qsz13 / igraphhpp Goto Github PK

* Add documentation
* Add test cases
* Use a consistence default value

What steps will reproduce the problem?
Compiling the attached file. located at resize() and remove() function

What is the expected output? What do you see instead?
../igraph/cpp/impl/referencevector.cpp:168: warning: deleting ‘void*’ is
undefined
../igraph/cpp/impl/referencevector.cpp:146: warning: deleting ‘void*’ is
undefined

Compiler: g++4.4 

The following problems need some change of interface to resolve

1. There is ambiguity between two constructors:

ReferenceVector(const long count) MAY_THROW_EXCEPTION;
ReferenceVector(const long count, ...) MAY_THROW_EXCEPTION;

2. There is problem while calling push_back. The first take const input and
past to second which take non-const input! And it willl remove the ownership

ReferenceVector<T>& push_back(const_reference obj, const
::tempobj::OwnershipTransfer transfer = ::tempobj::OwnershipTransferCopy)
MAY_THROW_EXCEPTION { return push_back(&obj, transfer); }

ReferenceVector<T>& push_back(pointer e, const ::tempobj::OwnershipTransfer
transfer = ::tempobj::OwnershipTransferMove) MAY_THROW_EXCEPTION;

As titled. The internal ones give multigraphs which are not suitable for our 
need.

Wrapping the disjoint union, union and intersection for igraph. The
operator interface should also be supported.

While calling the constructor of the vector with size 0, it gives an
ambiguous result. Is it appropriate? Or can it be solved.

Vector p(0);

mincut_test.cpp:35: error: call of overloaded ‘BasicVector(int)’ is 
ambiguous


Also, the default constructor seems give a segfault.

Vector p;
p.print();


I think it should be mentioned in the documentation. Or adding a null state
vector support.

Can it have a stable branch? The igraphhpp is basically sufficient at that
time, but the recent trunk can not be compiled at all. We need to use it
while develop it. 

The test should be automated. Maybe the make file can be written to test
different platform and flags(g++-4.2, g++-4.4, -std=gnu++0x) at once. This
can be schedule to run in a computer periodically. I am not skillful on
make file yet.

Reasons:
(1) g++-4.3's C++0x does not support initialization list nor deleted functions, 
which always causes 
errors. That's not a regression at all.
(2) No one uses C++0x mode in GCC 4.3 (I have 4.2, you have 4.4, MOSIX has 4.1.)

As title, it should not be the default behaviour. Maybe a reference class
is required to solve this problem.

What steps will reproduce the problem?
compile in the path igraphhpp/tests/ with 

g++-4.4 -std=gnu++0x -Wall -Wno-unknown-pragmas -O2 -ligraph -lgsl
-lgslcblas -I../ -I/opt/local/include -L/opt/local/lib -Wno-attributes
-fno-strict-aliasing xxx.cpp

The Graph::mincut is added recently. It returns a Mincut object and the
usage pattern is 

Graph g = Graph::full(4);

Vector p(1);
g.mincut().get_partition1(p).compute();

Is this interface appropriate compared with the Community one? It use the
mechanism of first construct a object, so that the variable can be set. The
computation is then carried out (or done when the destructor is called).

For example, I wanna know why int_ptr_overload.cpp  pass in MSVC, but it isn't 
on svn yet.

The current igraph_cliques.cpp will result in segfault because
XXINTRNL_TEMP_RETURN_PTRVEC is improperly used.

Note that if you want to return a ReferenceVector of an internal igraph
type, you must use it like this:

igraph_vector_ptr_t res
...
return ReferenceVector<Cpp_Type>::adopt<C_Type>(res);

Please rewrite XXINTRNL_TEMP_RETURN_PTRVEC to use this instead.

(I've reverted those for cliques back to r85 to make igraph_cliques.cpp
works properly, please check the rest yourself.)

What steps will reproduce the problem?
1. compile attached by 

g++ -I ~/svn/igraphhpp/ bugs.cpp -ligraph

What is the expected output? What do you see instead?
It gives Segmentation fault, possibly located at GraphWrite, occurs while
destroying