Virtual octreedatanode

[lorenzoriano]

Subclassing OcTreeNode is still problematic, since no method is virtual and the class is not polymorphic. The main problem is with static_cast in OcTreeNode::createChild, where any information about a subclass is lost. Effectively creating a subclass is still not very easy, as a number of methods need to be rewritten (including the copy constructor) but these changes allow for polymorphism, thus easier extension of the node.

This pull request should solve issue #93 and potentially #95

Merry Christmas! 😄 🎄

[ahornung]

Hi Lorenzo!

Indeed, polymorphism would make many things easier. However, the cost would be one pointer for each node object for the vtable, as mentioned in our OctoMap paper (Hornung et al. 2013). This is a significant increase particularly for the leaf nodes and usually not needed because the single nodes of one tree class can't change during runtime.

That's why the implementation explicitly does not use virtual inheritance for the nodes. I created a new issue (#98) to track improvements to the current situation, although I don't have a clear idea yet how they can look like. Maybe there are some implementation tricks that could help, otherwise it'll be with a better documentation.

[ahornung]

...and Merry Christmas as well, thanks! 🎄 🎅