seen as a mask that you put 
over your original graph, leaving the original graph untouched. What you need 
to provide filtered_graph<> with is a predicate that decides if an edge/vertex 
should be visible or not. Here, I could think of a map that is updated at each 
run and hence directly keeps track of the edges/vertices to be visible.
This seems to be the most elegant solution using the BGL IMHO.
If you have an idea how to allow for constant access of the visible/non-
visible property, this would greatly speed up this solution.

> I should be able to access all the Graphs in each stage and my original
> graph information should be
> untouched.
> 

The above suggestion leaves you with exactly two graphs: the original as 
adjacency_list<> and the filtered one as filtered_graph<>. So if you need the 
intermediate graphs, you could copy the filtered_graph<> to an adjacency_list<> 
which is of course potentially time-expensive as it will copy all the visible 
vertices and edges. But since you said that you need the intermediates you 
probably won't be able to avoid something like this; after all, these graphs 
need to be created. Please correct me if I'm wrong.

> What is the best way to do it ? Any suggestions ?
> 

Don't know if it is the best way, but definitely one feasible way, closely 
oriented on the functionality of BGL.

> (Actually, I am trying to solve one problem using  "Interval Based
> Analysis" of Control Flow Graphs. I would like to divide the given
> main control flow graph into disjoint intervals. You can have a look at the
> following link for more information
> http://nptel.iitm.ac.in/courses/106108052/module9/control-flow-ana-2.pdf
> )
> 
> Regards,
> Srinivas

Hope that helps.

Best,

Cedric