[breaking] change DualOptimizer to not add bridges by default

[blegat]

It's a bit counter intuitive to still add caches and bridges when the user explicitly give an instantiated model. I guess the use case is the DualizableModel but maybe we don't need this. We're using a UniversalFallback on top of it anyway so what's the use for it instead of just MOI.Utilities.Model ?

[blegat]

This is consistent with ParametricOptInterface which does some magic if it takes an optimizer constructor as argument but when it takes an instantiated optimizer, it does not modify it.

[joaquimg]

Interesting,
Should we document this pattern in MOI docs?

[blegat]

Interesting, Should we document this pattern in MOI docs?

Yes, I think we should discuss this, and it should also go in parallel with the layer design that @odow suggested in
jump-dev/JuMP.jl#4014 (comment)
In the meantime, we can try to see if we can simplify the logic of each layer. For POI, we could improve MatrixOfConstraints to get rid of
https://github.com/jump-dev/ParametricOptInterface.jl/blob/2219828f43a81268a3c0b4bc787b1ea48a9e768f/src/ParametricOptInterface.jl#L304-L317

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.08%. Comparing base (685be6c) to head (4895b1e).
⚠️ Report is 8 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #209   +/-   ##
=======================================
  Coverage   95.08%   95.08%           
=======================================
  Files          16       16           
  Lines         834      834           
=======================================
  Hits          793      793           
  Misses         41       41           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[hz-xiaxz]

I'm very excited about this new feature and want to try it out to solve the problem discribed in https://discourse.julialang.org/t/memory-blow-up-when-solving-a-sparse-affine-hermitian-sdp-translated-from-yalmip-to-jump/137744/7

Can I adapt this branch locally to try out, or do I need to make any modification to the MOI?

[odow]

For now, you can just do:

function dual_optimizer()
    inner = MosekTools.Optimizer()
    dual_problem = Dualization.DualProblem{Float64}(inner)
    return Dualization.DualOptimizer{Float64,typeof(inner)}(dual_problem)
end

model = Model(dual_optimizer)

which I linked in https://discourse.julialang.org/t/bottleneck-of-jump/136609/21?u=odow