如何在Julia中定义typealias的构造函数

在Optim.jl中，我们有一个类型别名

OptimizationTrace{T}

，它是

Vector{OptimizationState{T}

的别名，其中

T

属于

优化器

类型。我想创建一个构造函数

OptimizationTrace{T}（method:：T）=OptimizationTrace{T}（）

，但我正在努力让它工作

abstract Optimizer

immutable NelderMead <: Optimizer
end

immutable OptimizationState{T <: Optimizer}
    iteration::Int
    value::Float64
    g_norm::Float64
    metadata::Dict
end

typealias OptimizationTrace{T} Vector{OptimizationState{T}}

OptimizationTrace{NelderMead}() # works

# A different approach, define a convert-method.
OptimizationTrace{T}(method::T) = OptimizationTrace{T}() #     Doesn't work:
 ERROR: cannot define function OptimizationTrace; it already has a value

import Base.convert
convert{T<:Optimizer}(::Type{Array{OptimizationState{T},1}}, method::T) = OptimizationTrace{T}()
OptimizationTrace(NelderMead())
ERROR: MethodError: `convert` has no method matching convert(::Type{Array{OptimizationState{T},1}}, ::NelderMead)
This may have arisen from a call to the constructor Array{OptimizationState{T},1}(...),

抽象优化器
不变的NelderMead最直接的方法是在v0.5中

abstract Optimizer

immutable NelderMead <: Optimizer
end

immutable OptimizationState{T <: Optimizer}
                  iteration::Int
                  value::Float64
                  g_norm::Float64
                  metadata::Dict
              end

typealias OptimizationTrace{T} Vector{OptimizationState{T}}

OptimizationTrace{T<:Optimizer}(method::T) = Vector{OptimizationState{T}}()
OptimizationTrace(NelderMead())

抽象优化器
不变内尔德米德
call{T <: Optimizer}(::Type{OptimizationTrace}, method::T) = Vector{OptimizationState{T}}()