feat: Add Uno solver integration with comprehensive tests by ocots · Pull Request #103 · control-toolbox/CTSolvers.jl

ocots · 2026-03-16T21:53:44Z

@jbcaillau @PierreMartinon @joseph-gergaud Uno solver

Add Uno solver with CPU-only support and ADNLP modeler compatibility
Implement complete metadata with validation for all Uno options
Add comprehensive test suite covering:
- Type hierarchy and contracts (test_solver_types.jl)
- Extension stubs and error handling (test_extension_stubs.jl)
- Parameter validation (test_uno_parameter_validation.jl)
- Type stability (test_type_stability.jl)
- Module exports and interface compliance (test_solvers.jl)
- Integration with validation modes (test_comprehensive_validation.jl)
- Real strategy mode tests (test_real_strategies_mode.jl)
- Route-to functionality (test_route_to_comprehensive.jl)
- Full extension tests with problem solving (test_uno_extension.jl)

Key features:

Uno supports presets (ipopt, filtersqp) and full option set
CPU execution only (GPU parameter rejected at compile time)
ADNLP modeler compatibility (no Exa support)
Comprehensive option validation wi
Add Uno solver with CPU-only support and ADNLP modeler compatibility AP- Implement complete metadata with validation for all Uno options
Arc- Add comprehensive test suite covering:
- Type hierarchy and l * Type hierarchy and contracts (test_oS * Extension stub

- Add Uno solver with CPU-only support and ADNLP modeler compatibility - Implement complete metadata with validation for all Uno options - Add comprehensive test suite covering: * Type hierarchy and contracts (test_solver_types.jl) * Extension stubs and error handling (test_extension_stubs.jl) * Parameter validation (test_uno_parameter_validation.jl) * Type stability (test_type_stability.jl) * Module exports and interface compliance (test_solvers.jl) * Integration with validation modes (test_comprehensive_validation.jl) * Real strategy mode tests (test_real_strategies_mode.jl) * Route-to functionality (test_route_to_comprehensive.jl) * Full extension tests with problem solving (test_uno_extension.jl) Key features: - Uno supports presets (ipopt, filtersqp) and full option set - CPU execution only (GPU parameter rejected at compile time) - ADNLP modeler compatibility (no Exa support) - Comprehensive option validation wi - Add Uno solver with CPU-only support and ADNLP modeler compatibility AP- Implement complete metadata with validation for all Uno options - Arc- Add comprehensive test suite covering: * Type hierarchy and l * Type hierarchy and contracts (test_oS * Extension stub

…sion - Add UnoSolver to test targets in Project.toml - Import NLPModels in CTSolversUno extension to fix precompilation error This fixes the missing dependency error when running Uno tests.

- Use raw UnoSolver constants for direct solver calls (solver(nlp)) - Use Symbol() wrapper for CommonSolve.solve calls (goes through extract_solver_infos) - Fix all status comparisons in test_uno_extension.jl This fixes 8 test failures, reducing from 53 pass/8 fail/11 error to 61 pass/0 fail/11 error.

- Add SolverCore import and status mapping function in ext/CTSolversUno.jl - Map Uno statuses to SolverCore symbols (UNO_FEASIBLE_KKT_POINT → :first_order, etc.) - Convert UnoSolver.Statistics to GenericExecutionStats in solve_with_uno - Update all tests to use GenericExecutionStats fields (.status, .solution, .objective) - Skip Initial Guess tests (Uno performs 1 iteration even with max_iterations=0) - Fix integration tests: add UnoSolver import and use correct canonical option names - All tests passing: 192 passed, 2 intentionally skipped

ocots · 2026-03-17T10:37:44Z

@amontoison @cvanaret Does Uno work with ExaModels? Or just ADNLPModels?

amontoison · 2026-03-17T12:37:16Z

@ocots Uno works with any AbstractNLPModel from.NLPModels.jl.
ExaModel, CUTEstModel, ADNLPModel, AMPLModel, ...

