Allow MOI.ModelLike as the optimizer instead of MOI.AbstractOptimizer

[odow]

Closes jump-dev/MathOptInterface.jl#2417

The only downside is this:

julia> model = Model(MOI.FileFormats.MPS.Model)
A JuMP Model
Feasibility problem with:
Variables: 0
Model mode: AUTOMATIC
CachingOptimizer state: EMPTY_OPTIMIZER
Solver name: unknown

julia> @variable(model, x >= 0)
x

julia> optimize!(model)
ERROR: MethodError: no method matching optimize!(::MathOptInterface.Utilities.GenericModel{…})

Closest candidates are:
  optimize!(::Any, ::Any)
   @ MathOptInterface ~/.julia/packages/MathOptInterface/3JqTJ/src/MathOptInterface.jl:82
  optimize!(::MathOptInterface.Bridges.AbstractBridgeOptimizer)
   @ MathOptInterface ~/.julia/packages/MathOptInterface/3JqTJ/src/Bridges/bridge_optimizer.jl:378
  optimize!(::MathOptInterface.Utilities.CachingOptimizer)
   @ MathOptInterface ~/.julia/packages/MathOptInterface/3JqTJ/src/Utilities/cachingoptimizer.jl:302
  ...

Stacktrace:
 [1] optimize!(b::MathOptInterface.Bridges.LazyBridgeOptimizer{MathOptInterface.Utilities.GenericModel{…}})

It violates the MethodError principle because we're throwing an error deep in the stack.

Should we check whether the backend is an optimizer?

julia> model = Model(MOI.FileFormats.MPS.Model)
A JuMP Model
Feasibility problem with:
Variables: 0
Model mode: AUTOMATIC
CachingOptimizer state: EMPTY_OPTIMIZER
Solver name: unknown

julia> unsafe_backend(model)
A Mathematical Programming System (MPS) model

julia> unsafe_backend(model) isa MOI.AbstractOptimizer
false

[odow]

Now

julia> optimize!(model)
ERROR: Cannot call `optimize!` because the provided optimizer is not a subtype of `MOI.AbstractOptimizer`.
Stacktrace:

I didn't add the type because I don't know if this adds anything:

julia> print(typeof(unsafe_backend(model)))
MathOptInterface.Utilities.GenericModel{Float64, MathOptInterface.Utilities.ObjectiveContainer{Float64}, MathOptInterface.Utilities.VariablesContainer{Float64}, MathOptInterface.FileFormats.MPS.ModelFunctionConstraints{Float64}}

I guess I could show it instead

julia> sprint(show, unsafe_backend(model))
"A Mathematical Programming System (MPS) model"

[odow]

julia> optimize!(model)
ERROR: Cannot call `optimize!` because the provided optimizer is not a subtype of `MOI.AbstractOptimizer`.

The optimizer is:

A Mathematical Programming System (MPS) model

Stacktrace:
 [1] error(s::String)
   @ Base ./error.jl:35
 [2] optimize!(model::Model; ignore_optimize_hook::Bool, _differentiation_backend::MathOptInterface.Nonlinear.SparseReverseMode, kwargs::@Kwargs{})
   @ JuMP ~/.julia/dev/JuMP/src/optimizer_interface.jl:449
 [3] optimize!(model::Model)
   @ JuMP ~/.julia/dev/JuMP/src/optimizer_interface.jl:409
 [4] top-level scope
   @ REPL[55]:1

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (e4d2d7f) 98.29% compared to head (661dc0a) 98.32%.
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3667      +/-   ##
==========================================
+ Coverage   98.29%   98.32%   +0.03%     
==========================================
  Files          43       43              
  Lines        5676     5682       +6     
==========================================
+ Hits         5579     5587       +8     
+ Misses         97       95       -2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mlubin]

Looks pretty reasonable. Are there any other consequences of this change?

[odow]

Are there any other consequences of this change?

I don't think so. All of the solvers are AbstractOptimizer, so passing something that creates a ModelLike is non-trivial, and I expect that people who do so know what and why they're doing it.

But here's a extension-test to confirm: https://github.com/jump-dev/JuMP.jl/actions/runs/7804969693