Back to Julia

Scoped Values

doc/src/base/scopedvalues.md

1.12.68.4 KB
Original Source

[Scoped Values](@id scoped-values)

Scoped values provide an implementation of dynamic scoping in Julia.

!!! note "Lexical scoping vs dynamic scoping" [Lexical scoping](@ref scope-of-variables) is the default behavior in Julia. Under lexical scoping the scope of a variable is determined by the lexical (textual) structure of a program. Under dynamic scoping a variable is bound to the most recent assigned value during the program's execution.

The state of a scoped value is dependent on the execution path of the program. This means that for a scoped value you may observe multiple different values concurrently.

!!! compat "Julia 1.11" Scoped values were introduced in Julia 1.11. In Julia 1.8+ a compatible implementation is available from the package ScopedValues.jl.

In its simplest form you can create a [ScopedValue](@ref Base.ScopedValues.ScopedValue) with a default value and then use [with](@ref Base.ScopedValues.with) or [@with](@ref Base.ScopedValues.@with) to enter a new dynamic scope. The new scope will inherit all values from the parent scope (and recursively from all outer scopes) with the provided scoped value taking priority over previous definitions.

Let's first look at an example of lexical scope. A let statement begins a new lexical scope within which the outer definition of x is shadowed by it's inner definition.

julia
x = 1
let x = 5
    @show x # 5
end
@show x # 1

In the following example, since Julia uses lexical scope, the variable x in the body of f refers to the x defined in the global scope, and entering a let scope does not change the value f observes.

julia
x = 1
f() = @show x
let x = 5
    f() # 1
end
f() # 1

Now using a ScopedValue we can use dynamic scoping.

julia
using Base.ScopedValues

x = ScopedValue(1)
f() = @show x[]
with(x=>5) do
    f() # 5
end
f() # 1

Note that the observed value of the ScopedValue is dependent on the execution path of the program.

It often makes sense to use a const variable to point to a scoped value, and you can set the value of multiple ScopedValues with one call to with.

julia
using Base.ScopedValues

f() = @show a[]
g() = @show b[]

const a = ScopedValue(1)
const b = ScopedValue(2)

f() # a[] = 1
g() # b[] = 2

# Enter a new dynamic scope and set value.
with(a => 3) do
    f() # a[] = 3
    g() # b[] = 2
    with(a => 4, b => 5) do
        f() # a[] = 4
        g() # b[] = 5
    end
    f() # a[] = 3
    g() # b[] = 2
end

f() # a[] = 1
g() # b[] = 2

ScopedValues provides a macro version of with. The expression @with var=>val expr evaluates expr in a new dynamic scope with var set to val. @with var=>val expr is equivalent to with(var=>val) do expr end. However, with requires a zero-argument closure or function, which results in an extra call-frame. As an example, consider the following function f:

julia
using Base.ScopedValues
const a = ScopedValue(1)
f(x) = a[] + x

If you wish to run f in a dynamic scope with a set to 2, then you can use with:

julia
with(() -> f(10), a=>2)

However, this requires wrapping f in a zero-argument function. If you wish to avoid the extra call-frame, then you can use the @with macro:

julia
@with a=>2 f(10)

!!! note Dynamic scopes are inherited by Tasks, at the moment of task creation. Dynamic scopes are not propagated through Distributed.jl operations.

In the example below we open a new dynamic scope before launching a task. The parent task and the two child tasks observe independent values of the same scoped value at the same time.

julia
using Base.ScopedValues
import Base.Threads: @spawn

const scoped_val = ScopedValue(1)
@sync begin
    with(scoped_val => 2)
        @spawn @show scoped_val[] # 2
    end
    with(scoped_val => 3)
        @spawn @show scoped_val[] # 3
    end
    @show scoped_val[] # 1
end

Scoped values are constant throughout a scope, but you can store mutable state in a scoped value. Just keep in mind that the usual caveats for global variables apply in the context of concurrent programming.

