Adjoint times Diagonal times Vector for n = 0

[ettersi]

# This works
julia> zeros(0)'*zeros(0)
0.0

# This does not, which is inconsistent
julia> zeros(0)'*Diagonal(zeros(0))*zeros(0)
ERROR: ArgumentError: reducing over an empty collection is not allowed
Stacktrace:
 [1] _empty_reduce_error() at ./reduce.jl:212
 [2] mapreduce_empty(::Function, ::Function, ::Type) at ./reduce.jl:246
 [3] mapreduce_empty_iter(::Function, ::Function, ::Base.Iterators.Zip{Tuple{Adjoint{Float64,Array{Float64,1}},Array{Float64,1},Array{Float64,1}}}, ::Base.HasEltype) at ./reduce.jl:254
 [4] mapfoldl_impl at ./reduce.jl:57 [inlined]
 [5] #mapfoldl#186 at ./reduce.jl:72 [inlined]
 [6] mapfoldl at ./reduce.jl:72 [inlined]
 [7] #mapreduce#194 at ./reduce.jl:200 [inlined]
 [8] mapreduce at ./reduce.jl:200 [inlined]
 [9] *(::Adjoint{Float64,Array{Float64,1}}, ::Diagonal{Float64,Array{Float64,1}}, ::Array{Float64,1}) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.3/LinearAlgebra/src/diagonal.jl:640
 [10] top-level scope at none:0

[MasonProtter]

The bug seems to come from 3 argument multiplication:

julia> Diagonal(zeros(0))*zeros(0)
0-element Array{Float64,1}

julia> zeros(0)'Diagonal(zeros(0))*zeros(0)
0.0

julia> zeros(0)'*Diagonal(zeros(0))*zeros(0)
ERROR: ArgumentError: reducing over an empty collection is not allowed
Stacktrace:
 [1] _empty_reduce_error() at ./reduce.jl:295
 [2] mapreduce_empty(::Function, ::Base.BottomRF{typeof(+)}, ::Type{T} where T) at ./reduce.jl:334
 [3] reduce_empty(::Base.MappingRF{LinearAlgebra.var"#160#161",Base.BottomRF{typeof(+)}}, ::Type{T} where T) at ./reduce.jl:321
 [4] reduce_empty_iter at ./reduce.jl:347 [inlined]
 [5] reduce_empty_iter at ./reduce.jl:346 [inlined]
 [6] foldl_impl at ./reduce.jl:46 [inlined]
 [7] mapfoldl_impl at ./reduce.jl:41 [inlined]
 [8] #mapfoldl#189 at ./reduce.jl:157 [inlined]
 [9] mapfoldl at ./reduce.jl:157 [inlined]
 [10] #mapreduce#193 at ./reduce.jl:283 [inlined]
 [11] mapreduce at ./reduce.jl:283 [inlined]
 [12] *(::Adjoint{Float64,Array{Float64,1}}, ::Diagonal{Float64,Array{Float64,1}}, ::Array{Float64,1}) at /home/mason/julia/usr/share/julia/stdlib/v1.4/LinearAlgebra/src/diagonal.jl:647
 [13] top-level scope at REPL[7]:1
 [14] eval(::Module, ::Any) at ./boot.jl:331
 [15] eval_user_input(::Any, ::REPL.REPLBackend) at /home/mason/julia/usr/share/julia/stdlib/v1.4/REPL/src/REPL.jl:86
 [16] run_backend(::REPL.REPLBackend) at /home/mason/.julia/packages/Revise/Pcs5V/src/Revise.jl:1073
 [17] top-level scope at none:0

[dkarrasch]

Yes

julia> @which zeros(0)'Diagonal(zeros(0))*zeros(0)
*(u::Adjoint{T,#s662} where #s662<:(AbstractArray{T,1} where T) where T, v::AbstractArray{T,1} where T) in LinearAlgebra at /Applications/Julia-1.4.app/Contents/Resources/julia/share/julia/stdlib/v1.4/LinearAlgebra/src/adjtrans.jl:245

julia> @which zeros(0)'*Diagonal(zeros(0))*zeros(0)
*(x::Adjoint{#s662,#s661} where #s661<:(AbstractArray{T,1} where T) where #s662, D::Diagonal, y::AbstractArray{T,1} where T) in LinearAlgebra at /Applications/Julia-1.4.app/Contents/Resources/julia/share/julia/stdlib/v1.4/LinearAlgebra/src/diagonal.jl:647

The diagonal.jl version (the second) needs to be expanded to handle the length-0 case. I'm a bit surprised though that the first is not parsed like the second...