ModelingToolkit + IntervalConstraintProgramming: how to build expressions and contractors without "world age" problems or ambiguities

General Usage
1

Hi,

I’m facing a problem of “world age” that my Julia experience doesn’t allow me to comprehend very well. Here is a MWE that uses the ModelingToolkit and IntervalConstraintProgramming packages:

using ModelingToolkit
using IntervalArithmetic
using IntervalConstraintProgramming

# dynamic reformulation of the expression by adding a slack
function add_slack(constraint_expression)
    new_variable = @variables s
    return (constraint_expression - s, new_variable[1])
end

# contract the box wrt the constraint
function contract(box, constraint_expression, variables)
    contractor = Contractor(variables, constraint_expression)
    box = contractor(0..0, box)
    return box
end


function constraint(x)
    return x[1]*x[2]
end

# create the original variables
variables = @variables x[1:2]
# flatten to an array
variables = vcat(variables...)

# add the slack
constraint_expression = constraint(x)
(constraint_expression, new_variable) = add_slack(constraint_expression)
# create the array [x1, x2, s]
append!(variables, new_variable)

# create a box and contract it wrt constraint
box = IntervalBox(0..1, 0..1, 0..1)
box = contract(box, constraint_expression, variables)

I define a problem, reformulate it somehow (here I add a slack variable, but I could add arbitrarily many variables), generate a contractor and call it.
I get this error:

ERROR: LoadError: MethodError: no method matching (::IntervalConstraintProgramming.var"#31#32")(::IntervalBox{3,Float64})
The applicable method may be too new: running in world age 27984, while current world is 27986.

Somehow my logic is not Julia-compliant :slight_smile: I must say that the scope of the ModelingToolkit variables eludes me.
Can you help me out? If possible, I’d like to keep the same structure (problem definition + reformulation + creation of contractor independently).
Thanks,

Charlie

2

Whatever is evaling in there, make it not use eval and instead use Runtime generated functions.

3

As Chris suggests, the problem is that Contractor is using eval to evaluate an expression, but then you are immediately calling contractor within the same function.

To avoid the world-age problem you need to return to the “top level” before using a function created by eval. Maybe we should indeed be using RuntimeGeneratedFunctions instead of eval.

1 Like
4

You should :wink:

1 Like
5

Thanks for your answers! I tried something like this:

using ModelingToolkit
using IntervalArithmetic
using IntervalConstraintProgramming

function add_slack(constraint_expression)
    new_variable = @variables s
    return (constraint_expression - s, new_variable[1])
end

function create_contractor(variables, constraint_expression)
    return Contractor(variables, constraint_expression)
end

function constraint(x)
    return x[1]*x[2]
end

function main()
    # create the original variables
    variables = @variables x[1:2]
    # flatten to an array
    variables = vcat(variables...)

    # add the slack
    constraint_expression = constraint(x)
    (constraint_expression, new_variable) = add_slack(constraint_expression)
    # create the array [x1, x2, s]
    append!(variables, new_variable)

    # create a box and contract it wrt constraint
    box = IntervalBox(0..1, 0..1, 0..1)
    contractor = create_contractor(variables, constraint_expression)
    box = contractor(0..0, box)
    println(box)
    return box
end

main()

with the same result. What happens is absolutely not intuitive for me. The Contractor is now created and called in 2 different functions :thinking:

6

That’s still the same world age. You want to call contractor via Base.invokelatest(contractor, 0..0, box), but note that this will infer to Any and have some overhead.

7

Or just don’t put it in a main function.

But @dpsanders knows what’s up and this’ll presumably get fixed in the package soon.

2 Likes
8

You’re still inside the main() function. You don’t get back to top level (global scope).

9

Got it this time :slight_smile: Thanks a lot!

1 Like