Re: [boost] [graph] transitive_closure_test failures on gcc-4.1.x and later

From: Markus Schöpflin (markus.schoepflin_at_[hidden])
Date: 2007-11-19 06:19:20

Aaron Windsor wrote:
> On Nov 12, 2007 7:21 AM, Markus Schöpflin <markus.schoepflin_at_[hidden]> wrote:
>> Hello,
>>
>> I just wanted to ask whether the authors of the graph library are aware of
>> these test failures. (See e.g. http://tinyurl.com/3dy79o)
>>
>> The compiler fails to find a suitable constructor for vertices.
>>
>> Regards,
>> Markus
> 
> I've just been looking at this test - I can't figure out why it would
> fail to compile. The actual test code (transitive_closure_test.cpp)
> starts off by including two files: depth_first_search.hpp and then
> vector_as_graph.hpp. The errors are complaining that
> depth_first_search can't find the function "vertices" specialized for
> a vector of vectors, but that function is supplied in
> vector_as_graph.hpp. If you switch the order of these two includes (or
> just leave off the depth_first_search include, since it's redundant -
> transitive_closure includes topological_sort, which includes
> depth_first_search), everything compiles and runs fine on gcc 4.1 and
> above.
> 
> Both depth_first_search and the vertices function are templated, and
> not instantiated until transitive_closure is called, so I'm at a loss
> to explain why the order of these two includes is causing gcc to get
> confused.

I don't think it is confused, it just checks these templates at the point 
of definition, and not at the point of instantiation.

> Does anyone object if I commit this fix (removing the
> depth_first_search include) to clear up these test failures?

Not me, this fix might even clear the failures for Tru64/CXX.

Thanks,
Markus