A library that helps you to organize your Elixir code in more domain-driven way. Exop provides macros which helps you to encapsulate business logic and offers you additionally: incoming params validation (with predefined contract), params coercion, policy check, fallback behavior, operations chaining and more.
Interested in property-based testing? Check out new Exop family member - ExopData. If you use Exop to organize your code with ExopData you can get property generators in the most easiest way.
The new ExopPlug library provides a convenient way to validate incoming parameters of your Phoenix application's controllers by offering you small but useful DSL.
Here is the CHANGELOG that was started from ver. 0.4.1 ¯\_(ツ)_/¯
def deps do
[{:exop, "~> 1.4"}]
end
defmodule IntegersDivision do
use Exop.Operation
parameter :a, type: :integer, default: 1
parameter :b, type: :integer, required: false,
numericality: %{greater_than: 0}
def process(params) do
result = params[:a] / params[:b]
IO.inspect "The division result is: #{result}"
end
end
Exop.Operation
provides parameter
macro, which is responsible for the contract definition.
Its spec is @spec parameter(atom | String.t, Keyword.t) :: none
, we define parameter name as the first argument and parameter options as the second Keyword
argument.
A parameter name could be either an atom or a string. You could even mix atom-named and string-named parameters in an operation's contract.
Parameter options determine a contract of a parameter, a set of parameters contracts is an operation contract.
Business logic of an operation is defined in process/1
function, which is required by the Exop.Operation module
behaviour.
After the contract and business logic were defined, you can invoke the operation simply by calling run/1
function:
iex> IntegersDivision.run(a: 50, b: 5)
{:ok, "The division result is: 10"}
Return type will be either {:ok, any()}
(where the second item in the tuple is process/1
function's result) or
{:error, {:validation, map()}}
(where the map()
is validation errors map).
for more information see Operation results section
A parameter options could have various checks. Here the list of available checks:
type
required
default
numericality
equals
(exactly
)in
not_in
format
(regex
)length
inner
struct
list_item
func
allow_nil
from
subset_of
Checks whether a parameter's value is of declared type.
parameter :some_param, type: :map
Exop handle almost all Elixir types and some additional:
- :boolean
- :integer
- :float
- :string
- :tuple
- :map
- :keyword
- :list
- :atom
- :module
- :function
- :uuid
Unknown type always generates ArgumentError exception on compile time.
module
'type' means Exop expects a parameter's value to be an atom (a module name) and this module should be already loaded (ready to call it's functions)
uuid
is not actually a "type" but I placed this under :type
check because there is no reason to have dedicated :uuid
check.
Checks the presence/absence of a parameter in passed to run/1
params collection.
Given parameters collection fails the validation only if required parameter is missed,
if required parameter's value is nil
this parameter will pass this check.
parameter :param_a # the same as required: true, required by default
parameter :param_b, required: false # this parameter is not required
By default, a parameter is required (since version 1.2.0, required: true
).
If you want to specify a parameter is not required, provide required: false
.
Why? Because you might find that you repetitively type required: true
for almost every parameter in a contract. I think if you provide a parameter to an operation (define it in a contract) you expect to get it. Cases, when you need a parameter passed into an operation (and don't really care whether it is present or not), are pretty rare.
Since version 1.1.0 the behavior of this check has been changed. Check out CHANGELOG for more info.
Checks the presence of a parameter in passed to run/1
params collection,
and if the parameter is missed - assign default value to it.
parameter :some_param, default: "default value"
# default value can be also a 1-arity function output
parameter :a, type: :integer, default: &__MODULE__.default_a/1
parameter :b, type: :integer
# this function takes params given to `run/1`
def default_a(params), do: params.b + 1
#iex> YourOperation.run(b: 1)
#iex> %{a: 2, b: 1}
Checks whether a parameter's value is a number and other numeric constraints. All possible constraints are listed in the example below.
parameter :some_param, numericality: %{equal_to: 10, # (aliases: `equals`, `is`, `eq`)
greater_than: 0, # (alias: `gt`)
greater_than_or_equal_to: 10 # (aliases: `min`, `gte`),
less_than: 20, # (alias: `lt`)
less_than_or_equal_to: 10 # (aliases: `max`, `lte`)}
(alias: exactly
)
Checks whether a parameter's value exactly equals given value (with type equality).
parameter :some_param, equals: 100.5
parameter :some_param, exactly: 100.5
Checks whether a parameter's value is within a given list.
parameter :some_param, in: ~w(a b c)
Checks whether a parameter's value is not within a given list.
parameter :some_param, not_in: ~w(a b c)
(alias: regex
)
Checks wether parameter's value matches given regex.
parameter :some_param, format: ~r/foo/
parameter :some_param, regex: ~r/foo/
Checks the length of a parameter's value. The value should be one of handled types:
- list (items count)
- string (chars count)
- atom (treated as string)
- map (key-value pairs count)
- tuple (items count)
length
check is complex as numericality
(should define map of inner checks).
