Incorrect parsing/evaluation of NLExpressions

[anriseth]

Using a NLExpr in the definition of another NLExpr does not seem to work correctly all the time.
I've tried to define an expression $v(x)^TAv(x)$, where $v$ is a nonlinear vector field in $x$ and $A$ is a matrix. In the example below, you can see that the evaluation of the functions f2 and f3 are different, while I expect it to be the same.

using JuMP
n = 2
x0 = [2.25, 2.4]
β = [-2 0.15; 0.23 -2.1]
Σ = [0.02 0.007; 0.007 0.014]
x0 = [1.5, 2.0]
m = Model()
@defVar(m, x[i=1:n] >= 0, start=x0[i])

@defNLExpr(f1[i=1:n], prod{x[j]^β[j,i], j = 1:n})
@defNLExpr(f2, sum{Σ[i,j]*f1[i]*f1[j], i=1:n, j=1:n}) # bug
@defNLExpr(f3, sum{Σ[i,j]*prod{x[k]^β[k,i], k = 1:n}*prod{x[k]^β[k,j], k = 1:n}, i=1:n, j=1:n})

println(getValue(f2))
println(getValue(f3))

Output:

0.0021786119842496466
0.008103475980157478

[mlubin]

Looks like there's a scoping issue. The j in f1 is the same symbol as the j in f2. Definitely a bug, but the easy workaround is to use a different indexing variable in f1.

[anriseth]

Thanks, I didn't think about that. @defNLExpr(f1[i=1:n], prod{x[k]^β[k,i], k = 1:n}) works.

[mlubin]

Will be fixed by #638, confusion of scope is not possible anymore