solveRobust( ... ) vs solve(...) with multiple dispatch?

[vgupta1]

Minor issue: it'd be nice to call solve() and print() instead of solveRobust() and printRobust() using multiple dispatch if possible. Is there a reason we can't do it this way?

[IainNZ]

Can't be done (yet) Robust models are JuMP models with an "extension".

function RobustModel(;solver=nothing,cutsolver=nothing)
    m = Model(solver=solver)
    m.ext[:Robust] = RobustData(cutsolver)
    return m
end

Alternatives that would let you do multiple dispatch including using an AbstractModel from which Model and a hypothetical RobustModel inherit from is one option, another was making Model parametric. Both had issues... This is something @mlubin and I talked about a lot, and this "extension" mechanism was the best we could do without compromising any JuMP speed. Somewhat related to https://github.com/JuliaLang/julia/issues/4935

[vgupta1]

I noticed that... Do you know if the language extension is coming soon? (i.e. requiring abstract types to have certain members?)

Alternatively, although it seems a bit dodgy, (for my own edifiaton) couldn't you have done this?

type RobustModel <: Model
    m  #This is actually a model
     rd #this is the robust data and extension stuff
end

In other words, Robust models implement BOTH "is a" and "has a" in the inheritance scheme?

---

Vishal Gupta PhD Candidate, 2014 Operations Research Center Massachusetts Institute of Technology www.mit.edu/~vgupta1/

On Thu, Apr 10, 2014 at 9:50 PM, Iain Dunning notifications@github.comwrote:

Can't be done (yet) Robust models are JuMP models with an "extension".

function RobustModel(;solver=nothing,cutsolver=nothing) m = Model(solver=solver) m.ext[:Robust] = RobustData(cutsolver) return m end

Alternatives that would let you do multiple dispatch including using an AbstractModel from which Model and a hypothetical RobustModel inherit from is one option, another was making Model parametric. Both had issues... This is something @mlubin https://github.com/mlubin and I talked about a lot, and this "extension" mechanism was the best we could do without compromising any JuMP speed. Somewhat related to JuliaLang/julia#4935https://github.com/JuliaLang/julia/issues/4935

Reply to this email directly or view it on GitHubhttps://github.com/IainNZ/JuMPeR.jl/issues/13#issuecomment-40163671 .

[mlubin]

The issue is that variables contain a reference to their parent model, which needs to be a Model. We tried making variables parametric on the model type, but that led to Julia's version of C++ template hell. Having different model types also makes extensions mutually exclusive.

[mlubin]

Also you can't inherit from concrete types in Julia. Model would need to be an abstract type.