feat: Refactor NLP modelers to use Strategies architecture

[ocots]

Summary

This PR refactors the NLP modelers (ADNLPModeler, ExaModeler) to use the new Strategies architecture instead of the legacy AbstractOCPTool interface that will be completely removed.

🔄 Migration from OCPTool to Strategies

Before (Legacy):

# Using AbstractOCPTool with _option_specs and _build_ocp_tool_options
function _option_specs(::Type{<:ADNLPModeler})
    return (
        show_time=OptionSpec(; type=Bool, default=false, ...),
        backend=OptionSpec(; type=Symbol, default=:optimized, ...),
    )
end

function ADNLPModeler(; kwargs...)
    values, sources = _build_ocp_tool_options(ADNLPModeler; kwargs...)
    return ADNLPModeler{typeof(values),typeof(sources)}(values, sources)
end

After (Strategies):

# Using AbstractStrategy with StrategyMetadata and StrategyOptions
struct ADNLPModeler <: AbstractStrategy
    options::StrategyOptions
end

Strategies.id(::Type{ADNLPModeler}) = :adnlp

Strategies.metadata(::Type{ADNLPModeler}) = StrategyMetadata(
    OptionDefinition(
        name = :show_time,
        type = Bool,
        default = false,
        description = "Whether to show timing information while building the ADNLP model."
    ),
    OptionDefinition(
        name = :backend,
        type = Symbol,
        default = :optimized,
        description = "Automatic differentiation backend used by ADNLPModels."
    )
)

function ADNLPModeler(; kwargs...)
    options = Strategies.build_strategy_options(ADNLPModeler; kwargs...)
    return ADNLPModeler(options)
end

📁 Files Modified

• src/nlp/nlp_backends.jl - Refactor ADNLPModeler and ExaModeler to Strategies
• src/nlp/model_api.jl - Update modeler registration and lookup functions
• src/nlp/problem_core.jl - Remove OCPTool dependencies
• src/nlp/discretized_ocp.jl - Update to use Strategies interface
• src/nlp/extract_solver_infos.jl - Update solution extraction

✅ Benefits

1. Type Safety: Options are stored in StrategyOptions with proper type checking
2. Validation: Automatic validation against OptionDefinition contracts
3. Aliases: Support for option aliases (e.g., max_iter vs max_iterations)
4. Introspection: Programmatic access to available options and defaults
5. Consistency: Same pattern as discretizers and solvers
6. Future-Ready: Compatible with Orchestration routing and disambiguation

🧪 Testing

• All existing modeler functionality preserved
• Option validation and error handling improved
• Integration with Orchestration module tested
• Backward compatibility maintained where possible

📋 Breaking Changes

• Removed: Legacy AbstractOCPTool interface for modelers
• Changed: Constructor now uses StrategyOptions instead of custom parameterized types
• Updated: Option access through Strategies.option_value(modeler, :option_name)

🔄 Migration Guide for Users

Old code:

modeler = ADNLPModeler(show_time=true, backend=:sparse)

New code:

modeler = ADNLPModeler(show_time=true, backend=:sparse)
# Same interface, better validation and type safety!

This completes the migration of all NLP components to the Strategies architecture, paving the way for removing the legacy AbstractOCPTool system entirely.

[github-actions]

Breakage test results
Date: 2026-01-26 16:04:22

Name	Latest	Stable
CTDirect	compat: v0.6.10-beta	compat: v0.7.1-beta
CTFlows	compat: v0.6.10-beta	compat: v0.7.1-beta
OptimalControl	compat: v0.6.10-beta	compat: v0.6.10-beta