Merge branch 'fix-attribute-types'

src/cloud.jl

@@ -44,14 +44,14 @@ nearby = cloud[inrange(cloud, [1,1,1], 5.0)]
 type PointCloud{Dim,T,SIndex}
     positions::Vector{SVector{Dim,T}}
     spatial_index::SIndex
-    attributes::Dict{Symbol,Vector{Any}}
+    attributes::Dict{Symbol,Vector}
 end
 
 # Create a `PointCloud` for FixedSizeArrays Vec using a KDTree for spatial
 # indexing based on `positions`.
 # TODO should be able to change the spatial index
-function PointCloud{T <: AbstractFloat}(positions::Vector{SVector{3, T}})
-    PointCloud(positions, KDTree(positions), Dict{Symbol,Vector{Any}}())
+function PointCloud{Dim, T <: AbstractFloat}(positions::Vector{SVector{Dim, T}})
+    PointCloud(positions, KDTree(positions), Dict{Symbol,Vector}())
 end
 
 # Create a `PointCloud` from an 3xN array of points, using a KDTree for spatial
@@ -121,7 +121,7 @@ endof(cloud::PointCloud) = length(cloud.positions)
 function getindex{Dim,T,SIndex}(cloud::PointCloud{Dim,T,SIndex}, row_inds::AbstractVector)
     pos = positions(cloud)[row_inds]
     tree = KDTree(pos)  # TODO NEED TO GET SPATIAL INDEX FROM TYPE
-    attrs = Dict{Symbol,Vector{Any}}()
+    attrs = Dict{Symbol,Vector}()
     for (k,v) in cloud.attributes
         attrs[k] = v[row_inds]
     end
@@ -130,7 +130,13 @@ end
 
 # Concatenate point clouds
 function vcat{Dim,T,SIndex}(cloud1::PointCloud{Dim,T,SIndex}, clouds::PointCloud{Dim,T,SIndex}...)
-    attrs = deepcopy(cloud1.attributes)
+    # Workaround for a weird compiler bug, which causes the deepcopy type assert to fail.
+    # See https://github.com/JuliaLang/julia/issues/19041
+    #attrs = deepcopy(cloud1.attributes)
+    attrs = Dict{Symbol,Vector}()
+    for (k,v) in cloud1.attributes
+        attrs[k] = deepcopy(v)
+    end
     pos = positions(cloud1)
     ks = Set(keys(attrs))
     for cloud in clouds
@@ -180,19 +186,21 @@ Normals are returned in the 3xN array `normals`.
 
 TODO: Need a generalized version of this for the PCA covariance computation.
 """
-function pca_normals!{D, T}(normals, query_points::Vector{SVector{D,T}}, cloud; neighbours::Int=10)
-    cov_tmp = MMatrix{3, 3, T}()
+function pca_normals!{V<:AbstractVector}(normals::Vector{V}, query_points, cloud; neighbours::Int=10)
+    position = positions(cloud)
     for i = 1:length(query_points)
         @inbounds (inds, _) = knn(cloud, query_points[i], neighbours)
         isempty(inds) && continue
-        fill!(cov_tmp, zero(T))
+        cov_tmp = zeros(SMatrix{3,3,eltype(V)})
         for j = 1:length(inds)
-            @inbounds p = query_points[inds[j]] - query_points[i]
-            cov_tmp .+= p * p'
+            @inbounds p = position[inds[j]] - query_points[i]
+            cov_tmp += p * p'
         end
+        # FIXME: Avoid converting to Matrix when eig(SMatrix) is solid
+        # See https://github.com/JuliaArrays/StaticArrays.jl/issues/80
+        evals, evecs = eig(Symmetric(Matrix(cov_tmp)))
         # Select evec of smallest eval
-        evals, evecs = eig(Symmetric(cov_tmp), 1:1)
-        normals[i] = evecs
+        normals[i] = evecs[:,1]
     end
     return normals
 end