All possible checks are listed in the example below.
parameter :some_param, length: %{gte: 5, gt: 4, min: 5, lte: 10, lt: 9, max: 10, is: 7, in: 5..8}
Checks the inner of either Map or Keyword parameter. It applies checks described in inner
map to
related inner items.
# some_param = %{a: 3, b: "inner_b_attr"}
parameter :some_param, type: :map, inner: %{
a: [type: :integer],
b: [type: :string, length: %{min: 1, max: 6}]
}
# you can omit `type` and `inner` checks keywords in order to check inner of your parameter,
# when `type` hasn't been specified explicitly, both keyword and map types pass the `type` validation
parameter :some_param, %{
a: [type: :integer],
b: [type: :string, length: %{min: 1, max: 6}]
}
And, of course, all checks on a parent parameter (:some_param
in the example) are still applied.
Checks whether the given parameter is expected structure.
parameter :some_param, struct: %SomeStruct{}
# or
parameter :some_param, struct: SomeStruct
Checks whether each of list items conforms defined checks. An item's checks could be any that Exop offers:
# list_param = ["1234567", "7chars"]
# you can omit `type` check while you're passing a list to an operation
parameter :list_param, list_item: %{type: :string, length: %{min: 7}}
Even more complex like inner
:
# list_param = [
# %TestStruct{a: 3, b: "6chars"},
# %TestStruct{a: nil, b: "7charss"}
# ]
parameter :list_param, list_item: %{inner: %{
a: %{type: :integer},
b: %{type: :string, length: %{min: 7}}
}}
Moreover, coerce_with
and default
options are available too.
Checks whether an item is valid over custom validation function.
A validation is treated as failed if callback function returns one of results:
{:error, _your_error_message_or_payload}
:error
false
Everything else is treaded as successful validation result.
If the validation function returns either false
or :error
, the default message "not valid"
is used in your operation's validation results.
The validation function receives two arguments:
- a tuple with a validating parameter's name and value
- a map of all parameters given to an operation
Those arguments allow you to make a parameter validations which depend on other parameters and their values.
parameter :some_param, func: &__MODULE__.your_validation/2
@spec your_validation({atom() | String.t(), any()}, map()) :: any()
def your_validation({param_name, param_value}, all_received_params_map) do
# your validation logic based on given arguments is here
end
it is possible to combine :func check with others (though not preferable), just make sure this check is the last check in the list
It is a parameter attribute which allow you to have other checks for a parameter whilst have a possibility to pass nil
as the parameter's value.
If allow_nil: true
and nil
is passed as a parameter value all the parameter's checks are ignored during validation.
defmodule YourOperation do
use Exop.Operation
parameter :a, type: :integer, allow_nil: true
parameter :b, type: :integer, allow_nil: false
def process(params), do: params
end
{:ok, %{a: 1, b: 1}} = YourOperation.run(a: 1, b: 1)
{:ok, %{a: nil, b: 1}} = YourOperation.run(a: nil, b: 1)
{:error, {:validation, %{b: ["doesn't allow nil"]}}} = YourOperation.run(a: nil, b: nil)
By default (if you omit allow_nil
), a parameter is treated as allow_nil: false
This option allows you to pass a parameter to run/1
and run!/1
functions with one name and work with this parameter within an operation under another name.
defmodule YourOperation do
use Exop.Operation
parameter :a, type: :integer, from: "a"
parameter :b, type: :string, from: :bB
def process(params), do: params
end
# now you can invoke YourOperation with such params:
# Youroperation.run(%{"a" => 1, b: "1"})
# Youroperation.run(%{a: 1, bB: "1"})
# Youroperation.run(%{"a" => 1, bB: "1"})
# and get:
# {:ok, %{a: 1, b: "1"}}
The same works for parameters given as a Keyword as well (in this case :from
value should be an atom):
defmodule YourOperation do
use Exop.Operation
parameter :a, type: :integer, from: :aA
parameter :b, type: :string, from: :bB
def process(params), do: params
end
# Youroperation.run(a: 1, b: "1")
# Youroperation.run(aA: 1, bB: "1")
# {:ok, %{a: 1, b: "1"}}
Why? Simply because sometimes you're not in control of incoming parameters but don't want to map them each time you need to use'em by yourself (good example: params in Phoenix controller's action, which come as a map with string keys).
