I am grudgingly thankful to CRAN for catching these errors: https://www.r-project.org/nosvn/R.check/r-devel-linux-x86_64-fedora-clang/mlpack-00install.html

The actual error is:

../inst/include/mlpack/core/tree/build_tree.hpp:21:11: note: candidate template ignored: substitution failure [with TreeType = Tree]: ambiguous partial specializations of 'TreeTraits<mlpack::BinarySpaceTree<mlpack::LMetric<2>, mlpack::NeighborSearchStat<mlpack::FurthestNS>, arma::Mat<double>, mlpack::CellBound, mlpack::UBTreeSplit>>'
   21 | TreeType* BuildTree(
      |           ^
   22 |     MatType&& dataset,
   23 |     std::vector<size_t>& oldFromNew,
   24 |     const std::enable_if_t<TreeTraits<TreeType>::RearrangesDataset>* = 0)

It appears that clang does not like the series of partial template specializations that we use for different TreeTraits. I worked around this issue by, for each individual tree type, making only one specialization of TreeTraits, and using other traits classes internally (like IsRPTreeSplitType<> and a few others I added) to set the boolean values in TreeTraits appropriately, depending on the configuration of the tree.

There is no actual behavior change, this is just a refactoring that makes compilation work on newer clang versions.

I also fixed a few unused variable warnings and other warnings that clang emitted.