Care is also required when storing references to mutable state in scoped values. You might want to explicitly [unshare mutable state](@ref unshare_mutable_state) when entering a new dynamic scope.

julia
using Base.ScopedValues
import Base.Threads: @spawn

const sval_dict = ScopedValue(Dict())

# Example of using a mutable value wrongly
@sync begin
    # `Dict` is not thread-safe the usage below is invalid
    @spawn (sval_dict[][:a] = 3)
    @spawn (sval_dict[][:b] = 3)
end

@sync begin
    # If we instead pass a unique dictionary to each
    # task we can access the dictionaries race free.
    with(sval_dict => Dict()) do
        @spawn (sval_dict[][:a] = 3)
    end
    with(sval_dict => Dict()) do
        @spawn (sval_dict[][:b] = 3)
    end
end

Example

In the example below we use a scoped value to implement a permission check in a web-application. After determining the permissions of the request, a new dynamic scope is entered and the scoped value LEVEL is set. Other parts of the application can query the scoped value and will receive the appropriate value. Other alternatives like task-local storage and global variables are not well suited for this kind of propagation; our only alternative would have been to thread a value through the entire call-chain.

julia
using Base.ScopedValues

const LEVEL = ScopedValue(:GUEST)

function serve(request, response)
    level = isAdmin(request) ? :ADMIN : :GUEST
    with(LEVEL => level) do
        Threads.@spawn handle(request, response)
    end
end

function open(connection::Database)
    level = LEVEL[]
    if level !== :ADMIN
        error("Access disallowed")
    end
    # ... open connection
end

function handle(request, response)
    # ...
    open(Database(#=...=#))
    # ...
end

Idioms

[Unshare mutable state](@id unshare_mutable_state)

julia
using Base.ScopedValues
import Base.Threads: @spawn

const sval_dict = ScopedValue(Dict())

# If you want to add new values to the dict, instead of replacing
# it, unshare the values explicitly. In this example we use `merge`
# to unshare the state of the dictionary in parent scope.
@sync begin
    with(sval_dict => merge(sval_dict[], Dict(:a => 10))) do
        @spawn @show sval_dict[][:a]
    end
    @spawn sval_dict[][:a] = 3 # Not a race since they are unshared.
end

Scoped values as globals

In order to access the value of a scoped value, the scoped value itself has to be in (lexical) scope. This means most often you likely want to use scoped values as constant globals.

julia
using Base.ScopedValues
const sval = ScopedValue(1)

Indeed one can think of scoped values as hidden function arguments.

This does not preclude their use as non-globals.

julia
using Base.ScopedValues
import Base.Threads: @spawn

function main()
    role = ScopedValue(:client)

    function launch()
        #...
        role[]
    end

    @with role => :server @spawn launch()
    launch()
end

But it might have been simpler to just directly pass the function argument in these cases.

Very many ScopedValues

If you find yourself creating many ScopedValue's for one given module, it may be better to use a dedicated struct to hold them.

julia
using Base.ScopedValues

Base.@kwdef struct Configuration
    color::Bool = false
    verbose::Bool = false
end

const CONFIG = ScopedValue(Configuration(color=true))

@with CONFIG => Configuration(color=CONFIG[].color, verbose=true) begin
    @show CONFIG[].color # true
    @show CONFIG[].verbose # true
end

API docs

@docs
Base.ScopedValues.ScopedValue
Base.ScopedValues.with
Base.ScopedValues.@with
Base.isassigned(::Base.ScopedValues.ScopedValue)
Base.ScopedValues.get

Implementation notes and performance

Scopes use a persistent dictionary. Lookup and insertion is O(log(32, n)), upon dynamic scope entry a small amount of data is copied and the unchanged data is shared among other scopes.

The Scope object itself is not user-facing and may be changed in a future version of Julia.

Design inspiration

This design was heavily inspired by JEPS-429, which in turn was inspired by dynamically scoped free variables in many Lisp dialects. In particular Interlisp-D and its deep binding strategy.

A prior design discussed was context variables ala PEPS-567 and implemented in Julia as ContextVariablesX.jl.