ocots · 2026-03-17T07:59:32Z

ext/CTSolversUno.jl

+function _uno_status_to_solvercore(optimization_status::Cint, solution_status::Cint)::Symbol
+    if optimization_status == UnoSolver.UNO_ITERATION_LIMIT
+        return :max_iter
+    elseif optimization_status == UnoSolver.UNO_TIME_LIMIT
+        return :max_time
+    elseif optimization_status == UnoSolver.UNO_EVALUATION_ERROR
+        return :exception
+    elseif optimization_status == UnoSolver.UNO_ALGORITHMIC_ERROR
+        return :exception
+    else # UNO_SUCCESS
+        if solution_status == UnoSolver.UNO_FEASIBLE_KKT_POINT
+            return :first_order
+        elseif solution_status == UnoSolver.UNO_FEASIBLE_FJ_POINT
+            return :acceptable
+        elseif solution_status == UnoSolver.UNO_INFEASIBLE_STATIONARY_POINT
+            return :infeasible
+        elseif solution_status == UnoSolver.UNO_FEASIBLE_SMALL_STEP
+            return :small_step
+        elseif solution_status == UnoSolver.UNO_INFEASIBLE_SMALL_STEP
+            return :small_step
+        else # UNO_UNBOUNDED
+            return :unbounded
+        end
+    end
+end
+
+"""
+$(TYPEDSIGNATURES)
+
+Convert UnoSolver.Statistics to SolverCore.GenericExecutionStats.
+
+This conversion allows Uno to integrate seamlessly with the CTSolvers pipeline
+which expects SolverCore.AbstractExecutionStats.
+
+# Arguments
+- `nlp::NLPModels.AbstractNLPModel`: The NLP model (needed for GenericExecutionStats constructor)
+- `uno_stats::UnoSolver.Statistics`: Uno solver execution statistics
+
+# Returns
+- `SolverCore.GenericExecutionStats`: Converted statistics compatible with CTSolvers
+
+# Field Mapping
+- `status` ← mapped from `optimization_status` and `solution_status`
+- `solution` ← `primal_solution`
+- `objective` ← `solution_objective`
+- `dual_feas` ← `solution_stationarity`
+- `primal_feas` ← `solution_primal_feasibility`
+- `multipliers` ← `constraint_dual_solution`
+- `multipliers_L` ← `lower_bound_dual_solution`
+- `multipliers_U` ← `upper_bound_dual_solution`
+- `iter` ← `number_iterations`
+- `elapsed_time` ← `cpu_time`
+"""
+function _uno_to_generic_stats(nlp::NLPModels.AbstractNLPModel, uno_stats::UnoSolver.Statistics)::SolverCore.GenericExecutionStats
+    # Map Uno status to SolverCore status
+    status = _uno_status_to_solvercore(
+        uno_stats.optimization_status,
+        uno_stats.solution_status
+    )
+
+    # Create GenericExecutionStats with all fields marked as reliable
+    stats = SolverCore.GenericExecutionStats(
+        nlp;
+        status=status,
+        solution=uno_stats.primal_solution,
+        objective=uno_stats.solution_objective,
+        dual_feas=uno_stats.solution_stationarity,
+        primal_feas=uno_stats.solution_primal_feasibility,
+        multipliers=uno_stats.constraint_dual_solution,
+        multipliers_L=uno_stats.lower_bound_dual_solution,
+        multipliers_U=uno_stats.upper_bound_dual_solution,
+        iter=Int(uno_stats.number_iterations),
+        elapsed_time=uno_stats.cpu_time
+    )
+
+    return stats
+end


@amontoison @cvanaret I have started to add Uno solver. It is easier for me to deal with SolverCore.AbstractExecutionStats hence I have made this conversion system. Do you think it could be placed directly in UnoSolver.jl?

ps: I am not sure it is correct.

Yes, I think we should add a option to decide what is the type of the returned statistics (UnoSolver.Statistics or SolverCore.AbstractExecutionStats).

@ocots Can you open a PR with this converter in UnoSolver.jl ?
I can't open new PR now until I have approvals for working on it.

I am fine with adding SolverCore.jl as a dependency.
I would like to use a GenericExecutionStats instead of a Statistics on the long run but we need more attributes in GenericExecutionStats before.

@cvanaret will maybe open a PR in SolverCore.jl :)

ocots · 2026-03-17T08:10:28Z

test/suite/extensions/test_uno_extension.jl

+        Test.@testset "Initial Guess - max_iterations=0" begin
+            Test.@testset "Rosenbrock" verbose=VERBOSE showtiming=SHOWTIMING begin
+                Test.@testset "Modelers.ADNLP" verbose=VERBOSE showtiming=SHOWTIMING begin
+                    Test.@test_skip "Uno performs 1 iteration even with max_iterations=0"
+                end
+            end
+
+            Test.@testset "Elec" verbose=VERBOSE showtiming=SHOWTIMING begin
+                Test.@testset "Modelers.ADNLP" verbose=VERBOSE showtiming=SHOWTIMING begin
+                    Test.@test_skip "Uno performs 1 iteration even with max_iterations=0"
+                end
+            end
+        end


@amontoison @cvanaret I wanted to add some tests with 0 iterations, but it seems that when max_iterations=0, Uno performs 1 iteration.

