Boost logo

Boost :

Subject: Re: [boost] [Graph] component_index crash on big graph
From: Jeremiah Willcock (jewillco_at_[hidden])
Date: 2009-07-07 21:18:42


On Tue, 7 Jul 2009, Sandeep Gupta wrote:

> On Tue, Jul 7, 2009 at 8:21 AM, Andrew Sutton <andrew.n.sutton_at_[hidden]>wrote:
>
>>>
>>
>>>> Thanks Andrew, I got confused with add_vertex and add_edge allocation
>>> properties. I have modified the code create numEdges. However the problem
>>> occurs while building component_index which does not access graph data
>>> structure but rather only the parent map. Attached is the modified file
>>> (also at http://codepad.org/5dZM6YBo).
>>
>>
>> I get confused too. I didn't write the class, so it's hard to remember.
>> Just
>> a heads up, though. The adjacency class has a constructor that takes a
>> number of vertices and pre-allocates them. For example:
>>
>> graph g(20);
>> cout << num_vertices(g); // prints 20.
>>
>> Hi Andrew,
> Got it. Back of my mind, I think I was uncomfortable about conversion from
> int to vertex_descriptor.
> The final status is that Jeremiah verified that the code runs fine on gcc
> over the large input. Hence its quite possible a bug, and should be filed as
> such.

Just a minor thing -- Valgrind did show a buffer overrun in the code, and
I added asserts and showed that it writes beyond the end of a temporary
array, so it is definitely a bug in component_index.

> He suggested a workaround i.e. use ds.compress_sets() and then collect the
> component. So my problem is resolved.
>
>
> I also have issues with the documentation of incremental component which I
> shall post separately.

I would appreciate any comments on that you have. They should be a
separate ticket from the component_index one, though.

-- Jeremiah Willcock


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk