Add is_solved_and_feasible

[odow]

x-ref https://discourse.julialang.org/t/jump-behaviour-of-value-for-infeasible-models/109662/4

Perhaps one reason the question keeps coming up is that the docs don't actually uniformly implement the checks (because we "know" that HiGHS will find the right answer) 😄 I'll take the blame for that.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (7181551) 98.32% compared to head (a9a9257) 98.33%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #3668   +/-   ##
=======================================
  Coverage   98.32%   98.33%           
=======================================
  Files          43       43           
  Lines        5686     5696   +10     
=======================================
+ Hits         5591     5601   +10     
  Misses         95       95           

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

[odow]

See https://discourse.julialang.org/t/jump-behaviour-of-value-for-infeasible-models/109662/5 for some additional thoughts.

[odow]

I think I like my suggestion from discourse:

"""
    has_solution(model::Model; dual::Bool = false)

Return `true` if the model has a feasible primal solution to return and the
solver terminated normally. If `dual`, additionally check that a feasible
dual solution is available.

If this function returns `false`, use [`termination_status`](@ref), [`result_count`](@ref),
[`primal_status`](@ref) and [`dual_status`](@ref) to understand what solutions
are available (if any).
"""
function has_solution(model::Model; dual::Bool = false)
    ret = termination_status(model) in (OPTIMAL, LOCALLY_SOLVED) &&
          primal_status(model) == FEASIBLE_POINT
    if dual
        ret &= dual_status(model) == FEASIBLE_POINT
    end
    return ret
end

Then these would all just become @assert has_solution(model) and @assert has_solution(model; dual = true).

We have has_values and has_duals, but they don't check the termination status.

[gdalle]

This is a very welcome addition! I would also add some details on the various problem statuses, as mentioned on discourse
https://discourse.julialang.org/t/jump-behaviour-of-value-for-infeasible-models/109662/6

[odow]

See https://jump.dev/JuMP.jl/previews/PR3668/manual/solutions/

[odow]

I've changed to is_solved_and_feasible