deprecate diagm(A::SparseMatrixCSC) in favor of spdiagm(sparsevec(A))

[fredrikekre]

To be consistent with

julia> diagm(rand(3,1))
ERROR: MethodError: no method matching diagm(::Array{Float64,2})
[...]

julia> diagm(rand(1,3))
ERROR: MethodError: no method matching diagm(::Array{Float64,2})
[...]

Step 1 in JuliaLang/LinearAlgebra.jl#457

[Sacha0]

lgtm pending CI! :)

Edit: Perhaps add a NEWS.md entry?

[fredrikekre]

Following the discussion in JuliaLang/LinearAlgebra.jl#457 it looks like we will not get rid of spdiagm any time soon, and no matter what, this should actually be a method of spdiagm, rather than diagm. Perhaps with a specialized spdiagm_internal

julia/base/sparse/sparsematrix.jl

Lines 3302 to 3336 in 1323561

	function spdiagm_internal(B, d)
	ndiags = length(d)
	if length(B) != ndiags; throw(ArgumentError("first argument should be a tuple of length(d)=$ndiags arrays of diagonals")); end
	ncoeffs = 0
	for vec in B
	ncoeffs += length(vec)
	end
	I = Vector{Int}(ncoeffs)
	J = Vector{Int}(ncoeffs)
	V = Vector{promote_type(map(eltype, B)...)}(ncoeffs)
	id = 0
	i = 0
	for vec in B
	id += 1
	diag = d[id]
	numel = length(vec)
	if diag < 0
	row = -diag
	col = 0
	elseif diag > 0
	row = 0
	col = diag
	else
	row = 0
	col = 0
	end
	range = 1+i:numel+i
	I[range] = row+1:row+numel
	J[range] = col+1:col+numel
	copy!(view(V, range), vec)
	i += numel
	end
	return (I,J,V)
	end

for SparseVector. I will update the PR :)

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (master@8f5f981). Click here to learn what that means.
The diff coverage is n/a.

@@            Coverage Diff            @@
##             master   JuliaLang/julia#23341   +/-   ##
=========================================
  Coverage          ?   54.98%           
=========================================
  Files             ?      275           
  Lines             ?    51145           
  Branches          ?        0           
=========================================
  Hits              ?    28121           
  Misses            ?    23024           
  Partials          ?        0

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8f5f981...4e7da40. Read the comment docs.

[tkelman]

How does spdiagm compare in performance to the old specialized method?