This option doesn't work for :inner
check's inner parameters currently.
Checks whether a parameter's value (list) is a subset of a defined check-list. To pass this check, all items within given into an operation parameter should be included into check-list, otherwise the check is failed.
parameter :some_param, subset_of: [1, 2, :a, "b", C]
# {:ok, _} = MyOperation.run(some_param: [1, :a, C])
# {:ok, _} = MyOperation.run(some_param: [:a])
# {:error, _} = MyOperation.run(some_param: [])
# {:error, _} = MyOperation.run(some_param: [3, :a, C])
In some cases you might want to make an 'early return' from process/1
function.
For this purpose you can call interrupt/1
function within process/1
and pass an interruption reason to it.
An operation will be interrupted and return {:interrupt, your_reason}
# ...
def process(_params) do
interrupt(%{fail: "oops"})
:ok # will not return it
end
# ...
SomeOperation.run(a: 1) # {:interrupt, %{fail: "oops"}}
run!/1
invocation doesn't affect interruption result: {:interrupt, _your_result} tuple will be returned anyway as expected and handled result.
It is possible to coerce a parameter before the contract validation, all validation checks
will be invoked on coerced parameter value.
Since coercion changes a parameter before any validation has been invoked,
default values are resolved (with :default
option) before the coercion.
The flow looks like: Resolve param default value -> Coerce -> Validate coerced
If coercion function returns an error tuple it will be treated as validation failure: an operation's invokation stops and that error tuple will be returned as a result.
:coerce_with
accepts 2-arity function. This function takes a tuple {coerced_param_name, coerced_param_value}
as the first argument and a map with all the parameters that have been passed to either run/1
or run!/1
function as the second argument.
parameter :a, type: :integer
parameter :b, type: :string, coerce_with: &__MODULE__.to_string/2
parameter :c, type: :string, coerce_with: &__MODULE__.to_string/2
def to_string({_, value}, %{a: _, b: _, c: _} = _received_params) when is_integer(value) do
Integer.to_string(c_value)
end
def to_string({_, value}, _received_params) when is_binary(c_value) do
value
end
def to_string({:c, c_value}, _received_params) do
# special coercion for :c parameter here
end
Why it is so? Because there are cases when you can use the same coercion function for multiple params and/or coerce a parameter depending on another's value and need to have all this information.
It is possible to define a policy that will be used for authorizing the possibility to invoke an operation. So far, there is a simple policy implementation and usage.
Just define a module with a bunch of functions, each takes a single argument (any type) and returns true/false (authorization result) or any term (which will be represented as authorization error reason.
defmodule MonthlyReportPolicy do
def can_read?(%{user_role: "admin"}), do: true
def can_read?("admin"), do: true
def can_read?(%User{role: "manager"}), do: true
def can_read?(:manager), do: true
def can_read?(_opts), do: false
def can_write?(%{user_role: "admin"}), do: true
def can_write?(%{user_role: "manager"}), do: false
def can_write?(_), do: [:your, "reason"] # (the result error will be: {:error, {:auth, [:your, "reason"]}})
end
In this policy two actions (checks) defined (can_read?/1
& can_write?/1
).
Every action expects an argument for a check. It's up to you how to handle this argument and turn it into the actual check.
- next step - link an operation and a policy
defmodule ReadOperation do
use Exop.Operation
policy MonthlyReportPolicy, :can_read?
parameter :user, struct: User
def process(%{user: %User{} = user}) do
# make some reading...
end
end
- finally - call
authorize/1
withinprocess/1
defmodule ReadOperation do
use Exop.Operation
policy MonthlyReportPolicy, :can_read?
parameter :user, struct: User
def process(params) do
authorize(params.user)
# make some reading...
end
end
Please, note: if authorization fails, any code after (below) auth check
will be postponed (an error {:error, {:auth, _policy_action_name}}
will be returned immediately as an operation result)
If you'd like to handle various operations fails with a certain logic (for example log it into Graylog)
you can use Exop.Fallback
.
