Re: [Boost-users] [boost] Create Graph in Function

Subject: Re: [Boost-users] [boost] Create Graph in Function
From: Jeremiah Willcock (jewillco_at_[hidden])
Date: 2009-10-29 08:47:56

On Wed, 28 Oct 2009, James wrote:

> Jeremiah Willcock wrote:
>> On Wed, 28 Oct 2009, James wrote:
>> 
>>> Jeremiah Willcock wrote:
>>>> You could also just declare the graph in whichever scope will end up 
>>>> storing it and then pass it by reference to the construction function:
>>>> 
>>>> void function(Graph& g) {
>>>>   // build g, or build a temporary and assign it to g
>>>> }
>>>> 
>>>> Graph the_graph;
>>>> function(the_graph);
>>>> 
>>>> Where are you going to be putting the final graph?  Will it be in a 
>>>> variable in your main(), even if it is not constructed there?
>>>> 
>>>> -- Jeremiah Willcock
>>> Jeremiah,
>>> 
>>> I'm sorry if I was not clear in my first post.  The majority of the code 
>>> (along with the main routine) is written in Fortran. I am using the graph 
>>> to update a number of arrays which are needed for this Fortran code.  I 
>>> will create the graph in a subroutine (or c++ function) and modify it in 
>>> another c++ function. The graph is not called in the main routine, only in 
>>> subroutines.
>>> 
>>> I need to know the way the graph is stored so I can pass it back to the 
>>> main function (or perhaps define it as a public variable).
>>> 
>>> In reference to the code snippet above, am I correct in reading it as the 
>>> creating routine will be defined as:
>>> void function(Graph& g) {
>>> // build g, or build a temporary and assign it to g
>>> }
>>> and then the function to modify it would be defined as:
>>> void modify_the_graph(Graph& g){
>>>   \\ some graph manipulations here
>>> }
>> 
>> Where is the variable holding the graph going to be?  If it doesn't have a 
>> Fortran type the variable cannot be in Fortran code.  Are you going to have 
>> a global variable (in C++) that contains the graph?  Otherwise, you might 
>> need to use a raw (not shared) pointer to the graph, managing the memory 
>> yourself, and pass it between Fortran and C++ as either some kind of 
>> interoperability C pointer type (in newer Fortran versions), a large enough 
>> integer, or maybe a Cray pointer.
>> 
>> -- Jeremiah Willcock
>
> That is the issue.  I agree that to manipulate the graph in the Fortran 
> portion of the code would be difficult (due the lack of a comparable data 
> type).  Luckily, I don't need to modify the graph in the Fortran portion of 
> the code, I just need to allow for a c++ function to access the graph.  If I 
> can define the graph as a global variable in c++ and negate the need to pass 
> the graph from the create_graph function to the modify_graph function via the 
> Fortran main routine:
>  Create_Graph.cpp -> Main.f -> Modify_Graph.cpp
> would be great.
>
> Would I accomplish this as:
> in create_graph.cpp
> Graph& g;
>
> void create_graph(){
> \\ create graph
> }
>
> and in modify_graph.cpp
> extern Graph& g;
>
> void modify_graph(){
> \\...
> }

The storage for the graph needs to be somewhere; making the variables of 
type Graph rather than Graph& would solve that problem.

-- Jeremiah Willcock