Re: [Boost-Users] Const Graph Property Maps

From: Douglas Gregor (gregod_at_[hidden])
Date: 2003-03-23 13:12:29

On Friday 21 March 2003 09:55 pm, dmatt001 wrote:
> Hi Guys,
>
> I have three interesting questions in regard to graph properties.
> Please assume Boost 1_29_0. I'm a little slow in upgrading ... ;-)
>
> 1. If I define:
>
>   MyGraph_t const myGraph;
>
> Why cannot I get a property map to access my property:
>
>  typedef boost::property_map<MyGraph_t, MyPropTag_t>::type MyPropMap_t;
>  MyPropMap_t myPropMap = get(MyPropTag_t(), myGraph);

You want:
  typedef boost::property_map<MyGraph_t, MyPropTag_t>::const_type MyPropMap_t;

> 2. Why not use a copy constructor if it exists when copying graph
> elements? Is the existence of a copy constructor a hard-yard to verfiy
> with MPL?

I don't know the rationale behind the decision, but I do know that it is 
_very_ hard to determine if there is a copy constructor in C++ because they 
are implicitly generated. The question isn't "is there a copy constructor" 
but "is there an _accessible_ copy constructor" and, unfortunately, we can't 
test accessibility in a metaprogram :(.

> 3. I noticed that remove_edge(...) complains in regard to a missing
> default constructor of a property object. Don't get me wrong; this
> happened during an editing mishap and I know that we need regular
> types here. Why is this required if you are removing an element?

The answer is probably linked to the answer to #2: if copy constructors aren't 
used, then edges (along with their internal properties) are copied by 
defalut-constructing the target then calling a function to perform the copy.

        Doug