Fix promotion rules for GenericNonlinearExpr

[odow]

Closes #3482

https://github.com/jump-dev/JuMP.jl/actions/runs/6067289679

[codecov]

Codecov Report

Patch coverage: 66.66% and project coverage change: -0.06% ⚠️

Comparison is base (7b98ffc) 98.09% compared to head (7002356) 98.04%.
Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3483      +/-   ##
==========================================
- Coverage   98.09%   98.04%   -0.06%     
==========================================
  Files          37       37              
  Lines        5518     5525       +7     
==========================================
+ Hits         5413     5417       +4     
- Misses        105      108       +3     

Files Changed	Coverage Δ
src/nlp_expr.jl	99.04% <50.00%> (-0.23%)	⬇️
src/aff_expr.jl	97.34% <100.00%> (+0.02%)	⬆️

... and 3 files with indirect coverage changes

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

[odow]

This fixes the UnitJuMP test failure, although UnitJuMP still commits type piracy.

[odow]

It's still possible to generate Vector{AbstractJuMPScalar}:

julia> using JuMP

julia> model = Model();

julia> @variable(model, x);

julia> y = [i == 1 ? x : sin(x) for i in 1:2]
2-element Vector{AbstractJuMPScalar}:
 x
 sin(x)

julia> @objective(model, Min, [i == 1 ? x : sin(x) for i in 1:2])
ERROR: MethodError: no method matching moi_function(::Vector{AbstractJuMPScalar})
Closest candidates are:
  moi_function(::AbstractConstraint) at /Users/oscar/.julia/dev/JuMP/src/constraints.jl:550
  moi_function(::AbstractVariableRef) at /Users/oscar/.julia/dev/JuMP/src/variables.jl:575
  moi_function(::Vector{var"#s65"} where var"#s65"<:AbstractVariableRef) at /Users/oscar/.julia/dev/JuMP/src/variables.jl:584
  ...
Stacktrace:
 [1] set_objective_function(model::Model, func::Vector{AbstractJuMPScalar})
   @ JuMP ~/.julia/dev/JuMP/src/objective.jl:146
 [2] set_objective(model::Model, sense::MathOptInterface.OptimizationSense, func::Vector{AbstractJuMPScalar})
   @ JuMP ~/.julia/dev/JuMP/src/objective.jl:175
 [3] macro expansion
   @ ~/.julia/dev/JuMP/src/macros.jl:1850 [inlined]
 [4] top-level scope
   @ REPL[5]:1

julia> @objective(model, Min, y)
ERROR: MethodError: no method matching moi_function(::Vector{AbstractJuMPScalar})

So perhaps we need a better error for the fallback.

[odow]

Thoughts @pulsipher?

[odow]

@trulsf fixed UnitJuMP, but I think this is the right thing regardless. We can always add back support for Vector{AbstractJuMPScalar} later.