Scope of index variables

[allbits]

m = Model()
@variable(m,x[1:5])
@constraint(m,[m=1:4],x[m]==x[m+1])

complains that there's no addconstraint method with an Int64 as the first argument. Should this work?

@constraint(m,[i=1:4],x[i]==x[i+1])

is fine.

[mlubin]

Should this work?

I'm struggling to see the benefit of allowing this to work.

[allbits]

I'm struggling to see the benefit of allowing this to work.

There's probably little benefit, since you always can change the variable. I see this more as a question of clear scoping rules. In
@constraint(m,[m=1:4],x[m]==x[m+1])
it seems natural to read "[m=1:4]" as a universal quantifier and expect m to be bound to 1:4 in the scope of the restrictions. Or consider:
@constraint(m,sum{x[m],m=1:5}==1)
where it seems natural to expect m to be bound to 1:5 in the scope of the sum.

[mlubin]

I'm not opposed to making things more consistent but I don't expect to spend any time fixing this soon. PRs will be considered.

[mlubin]

Possibly addressed in Julia 0.7 by JuliaLang/julia#22314

[mlubin]

This is not addressed by JuliaLang/julia#22314.

@constraint(m,[i=1:4],x[i]==x[i+1])

is translated essentially to:

for i in 1:4
    @constraint(m, x[i] == x[i + 1])
end

We would need to mangle the iteration variables to avoid the name conflict. This is feasible to do but still not high priority. The only cases I've seen where users run into this issue is when the model is named m. We've stopped recommending this name for the model.

[odow]

Rather than make this work, we should just throw an error if the index set has the same name as the model.