Define a fallback module:
defmodule FallbackModule do
use Exop.Fallback
def process(operation, params, error) do
# your error handling code here
:some_fallback_result
end
end
here you need to define and implement process/3
function which takes following params:
- failed operation module
- params that were passed into the operation
- an error result which was returned by the operation
Use your fallback in operations like this:
defmodule SomeOperation do
use Exop.Operation
fallback FallbackModule, return: true
parameter :a, type: :integer
parameter :b, type: :integer
def process(%{a: a, b: b}), do: a + b
end
The results of the operation executions:
# SomeOperation will fail
iex> SomeOperation.run(a: 1, b: "2")
:some_fallback_result
# SomeOperation will be successful
iex> SomeOperation.run(a: 1, b: 2)
{:ok, 3}
During a fallback definition you can add return: true
option so in the example case
SomeOperation.run/1
will return the result of the fallback (FallbackModule.process/3
function's result - :some_fallback_result
).
If you want SomeOperation.run/1
to return original result ({:error, {:validation, %{a: ["has wrong type"]}}}
)
specify return: false
option or just omit it in a fallback definition.
If you'd like to handle a side effect successful operation (for example brodcast results to Phoenix.PubSub)
you can use Exop.Callback
.
Define a callback module:
defmodule CallbackModule do
use Exop.Callback
def process(operation, params, result, opts) do
Phoenix.PubSub.broadcast(MyApp.PubSub, :topic, result)
end
end
here you need to define and implement process/4
function which takes following params:
- operation module
- params that were passed into the operation
- an successful result which was returned by the operation
- opts keyword list for additional metadata
Use your callback in operations like this:
defmodule SomeOperation do
use Exop.Operation
callback CallbackModule
parameter :a, type: :integer
parameter :b, type: :integer
def process(%{a: a, b: b}), do: a + b
end
You can use opts
keyword list to add needed metadata to your call like this:
defmodule CallbackModule do
use Exop.Callback
def process(operation, params, result, opts) do
Phoenix.PubSub.broadcast(MyApp.PubSub, opts[:topic], result)
end
end
and then use the callback like this
defmodule SomeOperation do
use Exop.Operation
callback CallbackModule, topic: :some_topic
parameter :a, type: :integer
parameter :b, type: :integer
def process(%{a: a, b: b}), do: a + b
end
As said earlier, operations in most cases called by run/1
function. This function
receives parameters collection. It's not required to pass to run/1
function parameters
only described in the operation's contract, but only described parameters will be validated and gived to process/1
function, all other will be filtered out from the process.
filtering out parameters which are not defined in a contract is here to support the whole idea of defining the right operation's contract and take care of what you really need for a certain business process / function
run/1
function validates received parameters over the contract and if all parameters passed
the validation, the run/1
function calls the code defined in process/1
function.
iex> SomeOperation.run(param1: 1, param2: "2")
_some_result_
If at least one of the given parameters didn't pass the validation process/1
function's code
will not be invoked and corresponding warning in the application's log will appear.
There is "bang" version of run/1
exists. Function run!/1
does the same things that its sibling does,
the only difference is a result of invocation, it might be:
- if a contract validation passed - the actual result of an operation (result of a code, described in
process/1
) - if a contract validation failed - an error
Exop.Validation.ValidationError
is raised - if an operation returns an error tuple - an error
Exop.Operation.ErrorResult
is raised - in case of manual interruption -
{:interrupt, _reason}
You always can bypass the validation simply by calling process/1
function itself, if needed.
If received parameters passed a contract validation, a code defined in process/1
will be invoked.
Or you will receive @type validation_error :: {:error, :validation_failed, map()}
as a result otherwise.
map()
as errors reasons might look like this:
%{param1: ["has wrong type"], param2: ["is required", "must be equal to 3"]}
An operation can return one of results listed below (depends on passed in params and operation definition):
- an operation was completed successfully:
{:error, _your_error_reason_}
(if an :error-tuple (any length, but:error
atom should be the first element) was returned byprocess/1
function){:ok, any()}
(otherwise, even if{:ok, _your_result_}
tuple was returned byprocess/1
function)
- a contract validation failed:
{:error, {:validation, map()}}
- if
interrupt/1
was invoked:{:interrupt, any()}
- policy check failed:
{:error, {:auth, :undefined_policy}}
{:error, {:auth, :undefined_action}}
{:error, {:auth, atom()}}
For the "bang" version of run/1
see results description above.
