Run all tests against bridged SDPA model

[blegat]

Closes #2375

When looking over the failing tests of https://github.com/blegat/SDPLR.jl/, I noticed some were due to MOI bugs. When looking at how to reproduce them, I found out that we could have detected them by running against all the tests. Looking back at the history, it seems I initially just called nametest because it wasn't so easy to run all the tests. When @odow migrated to the new test infrastructure, he used include = ["ConstraintName", "VariableName"], so that's its equivalent to nametest. This did not take the opportunity brought by this new test infrastructure to run it against all the tests. These tests are usually failing for all solvers so fixing them should help many solver wrappers at the same time.

This brings up a few bugs that I suggest we first fix in separate PRs with their own tests:

GeometricSDPA (so it should help SCS, ECOS, COSMO, SeDuMi, SDPT3, CDCS, Clarabel, ...)

• test_conic_SecondOrderCone_INFEASIBLE Fix RSOCtoPSD bridge for dimension 2 #2359
• test_model_LowerBoundAlreadySet [Bridges] fix BoundAlreadySet errors #2376
• test_model_UpperBoundAlreadySet [Bridges] fix BoundAlreadySet errors #2376

StandardSDPA (so it should help CSDP, SDPA, SDPLR, DSDP, ...)

• test_basic_VectorOfVariables_Nonpositives [Bridges] Fix ConstraintIndex conflicts between variable and constraint bridges #2362
• test_basic_VectorOfVariables_SecondOrderCone [Bridges] Fix ConstraintFunction getter of conversion bridge with small rounding errors #2360
• test_conic_SecondOrderCone_INFEASIBLE Fix RSOCtoPSD bridge for dimension 2 #2359
• test_model_ListOfVariablesWithAttributeSet [Bridges] Fix ListOfVariablesWithAttributeSet #2380
• test_model_LowerBoundAlreadySet [Bridges] Throw for bound already set on bridged variable #2383
• test_model_UpperBoundAlreadySet [Bridges] Throw for bound already set on bridged variable #2383
• test_model_ScalarFunctionConstantNotZero [Bridges] Throw in Constraint.SlackBridge if scalar constant is nonzero #2382
• test_model_delete [Bridges] Fix deletion corner cases #2377
• test_multiobjective_vector_nonlinear, test_multiobjective_vector_nonlinear_delete, test_multiobjective_vector_nonlinear_delete_vector, test_multiobjective_vector_nonlinear_modify: missing isapprox(::MOI.ScalarAffineFunction{Float64}, ::MOI.VariableIndex)
• test_multiobjective_vector_of_variables, test_multiobjective_vector_of_variables_delete_all, test_multiobjective_vector_of_variables_delete, test_multiobjective_vector_of_variables_delete_vector: missing substitute_variables(_, ::MOI.VectorOfVariables)

[odow]

test_multiobjective_vector_nonlinear I would have expected all of these to be skipped

[blegat]

They are skipped for the SDPA, CSDP, SDPLR, ... solvers but not for the tests in lazy_bridge_optimizer since we don't exclude it:

MathOptInterface.jl/test/Bridges/lazy_bridge_optimizer.jl

Lines 365 to 372 in 2fb6eb2

	function MOI.supports(
	::StandardSDPAModel{T},
	::MOI.ObjectiveFunction{
	<:Union{MOI.VariableIndex,MOI.ScalarQuadraticFunction{T}},
	},
	) where {T}
	return false
	end

[odow]

Oh. Let's just exclude it. They're not really meant to be bridged like this just yet.

[odow]

I'm taking a look at some of the remaining issues

[odow]

test_model_ListOfVariablesWithAttributeSet

import MathOptInterface as MOI
include("test/Bridges/sdpa_models.jl")
model =
    MOI.instantiate(StandardSDPAModel{Float64}; with_bridge_type = Float64)
config = MOI.Test.Config(
    exclude = Any[MOI.optimize!, MOI.SolverName, MOI.SolverVersion],
)
MOI.Test.test_model_ListOfVariablesWithAttributeSet(model, config)

test_model_LowerBoundAlreadySet

import MathOptInterface as MOI
include("test/Bridges/sdpa_models.jl")
model =
    MOI.instantiate(StandardSDPAModel{Float64}; with_bridge_type = Float64)
config = MOI.Test.Config(
    exclude = Any[MOI.optimize!, MOI.SolverName, MOI.SolverVersion],
)
MOI.Test.test_model_LowerBoundAlreadySet(model, config)

test_model_UpperBoundAlreadySet

import MathOptInterface as MOI
include("test/Bridges/sdpa_models.jl")
model =
    MOI.instantiate(StandardSDPAModel{Float64}; with_bridge_type = Float64)
config = MOI.Test.Config(
    exclude = Any[MOI.optimize!, MOI.SolverName, MOI.SolverVersion],
)
MOI.Test.test_model_UpperBoundAlreadySet(model, config)

test_model_ScalarFunctionConstantNotZero

import MathOptInterface as MOI
include("test/Bridges/sdpa_models.jl")
model =
    MOI.instantiate(StandardSDPAModel{Float64}; with_bridge_type = Float64)
config = MOI.Test.Config(
    exclude = Any[MOI.optimize!, MOI.SolverName, MOI.SolverVersion],
)
MOI.Test.test_model_ScalarFunctionConstantNotZero(model, config)

[odow]

So it makes sense that the test_model_UpperBoundAlreadySet and test_model_LowerBoundAlreadySet tests fail, because they're getting reformulated into something.

But look at this:

julia> model =
           MOI.instantiate(GeometricSDPAModel{Float64}; with_bridge_type = Float64);

julia> x = MOI.add_variable(model)
MOI.VariableIndex(1)

julia> set2 = MOI.GreaterThan(0.0)
MathOptInterface.GreaterThan{Float64}(0.0)

julia> set1  = MOI.EqualTo(0.0)
MathOptInterface.EqualTo{Float64}(0.0)

julia> ci = MOI.add_constraint(model, x, set1)
MathOptInterface.ConstraintIndex{MathOptInterface.VariableIndex, MathOptInterface.EqualTo{Float64}}(1)

julia> print(model)
Feasibility

Subject to:

VariableIndex-in-EqualTo{Float64}
 v[1] == 0.0

julia> ci = MOI.add_constraint(model, x, set1)
MathOptInterface.ConstraintIndex{MathOptInterface.VariableIndex, MathOptInterface.EqualTo{Float64}}(1)

julia> print(model)
Feasibility

Subject to:

VectorAffineFunction{Float64}-in-Zeros
 ┌               ┐
 │-0.0 + 1.0 v[1]│
 └               ┘ ∈ Zeros(1)

VariableIndex-in-EqualTo{Float64}
 v[1] == 0.0

[odow]

This one is a maybe. I didn't look too deeply if you know better

[odow]

https://github.com/jump-dev/MathOptInterface.jl/actions/runs/7304556597

[odow]

LGTM. There are still a couple of stragglers, but I don't now what we should do about them. We're much better than before.