Add TrajkovicKeypoint2D/3D to CMake build

[ThorstenHarter]

The files which have been added to PCL in #409 are missing in the CMakeLists.txt of the keypoints module.

Fix Microsoft Visual Studio 2015 compile error: "trajkovic_3d.hpp(235): error C3016: 'i': index variable in OpenMP 'for' statement must have signed integral type" by changing index type from size_t to int.

[SergioRAgostinho]

Have a look at

https://github.com/PointCloudLibrary/pcl/blob/master/common/include/pcl/pcl_macros.h#L44-L55
http://www.boost.org/doc/libs/1_55_0/boost/cstdint.hpp

The boost provided intmax_t might be the safest option in this scenario.

[ThorstenHarter]

You mean like this?

  for (intmax_t i = 0; i < indices.size (); ++i)
  {
    int idx = indices[i];

I automatically added the casts because we use -Wsign-conversion in our projects, but since this isn't activated in PCL, they are not mandatory.

[SergioRAgostinho]

You still need to cast indices.size() like you did before because of what of Wsign-comparison. It is active in PCL for clang (we do need to revise/unify our flags across different compilers). So please suppress all of those.

The point behind my comment was to make sure we use the highest range integer possible, that's available on each platform, to avoid potential overflows on very big indices vectors. int didn't ensure that but intmax_t seems to.

[ThorstenHarter]

So you're saying a PointCloud could have more than INT_MAX points?
Because only then the indices vector of the keypoints search could be bigger than that.

In that case the filters, features and keypoints algorithms won't work anyway, because the indices are always defined like this: std::vector<int> indices

In other words, I don't see the sense in changing the type of 'i' to intmax_t, while 'idx', 'indices' etc. are using the type int.

[SergioRAgostinho]

That's a valid point. I regret a little what our "big data" users deal with. Hopefully someone with those needs will have a look at things and decide to push some changes.