Error when parsing an inner product in an expression

[mtanneau]

It seems that an inner product between two vectors is not identified as a scalar when parsed by JuMP.

using LinearAlgebra
using JuMP

u = ones(1)
v = ones(1)

m = Model()

@variable(m, x)
@constraint(m, (1.0 + u'v) * x == 0.0)  # u, v are vectors so u'v is a scalar

raises the error

ERROR: MethodError: no method matching *(::Adjoint{Float64,Array{Float64,1}}, ::VariableRef)
Closest candidates are:
*(::Any, ::Any, ::Any, ::Any...) at operators.jl:502
*(::Adjoint{T,#s623} where #s623<:(AbstractArray{T,1} where T) where T, ::Adjoint{#s623,#s622} where #s622<:(Transpose{T,#s623} where #s623<:(AbstractArray{T,1} where T) where T) where #s623) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.1/LinearAlgebra/src/adjtrans.jl:215
*(::Adjoint{T,#s623} where #s623<:(AbstractArray{T,1} where T) where T, ::Adjoint{#s623,#s622} where #s622<:LinearAlgebra.AbstractTriangular where #s623) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.1/LinearAlgebra/src/triangular.jl:1954

Note that the two following lines do not raise an error

@constraint(m, (1.0 + dot(u, v)) * x == 0.0)

and

(1.0 + u'v) * x  # returns 2 x

[odow]

What's weirder is that:

m = Model()
@variable(m, x)
u = v = ones(1)
@constraint(m, u' * v * x == 0)    # Works
@constraint(m, (u' * v) * x == 0)  # Doesn't work
@constraint(m, x * (u' * v) == 0)  # Doesn't work

Tidying up @macroexpand @constraint(m, (u' * v) * x == 0) shows that it gets parsed as

ssa_0 = (JuMP.Val){false}()
ssa_1 = (JuMP._destructive_add_with_reorder!)(ssa_0, u', x, v)
ssa_2 = (JuMP._destructive_add_with_reorder!)(ssa_1, -1.0, 0)
ssa_3 = (JuMP._destructive_add_with_reorder!)(ssa_2, 0)

when it should be parsed as

ssa_0 = (JuMP.Val){false}()
ssa_1 = (JuMP._destructive_add_with_reorder!)(ssa_0, u', v, x)
ssa_2 = (JuMP._destructive_add_with_reorder!)(ssa_1, -1.0, 0)
ssa_3 = (JuMP._destructive_add_with_reorder!)(ssa_2, 0)