Implement `getkey` for DoubleDicts

[jd-lara]

Related to POI, in some cases to get the key from a particular key for a value. For instance, recover the parametrized original function when calling MOI.get(model::ParametricOptimizer, attr::MOI.ConstraintFunction, ci::MOI.ConstraintIndex{F, S})

[odow]

Can you link to the use-case?

[jd-lara]

@odow I want to recover the key from this dict https://github.com/jump-dev/ParametricOptInterface.jl/blob/0f9f0bb76a56eb70c93b0e30ba9460a83036e0f9/src/ParametricOptInterface.jl#L47

[odow]

Given what?

[odow]

getkey assumes you know the key?

help?> getkey
search: getkey

  getkey(collection, key, default)

  Return the key matching argument key if one exists in collection, otherwise return default.

  Examples
  ≡≡≡≡≡≡≡≡≡≡

  julia> D = Dict('a'=>2, 'b'=>3)
  Dict{Char, Int64} with 2 entries:
    'a' => 2
    'b' => 3
  
  julia> getkey(D, 'a', 1)
  'a': ASCII/Unicode U+0061 (category Ll: Letter, lowercase)
  
  julia> getkey(D, 'd', 'a')
  'a': ASCII/Unicode U+0061 (category Ll: Letter, lowercase)

[jd-lara]

Given what?

The index, and see if it is indexed already or maybe it was converted to quadratic and I need to go search in the quad cache. I can also use other patterns but it seemed easier to concatenate the searches.

[odow]

Do you mean haskey? I'm confused as to where this gets used. Can you sketch an example?

[jd-lara]

all getkey does is return a default value if the key isn't in the dict. It summarizes this pattern:

mydict = Dict("A" => 1, "B" =>3)

# this: 
value = haskey(mydict, "C") ? mydict["C"] : 0.0

# is the same as: 

value = getkey(mydict, "C", 0.0)

[jd-lara]

This is the function in POI I was using but I can write it with haskey instead.

function MOI.get(model::ParametricOptimizer, attr::MOI.ConstraintFunction, ci::MOI.ConstraintIndex{F, S}) where {F, S}
    # Check if the index was cached as Affine expression
    if !isempty(model.affine_constraint_cache)
        value = get(model.affine_constraint_cache, ci, MOI.ScalarAffineFunction(Array{MathOptInterface.ScalarAffineTerm{Float64}, 1}(), 0.0))
        !iszero(value) && return value
    # Check if the index was cached as a Quadratic
    elseif !isempty(model.quadratic_added_cache)
        value = getkey(model.quadratic_added_cache, ci, MOI.ScalaQuadraticFunction(Array{MathOptInterface.ScalarAffineTerm{Float64}, 1}(), 0.0))
        !iszero(value) && return value
    else
        return MOI.get(model.optimizer, attr, ci)
    end
end

[odow]

getkey doesn't do what you think it does, and your function has a few things wrong with it. In particular, it may return nothing, and it will only check the quadratic_added_cache if the affine_constriant_cache is empty.

Did you mean something like:

function MOI.get(
    model::ParametricOptimizer,
    attr::MOI.ConstraintFunction,
    ci::MOI.ConstraintIndex{F,S},
)::F where {F,S}
    value = get(model.affine_constraint_cache, ci, nothing)
    if value !== nothing
        return value
    end
    value = get(model.quadratic_added_cache, ci, nothing)
    if value !== nothing
        return value
    end
    return MOI.get(model.optimizer, attr, ci)
end

But this might still be wrong since value might not always be F? Don't you need to write separate functions for affine and quadratic functions?

[jd-lara]

@odow yeah the code I pasted was hasty, this is what I pushed jump-dev/ParametricOptInterface.jl@b0d3091