Workaround JuliaLang/julia#22907, indmax no longer returning a linear

index

src/ACME.jl

@@ -61,6 +61,15 @@ else
     end
 end
 
+function _indmax(a::AbstractMatrix)
+    ind = indmax(a)
+    if isa(ind, CartesianIndex) # since 0.7.0-DEV.1660
+        return (ind[1], ind[2])
+    else
+        return ind2sub(size(a), ind)
+    end
+end
+
 include("kdtree.jl")
 include("solvers.jl")
 
@@ -550,7 +559,7 @@ function model_matrices(circ::Circuit, t::Rational{BigInt})
         warn("State update depends on indeterminate quantity")
     end
     while size(nullspace, 2) > 0
-        i, j = ind2sub(size(nullspace), indmax(map(abs, nullspace)))
+        i, j = _indmax(map(abs, nullspace))
         nullspace = nullspace[[1:i-1; i+1:end], [1:j-1; j+1:end]]
         f = f[:, [1:j-1; j+1:end]]
         for k in [:fv; :fi; :c; :fq]
@@ -587,7 +596,7 @@ function tryextract(fq, numcols)
     for colcnt in 1:numcols
         # determine element with maximum absolute value in unprocessed columns
         # to use as pivot
-        i, j = ind2sub(size(fq), indmax(map(abs, fq[:,colcnt:end])))
+        i, j = _indmax(map(abs, fq[:,colcnt:end]))
         j += colcnt-1
 
         # swap pivot to first (unprocessed) column
@@ -1000,7 +1009,7 @@ function rank_factorize(a::SparseMatrixCSC)
     nullspace = gensolve(a', spzeros(size(a, 2), 0))[2]
     c = eye(eltype(a), size(a, 1))
     while size(nullspace, 2) > 0
-        i, j = ind2sub(size(nullspace), indmax(map(abs, nullspace)))
+        i, j = _indmax(map(abs, nullspace))
         c -= c[:, i] * nullspace[:, j]' / nullspace[i, j]
         c = c[:, [1:i-1; i+1:end]]
         nullspace -= nullspace[:, j] * vec(nullspace[i, :])' / nullspace[i, j]

src/kdtree.jl

@@ -38,7 +38,7 @@ function KDTree(p::AbstractMatrix, Np=size(p,2))
         return KDTree{typeof(cut_val),typeof(p)}(cut_dim, cut_val, [1], p)
     end
 
-    dim = indmax(var(p[:,1:Np],2))
+    dim = indmax(vec(var(p[:,1:Np],2)))
     p_idx = sortperm(vec(p[dim,:]))
 
     min_idx[1] = 1
@@ -56,7 +56,7 @@ function KDTree(p::AbstractMatrix, Np=size(p,2))
             min_idx[n] = cut_idx[parent_n]+1
             max_idx[n] = max_idx[parent_n]
         end
-        dim = indmax(var(p[:,p_idx[min_idx[n]:max_idx[n]]],2))
+        dim = indmax(vec(var(p[:,p_idx[min_idx[n]:max_idx[n]]],2)))
         idx = sortperm(vec(p[dim,p_idx[min_idx[n]:max_idx[n]]]))
         p_idx[min_idx[n]:max_idx[n]] = p_idx[idx + min_idx[n] - 1]
         cut_idx[n] = calc_cut_idx(min_idx[n], max_idx[n])

test/runtests.jl

@@ -88,7 +88,7 @@ end
 let ps = rand(6, 10000)
     t = ACME.KDTree(ps)
     p = rand(6)
-    best_p = ps[:,indmin(sum(abs2, broadcast(-, ps, p),1))]
+    best_p = ps[:,indmin(vec(sum(abs2, broadcast(-, ps, p),1)))]
     idx = ACME.indnearest(t, p)
     @test sum(abs2, p - best_p) ≈ sum(abs2, p - ps[:, idx])
 end