Test coverage - Githubissues

Just ran @timholy PR on Coverage (thanks, Tim!) and inspecting the .cov files, I see the following spots where our coverage could improve:

[ ] copy(::Model) with nonempty nlpdata
[ ] changing (QP) objective in model built in memory when !method_exists(MathProgBase.setquadobjterms!, ...)
[ ] adding linear constraint in model built in memory when !method_exists(MathProgBase.addconstr!, ...)
[ ] addSOS1/addSOS2 in model built in memory
[ ] adding quad. constraint in model built in memory when !method_exists(MathProgBase.addquadconstr!, ...)
[ ] Adding fixed variable to existing constraints
[ ] lines 95--97, 177--179, 213, in writers.jl (??)
[ ] UnsetSolver for NLP call solve (how is this not covered?)
[ ] Getting infeasible ray when solver does not support it
[ ] Getting unbounded ray
[ ] Grabbing solution stuff when status is not in [:Optimal, :Infeasible, :Unbounded]
[ ] MIP solvers not supporting hot-start
[ ] MIP w/o stat == :Optimal (??)
[ ] buildInternalModel for non-continuous model (and possibly QP/QCP/NLP as well)
[ ] Lazy callback not at :MIPSol
[ ] Adding an empty lazy/user cut
[ ] Printing a solver that is not UnsetSolver (there's actually a regression here on v0.4 I've noticed)
[ ] Tiebreaker in _isless
[ ] Needing :ExprGraph
[ ] Quadratic constraints in NLP
[ ] Linear constraints in NLP (I think?)
[ ] MINLP (this might just be because I don't have KNITRO locally)
[ ] parseExpr lines 237--239 in parseExpr_staged.jl (which I think is unreachable)
[x] Base.next(x::JuMPArray, ::Any) for ndims(x) > 1 (this code could use optimization anyway, from the looks of it)

Plus a whole bunch of test-throws cases we could add to try to catch bad behavior in user code.

jump-dev / JuMP.jl