Allows Julia function calls to be temporarily overloaded for purpose of testing.

• Usage
• Gotchas
• Overhead

Suppose you wrote the function randdev (UNIX only). How would you go about writing tests for it?

function randdev(n::Integer)
    open("/dev/urandom") do fp
        reverse(read(fp, n))
    end
end

The non-deterministic behaviour of this function makes it hard to test but we can write some tests dealing with the deterministic properties of the function:

using Test
using ...: randdev

n = 10
result = randdev(n)
@test eltype(result) == UInt8
@test length(result) == n

How could we create a test that shows the output of the function is reversed? Mocking.jl provides the @mock macro which allows package developers to temporarily overload a specific calls in their package. In this example we will apply @mock to the open call in randdev:

using Mocking

function randdev(n::Integer)
    @mock open("/dev/urandom") do fp
        reverse(read(fp, n))
    end
end

With the call site being marked as "mockable" we can now write a testcase which allows us to demonstrate the reversing behaviour within the randdev function:

using Mocking
using Test
using ...: randdev

Mocking.activate()  # Need to call `activate` before executing `apply`

n = 10
result = randdev(n)
@test eltype(result) == UInt8
@test length(result) == n

# Produces a string with sequential UInt8 values from 1:n
data = unsafe_string(pointer(convert(Array{UInt8}, 1:n)))

# Generate a alternative method of `open` which call we wish to mock
patch = @patch open(fn::Function, f::AbstractString) = fn(IOBuffer(data))

# Apply the patch which will modify the behaviour for our test
apply(patch) do
    @test randdev(n) == convert(Array{UInt8}, n:-1:1)
end

# Outside of the scope of the patched environment `@mock` is essentially a no-op
@test randdev(n) != convert(Array{UInt8}, n:-1:1)

Remember to:

• Use @mock at desired call sites
• Run Mocking.activate() before executing any apply calls

The @mock macro uses a conditional check of Mocking.activated() which only allows patches to be utilized only when Mocking has been activated. By default, Mocking starts as disabled which should result conditional being optimized away allowing for zero-overhead. Once activated via Mocking.activate() the Mocking.activated function will be re-defined, causing all methods dependent on @mock to be recompiled.

Mocking.jl is provided under the MIT "Expat" License.