Attributes canget too strict

[tkoolen]

In

https://github.com/JuliaOpt/MathOptInterfaceUtilities.jl/blob/c5b5c4bcae650a141101fb1e16d459c84e582bb9/src/instancemanager.jl#L359-L364

typeof(index) is used for canget, even though a concrete index is available. That seems too strict. It may be the case that you can get an attribute for certain indices, but not all.

[mlubin]

This was discussed at jump-dev/MathOptInterface.jl#191. Allowing the possibility for an attribute not to be available for a subset of indices forces inefficiencies in copy! (e.g., we can't query for all values at the same time). It also adds an extra level of state (and hence complexity) that I don't see a compelling need for.

[mlubin]

Also using indices instead of types in the code block referenced above would mean that for the same attribute, the instancemanager could sometimes return the value from the solver and sometimes from the instance. That's just confusing.

[tkoolen]

Ah, my mistake. I thought there was also a canget method with ::VariableIndex argument, in addition to the ::Type{VariableIndex} version, but I must have misread the MOI docs. I mistakenly implemented the former instead of the latter.