Caches that are not caches #2862
Description
Consider:
julia> import ECOS
julia> import MathOptInterface
julia> model = MOI.instantiate(ECOS.Optimizer; with_cache_type = Float64)
MOIU.CachingOptimizer
├ state: EMPTY_OPTIMIZER
├ mode: AUTOMATIC
├ model_cache: MOIU.UniversalFallback{ECOS.OptimizerCache}
│ ├ ObjectiveSense: FEASIBILITY_SENSE
│ ├ ObjectiveFunctionType: MOI.ScalarAffineFunction{Float64}
│ ├ NumberOfVariables: 0
│ └ NumberOfConstraints: 0
└ optimizer: ECOS.Optimizer
├ ObjectiveSense: unknown
├ ObjectiveFunctionType: unknown
├ NumberOfVariables: unknown
└ NumberOfConstraints: unknown
julia> x = MOI.add_variable(model)
MOI.VariableIndex(1)
julia> f = MOI.Utilities.vectorize([1.0 * x])
┌ ┐
│0.0 + 1.0 MOI.VariableIndex(1)│
└ ┘
julia> MOI.add_constraint(model, f, MOI.Nonnegatives(1))
MathOptInterface.ConstraintIndex{MathOptInterface.VectorAffineFunction{Float64}, MathOptInterface.Nonnegatives}(1)
julia> MOI.optimize!(model)
ECOS 2.0.8 - (C) embotech GmbH, Zurich Switzerland, 2012-15. Web: www.embotech.com/ECOS
It pcost dcost gap pres dres k/t mu step sigma IR | BT
0 +0.000e+00 -0.000e+00 +1e+00 5e-01 5e-01 1e+00 1e+00 --- --- 0 0 - | - -
1 +0.000e+00 -0.000e+00 +1e-02 4e-03 1e-02 1e-02 1e-02 0.9890 1e-04 0 0 1 | 0 0
2 +0.000e+00 -0.000e+00 +1e-04 4e-05 1e-04 1e-04 1e-04 0.9890 1e-04 0 1 1 | 0 0
3 +0.000e+00 -0.000e+00 +1e-06 5e-07 1e-06 1e-06 1e-06 0.9890 1e-04 0 1 1 | 0 0
4 +0.000e+00 -0.000e+00 +2e-08 5e-09 2e-08 2e-08 2e-08 0.9890 1e-04 0 0 0 | 0 0
5 +0.000e+00 -0.000e+00 +2e-10 6e-11 2e-10 2e-10 2e-10 0.9890 1e-04 0 0 0 | 0 0
OPTIMAL (within feastol=1.7e-10, reltol=nan, abstol=1.7e-10).
Runtime: 0.000077 seconds.
julia> MOI.add_constraint(model, f, MOI.Nonnegatives(1))
ERROR: MathOptInterface.AddConstraintNotAllowed{MathOptInterface.VectorAffineFunction{Float64}, MathOptInterface.Nonnegatives}:
## Cause
Adding `MathOptInterface.VectorAffineFunction{Float64}`-in-`MathOptInterface.Nonnegatives` constraints cannot be performed because:
MatrixOfConstraints does not allow modifications to be made to the model once
`MOI.Utilities.final_touch` has been called. This is called at the end of
`MOI.copy_to` and in `MOI.Utilities.attach_optimizer` (which is called by
`MOI.optimize!` in a `MOI.Utilities.CachingOptimizer`). In order to be able to
apply modifications to this model, you should add a layer
`MOI.Utilities.CachingOptimizer(MOI.Utilities.Model{Float64}(), model)`
where `model` is the current model. This will automatically empty `model` when
modifications are done after `MOI.Utilities.final_touch` is called and copy the
model again in `MOI.Utilities.attach_optimizer`.
## Fixing this error
An `MOI.NotAllowedError` error occurs when you have tried to do something that
is not implemented by the solver.
The most common way to fix this error is to wrap the optimizer in a
`MOI.Utilities.CachingOptimizer`.
For example, if you are using `JuMP.Model` or `JuMP.set_optimizer`, do:
```julia
model = JuMP.Model(optimizer; with_cache_type = Float64)
model = JuMP.GenericModel{T}(optimizer; with_cache_type = T)
JuMP.set_optimizer(model, optimizer; with_cache_type = Float64)
```
Similarly, if you are using `MOI.instantiate`, do:
```julia
model = MOI.instantiate(optimizer; with_cache_type = Float64)
```
Stacktrace:
[1] add_constraint(model::MathOptInterface.Utilities.MatrixOfConstraints{…}, func::MathOptInterface.VectorAffineFunction{…}, set::MathOptInterface.Nonnegatives)
@ MathOptInterface.Utilities ~/.julia/packages/MathOptInterface/zq9bo/src/Utilities/matrix_of_constraints.jl:403
[2] add_constraint(model::ECOS.ZerosOrNot{…}, func::MathOptInterface.VectorAffineFunction{…}, set::MathOptInterface.Nonnegatives)
@ MathOptInterface.Utilities ~/.julia/packages/MathOptInterface/zq9bo/src/Utilities/struct_of_constraints.jl:67
[3] add_constraint
@ ~/.julia/packages/MathOptInterface/zq9bo/src/Utilities/model.jl:325 [inlined]
[4] add_constraint(uf::MathOptInterface.Utilities.UniversalFallback{…}, func::MathOptInterface.VectorAffineFunction{…}, set::MathOptInterface.Nonnegatives)
@ MathOptInterface.Utilities ~/.julia/packages/MathOptInterface/zq9bo/src/Utilities/universalfallback.jl:851
[5] add_constraint(m::MathOptInterface.Utilities.CachingOptimizer{…}, func::MathOptInterface.VectorAffineFunction{…}, set::MathOptInterface.Nonnegatives)
@ MathOptInterface.Utilities ~/.julia/packages/MathOptInterface/zq9bo/src/Utilities/cachingoptimizer.jl:586
[6] top-level scope
@ REPL[112]:1
Some type information was truncated. Use `show(err)` to see complete types.But I aaaaaam using a caching optimizer you say.
That's because our cache is a ECOS.OptimizerCache instead of a Utilities.Model.
That's intended behaviour because we optimised for JuMP, which has a top-level cache. It's problematic for Juniper which embeds it as a sub-solver: lanl-ansi/Juniper.jl#269
I guess the solution is for it to use an explicit cache.