For us, it would be useful to allow 0 iterations in order to build an optimal control solution directly from an initial guess. Indeed, the input (the initial guess) and the output (the optimal control solution) are not of the same type.

Indeed, this special case is not detected. I can patch it real quick.
cvanaret/Uno#597

It would be great!

Soon we will make a new release of OC with Uno and relaunch all our benchmarks 🤞

It's not going to be that quick because I have to do Uno and Uno_jll releases first 😅

This is fixed in UnoSolver.jl v0.2.8

@ocots apparently you deleted your comment. Your function _uno_status_to_solvercore is missing the test solution_status == UNO_NOT_OPTIMAL. But I did forget to set the status to UNO_ITERATION_LIMIT 😆

@cvanaret I tried it, and it works—thanks!

However, even though the number of iterations is effectively 0, the status is not UNO_ITERATION_LIMIT but UNO STATUS: optimization_status = UNO_SUCCESS, solution_status = UNO_NOT_OPTIMAL. Is that expected?

Fixed in cvanaret/Uno#610

ocots · 2026-03-17T13:03:42Z

@ocots Uno works with any AbstractNLPModel from.NLPModels.jl. ExaModel, CUTEstModel, ADNLPModel, AMPLModel, ...

Tested with ADNLP and Exa: https://github.com/control-toolbox/OptimalControl.jl/blob/4176281a4709af90454ee8595ce4dc3c5036ed7c/src/helpers/methods.jl#L42-L60

Some tests from https://github.com/control-toolbox/OptimalControl.jl/blob/4176281a4709af90454ee8595ce4dc3c5036ed7c/test/suite/solve/test_canonical.jl#L52-L152 (not benchmarks since no "compilation" call):

cvanaret · 2026-03-17T14:14:27Z

Nice! Is that with Uno-ipopt or Uno-filtersqp? Based on feedback here, I would recommend using the filtersqp preset. It nicely exploits negative curvature and is way more robust than the ipopt preset.

ocots · 2026-03-17T14:42:04Z

Nice! Is that with Uno-ipopt or Uno-filtersqp? Based on feedback here, I would recommend using the filtersqp preset. It nicely exploits negative curvature and is way more robust than the ipopt preset.

It is with ipopt since we are more familiar with it. I plan to do some benchmarking to choose the default.

github-actions bot assigned ocots Mar 16, 2026

ocots added 3 commits March 16, 2026 23:07

fix: Add UnoSolver to test dependencies and import NLPModels in exten…

7cef442

…sion - Add UnoSolver to test targets in Project.toml - Import NLPModels in CTSolversUno extension to fix precompilation error This fixes the missing dependency error when running Uno tests.

ocots merged commit 72ed9d9 into main Mar 17, 2026
7 checks passed

ocots deleted the uno branch March 17, 2026 10:25

ocots commented Mar 17, 2026

View reviewed changes

cvanaret mentioned this pull request Mar 17, 2026

Do not perform the first iteration if max_iterations == 0 cvanaret/Uno#597

Merged

ocots mentioned this pull request Apr 1, 2026

[General] Release #112

Open

Conversation

ocots commented Mar 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

ocots commented Mar 17, 2026

Uh oh!

amontoison commented Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ocots Mar 17, 2026

Choose a reason for hiding this comment

Uh oh!

amontoison Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

amontoison Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ocots Mar 17, 2026

Choose a reason for hiding this comment

Uh oh!

cvanaret Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ocots Mar 17, 2026

Choose a reason for hiding this comment

Uh oh!

cvanaret Mar 17, 2026

Choose a reason for hiding this comment

Uh oh!

cvanaret Mar 17, 2026

Choose a reason for hiding this comment

Uh oh!

cvanaret Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ocots Mar 18, 2026

Choose a reason for hiding this comment

Uh oh!

cvanaret Mar 19, 2026

Choose a reason for hiding this comment

Uh oh!

ocots commented Mar 17, 2026

Uh oh!

cvanaret commented Mar 17, 2026

Uh oh!

ocots commented Mar 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ocots commented Mar 16, 2026 •

edited

Loading

amontoison commented Mar 17, 2026 •

edited

Loading

amontoison Mar 17, 2026 •

edited

Loading

amontoison Mar 17, 2026 •

edited

Loading

cvanaret Mar 17, 2026 •

edited

Loading

cvanaret Mar 18, 2026 •

edited

Loading