[boost] Re: [sanbox] container_algorithms problem

From: David Abrahams (dave_at_[hidden])
Date: 2003-10-17 08:56:44

Vladimir Prus wrote:

> Hi,
> I've just updated my copy of boost-sandbox and start to get the following 
> error from g++ 3.3:
   ^^^^^
That's a warning, not an error.

> /space/NM/boost-sandbox/boost/sequence_algo/container_algo.hpp: In function `
>    typename boost::container_traits<Container>::const_iterator 
>    boost::find(const Container&, const T&) [with Container = std::vector<int, 
>    std::allocator<int> >, T = int]':
> .../../lib/lvk/program_analysis/dataflow.cpp:145:   instantiated from here
> /space/NM/boost-sandbox/boost/sequence_algo/container_algo.hpp:150: warning: 
> cannot
>    pass objects of non-POD type `struct 
>    boost::contiguous_sequence_container_tag' through `...'; call will abort at 
>    runtime
> 
> When I compile the container_algo.cpp test program it produces lots of 
> similiar warnings, after which I get:
> 
> bash-2.05b$ ./a.out
> Segmentation fault
> 
> Seems like the warning is correct in predicting crash?

No, the function is never called.

> I suggest that instead of using "tag" types for determining the right function 
> to call, the pointer to "tag" is used. I.e. for one fragment one would get:
> 
>     template< typename C, typename T >
>     inline typename container_traits<C>::iterator 
>     find_impl( C& c, const T& value, associative_container_tag* ) ...
> 
>     template< typename C, typename T >
>     inline typename container_traits<C>::iterator 
>     find_impl( C& c, const T& value, ... ) .....

I don't know if that change is applicable in this case, but in general 
it is not, since the type to be passed may be a reference type.

> When I make the change, the test passes. The patch attached. It also fixes 
> several missing typenames, but the problem is that some of the lines are now 
> way longer than 80 symbols.
> 
> - Volodya

-Dave