Domain Specific Languages
Foreword
Domain Specific Languages (DSL) allow developers to tailor their application to a particular domain. You don’t need macros in order to have a DSL: every data structure and every function you define in your module is part of your Domain Specific Language.
For example, imagine we want to implement a Validator module which provides a data validation domain specific language. We could implement it using data structures, functions or macros. Let’s see how those different DSLs would look like:
# 1. data structures
import Validator
validate user, name: [length: 1..100],
email: [matches: ~r/@/]
# 2. functions
import Validator
user
|> validate_length(:name, 1..100)
|> validate_matches(:email, ~r/@/)
# 3. macros + modules
defmodule MyValidator do
use Validator
validate_length :name, 1..100
validate_matches :email, ~r/@/
end
MyValidator.validate(user)
Of all the approaches above, the first is definitely the most flexible. If our domain rules can be encoded with data structures, they are by far the easiest to compose and implement, as Elixir’s standard library is filled with functions for manipulating different data types.
The second approach uses function calls which better suits more complex APIs (for example, if you need to pass many options) and reads nicely in Elixir thanks to the pipe operator.
The third approach, uses macros, and is by far the most complex. It will take more lines of code to implement, it is hard and expensive to test (compared to testing simple functions), and it limits how the user may use the library since all validations need to be defined inside a module.
To drive the point home, imagine you want to validate a certain attribute only if a given condition is met. We could easily achieve it with the first solution, by manipulating the data structure accordingly, or with the second solution by using conditionals (if/else) before invoking the function. However it is impossible to do so with the macros approach unless its DSL is augmented.
In other words:
data > functions > macros
That said, there are still cases where using macros and modules to build domain specific languages is useful. Since we have explored data structures and function definitions in the Getting Started guide, this chapter will explore how to use macros and module attributes to tackle more complex DSLs.
Building our own test case
The goal in this chapter is to build a module named TestCase
that allows us to write the following:
defmodule MyTest do
use TestCase
test "arithmetic operations" do
4 = 2 + 2
end
test "list operations" do
[1, 2, 3] = [1, 2] ++ [3]
end
end
MyTest.run
In the example above, by using TestCase
, we can write tests using the test
macro, which defines a function named run
to automatically run all tests for us. Our prototype will rely on the match operator (=
) as a mechanism to do assertions.
The test
macro
Let’s start by creating a module that defines and imports the test
macro when used:
defmodule TestCase do
# Callback invoked by `use`.
#
# For now it returns a quoted expression that
# imports the module itself into the user code.
@doc false
defmacro __using__(_opts) do
quote do
import TestCase
end
end
@doc """
Defines a test case with the given description.
## Examples
test "arithmetic operations" do
4 = 2 + 2
end
"""
defmacro test(description, do: block) do
function_name = String.to_atom("test " <> description)
quote do
def unquote(function_name)(), do: unquote(block)
end
end
end
Assuming we defined TestCase
in a file named tests.exs
, we can open it up by running iex tests.exs
and define our first tests:
iex> defmodule MyTest do
...> use TestCase
...>
...> test "hello" do
...> "hello" = "world"
...> end
...> end
For now we don’t have a mechanism to run tests, but we know that a function named “test hello” was defined behind the scenes. When we invoke it, it should fail:
iex> MyTest."test hello"()
** (MatchError) no match of right hand side value: "world"
Storing information with attributes
In order to finish our TestCase
implementation, we need to be able to access all defined test cases. One way of doing this is by retrieving the tests at runtime via __MODULE__.__info__(:functions)
, which returns a list of all functions in a given module. However, considering that we may want to store more information about each test besides the test name, a more flexible approach is required.
When discussing module attributes in earlier chapters, we mentioned how they can be used as temporary storage. That’s exactly the property we will apply in this section.
In the __using__/1
implementation, we will initialize a module attribute named @tests
to an empty list, then store the name of each defined test in this attribute so the tests can be invoked from the run
function.
Here is the updated code for the TestCase
module:
defmodule TestCase do
@doc false
defmacro __using__(_opts) do
quote do
import TestCase
# Initialize @tests to an empty list
@tests []
# Invoke TestCase.__before_compile__/1 before the module is compiled
@before_compile TestCase
end
end
@doc """
Defines a test case with the given description.
## Examples
test "arithmetic operations" do
4 = 2 + 2
end
"""
defmacro test(description, do: block) do
function_name = String.to_atom("test " <> description)
quote do
# Prepend the newly defined test to the list of tests
@tests [unquote(function_name) | @tests]
def unquote(function_name)(), do: unquote(block)
end
end
# This will be invoked right before the target module is compiled
# giving us the perfect opportunity to inject the `run/0` function
@doc false
defmacro __before_compile__(env) do
quote do
def run do
Enum.each @tests, fn name ->
IO.puts "Running #{name}"
apply(__MODULE__, name, [])
end
end
end
end
end
By starting a new IEx session, we can now define our tests and run them:
iex> defmodule MyTest do
...> use TestCase
...>
...> test "hello" do
...> "hello" = "world"
...> end
...> end
iex> MyTest.run
Running test hello
** (MatchError) no match of right hand side value: "world"
Although we have overlooked some details, this is the main idea behind creating domain specific modules in Elixir. Macros enable us to return quoted expressions that are executed in the caller, which we can then use to transform code and store relevant information in the target module via module attributes. Finally, callbacks such as @before_compile
allow us to inject code into the module when its definition is complete.
Besides @before_compile
, there are other useful module attributes like @on_definition
and @after_compile
, which you can read more about in the docs for the Module
module. You can also find useful information about macros and the compilation environment in the documentation for the Macro
module and Macro.Env
.