Deleting variables does not work when a constraint evaluates a variable at multiple time points

[vyudu]

Describe the bug
In the following example the constraint evaluates Q[1] at 0.5 and 0.6. In this case deleting the variable Q[1] does not delete all the references to it:

To Reproduce
Please provide a minimum working example that causes this behavior (see how here). Do NOT copy your entire script, this makes debugging very difficult. Also, include a copy of the error you encounter if applicable.

using InfiniteOpt, JuMP [plus other packages]

model = InfiniteModel()
@infinite_parameter(model, t in [0, 10], num_supports = 100)
@variable(model, Q[1:3], Infinite(t))
@constraint(model, Q[1](0.6) == Q[1](0.5))
all_variables(model) #works
delete(model, Q[1])
all_variables(model) # throws an error

Error showing value of type Vector{GeneralVariableRef}:

SYSTEM (REPL): showing an error caused an error
ERROR: Invalid infinite variable reference, cannot find corresponding variable in the model. This is likely caused by using the reference of a deleted variable.
Stacktrace:
  [1] print_response(errio::IO, response::Any, backend::Union{Nothing, REPL.REPLBackendRef}, show_value::Bool, have_color::Bool, specialdisplay::Union{Nothing, AbstractDisplay})
    @ REPL ~/.julia/juliaup/julia-1.11.5+0.aarch64.apple.darwin14/share/julia/stdlib/v1.11/REPL/src/REPL.jl:446
  [2] (::REPL.var"#70#71"{REPL.LineEditREPL, Pair{Any, Bool}, Bool, Bool})(io::Any)
    @ REPL ~/.julia/juliaup/julia-1.11.5+0.aarch64.apple.darwin14/share/julia/stdlib/v1.11/REPL/src/REPL.jl:405
  [3] with_repl_linfo(f::Any, repl::REPL.LineEditREPL)
    @ REPL ~/.julia/juliaup/julia-1.11.5+0.aarch64.apple.darwin14/share/julia/stdlib/v1.11/REPL/src/REPL.jl:678
  [4] print_response(repl::REPL.AbstractREPL, response::Any, show_value::Bool, have_color::Bool)
    @ REPL ~/.julia/juliaup/julia-1.11.5+0.aarch64.apple.darwin14/share/julia/stdlib/v1.11/REPL/src/REPL.jl:403
  [5] (::REPL.var"#do_respond#100"{Bool, Bool, REPL.var"#116#134"{…}, REPL.LineEditREPL, REPL.LineEdit.Prompt})(s::REPL.LineEdit.MIState, buf::Any, ok::Bool)
    @ REPL ~/.julia/juliaup/julia-1.11.5+0.aarch64.apple.darwin14/share/julia/stdlib/v1.11/REPL/src/REPL.jl:1035
  [6] #invokelatest#2
    @ ./essentials.jl:1055 [inlined]
  [7] invokelatest
    @ ./essentials.jl:1052 [inlined]
  [8] run_interface(terminal::REPL.Terminals.TextTerminal, m::REPL.LineEdit.ModalInterface, s::REPL.LineEdit.MIState)
    @ REPL.LineEdit ~/.julia/juliaup/julia-1.11.5+0.aarch64.apple.darwin14/share/julia/stdlib/v1.11/REPL/src/LineEdit.jl:2755
  [9] run_frontend(repl::REPL.LineEditREPL, backend::REPL.REPLBackendRef)
    @ REPL ~/.julia/juliaup/julia-1.11.5+0.aarch64.apple.darwin14/share/julia/stdlib/v1.11/REPL/src/REPL.jl:1506
 [10] (::REPL.var"#79#85"{REPL.LineEditREPL, REPL.REPLBackendRef})()
    @ REPL ~/.julia/juliaup/julia-1.11.5+0.aarch64.apple.darwin14/share/julia/stdlib/v1.11/REPL/src/REPL.jl:497
Some type information was truncated. Use `show(err)` to see complete types.

Expected behavior
A clear and concise description of what you expected to happen.

Desktop (please complete the following information):

• OSX
• Package Versions: InfiniteOpt 0.5.9, JuMP 1.25.0
• Julia Version:1.11

Additional context
Add any other context about the problem here.

[pulsipher]

Thanks for reporting. This will be fixed by #377.

What is your use case for deletion? Usually there is a better way.

[vyudu]

Many thanks for the quick fix! My use case is this: I'm working on writing an interface from ModelingToolkit to JuMP/InfiniteOpt for optimal control problems SciML/ModelingToolkit.jl#3549. Whenever a problem gets made and solve is called, the ODE solver is baked in by adding collocation constraints to the system. I want to delete any previous collocation constraints whenever the user calls solve (possibly with a different solver as before), so they can call solve on the same problem multiple times without having to remake the problem. (In the implicit method case, I add another variable K to the system as well that is a stand-in for the intermediate points to solve for at each timestep, so I want to delete this from the model when I call solve again as well.)

[pulsipher]

Taking a quick look at the code for SciML/ModelingToolkit.jl#3549, it looks like you are setting up the variables/constraints in question in add_jump_solve_constraints!. It also appears that this is seeking to manually setup your own collocation scheme (i.e., not using one of InfiniteOpt's built-in discretization schemes). Is that correct?

If so, it seems like what you would really want to do is setup your own collocation scheme via InfiniteOpt's extendable backend (see https://infiniteopt.github.io/InfiniteOpt.jl/stable/develop/extensions/#Derivative-Evaluation-Methods). Using this API would be significantly more performant and would avoid the need to delete variables/constraints.