Sometimes you need to aggregate/group 'atom' operations into a single one operation responsible for
some complex business process/logic.
You have a few approaches to do it (with
for example) but mb you'll find Exop.Chain
more handy.
Exop.Chain
provides a simple way to organize a number of Exop.Operation modules into an invocation chain.
defmodule CreateUser do
use Exop.Chain
alias Operations.{User, Backoffice, Notifications}
operation User.Create
operation Backoffice.SaveStats
operation Notifications.SendEmail
# or you can use step alias:
# step User.Create
# step Backoffice.SaveStats
# step Notifications.SendEmail
end
This is how invoke this chain:
iex> CreateUser.run(name: "User Name", age: 37, gender: "m")
Exop.Chain
defines run/1
function under the hood (like common operations do) that accepts keyword()
, map()
or struct()
as params.
Those params will be passed into the first operation in the chain.
Bear in mind that each of chained operations (except the first one) awaits a returned result of
a previous operation as incoming params.
So in the example above CreateUser.run(name: "User Name", age: 37, gender: "m")
will invoke
the chain by passing [name: "User Name", age: 37, gender: "m"]
params to the first User.Create
operation.
The result of User.Create
operation will be passed to Backoffice.SaveStats
operation as its params and so on.
Once any of operations in the chain returns non-ok-tuple result (error result, interruption, auth error etc.)
the chain execution interrupts and error result returned (as the chain (CreateUser
) result).
You can pass additional parameters to any operation in a chain (with either an exact value or 0-arity function):
defmodule CreateUser do
use Exop.Chain
alias Operations.{User, Backoffice, Notifications}
operation User.Create
operation Backoffice.SaveStats, logger: MyFancyLoggerModule
# or
operation Backoffice.SaveStats, logger: &__MODULE__.logger/0
operation Notifications.SendEmail
def logger, do: MyFancyLoggerModule
end
use Exop.Chain
can take :name_in_error
option, when it is set to true
a failed operation in a chain returns the operation's module name as the first elements of output tuple {YourOperation, {:error, _}}
defmodule YourChain do
use Exop.Chain, name_in_error: true
operation Operation1
operation Operation2Fail
operation Operation3
end
iex> YourChain.run(a: "1", b: 2)
{Operation2Fail, {:error, {:validation, %{a: ["has wrong type"]}}}}
name_in_error: true
doesn't affect operations with a fallback defined (unmodified fallback result is returned).
It is possible to define an invokation condition for an operation in a chain. Meaning if the condition is truthy - the operation will be invoked.
The condition can be defined with if: your_func/1
option given to an operation.
defmodule YourChain do
use Exop.Chain
# operation/step - they are synonims in the context of Exop.Chain
operation Operation1
operation MultiplyByHundred, if: &__MODULE__.is_it_good_to_go?/1
operation DivisionByTen
def is_it_good_to_go?(previous_operation_output) do
# here your condition logic which should return a boolean
end
end
A condition function receives a single argument - the previous operation's output (the second element of {:ok, _}
tuple, not the tuple itself), turned into map (if the output is Keyword list) so it is easier to pattern-match.
An operation is invoked if a condition function returns true
, otherwise the operation won't be invoked.
And of course a chain invokation interrupts if the previous operation's result wasn't successful (is not {:ok, _}
tuple). The previous operation's result is not changed in this case.
If for some reason you need to change incoming parameters (which are the previous operation result) for your operation(-s) in the chain you can do it with :coerce_with
option. This option is provided for a particular operation (step) and refers to a 1-arity callback function. This single argument is a map of incoming parameters (the previous operation result). It is converted into a map even if the previous operation returns a keyword list.
defmodule YourChain do
use Exop.Chain
operation Sum
operation MultiplyByHundred, coerce_with: &__MODULE__.coerce/1
operation DivisionByTen
def coerce(%{a: a} = params), do: %{params | a: a * 10}
end
The coercion works right before any validation/invokation (how it is with the regular operation). It means that first of all the previous operation's result is coerced and then everything else happens based on those coerced parameters.
Coercion works well with other options like additional parameters and conditional invocation.
Here is the order of an operation handling in a chain:
- incoming parameters coercion
- additional parameters are added
- invocation condition is checked
Copyright © 2016 - 2020 Andrey Chernykh ( andrei.chernykh@gmail.com )
This work is free. You can redistribute it and/or modify it under the
terms of the MIT License. See the LICENSE file for more details.