Provides a standard API to handle keyword-list-based options.

This module began its life as a vendored form of NimbleOptions, meaning that we copied it from NimbleOptions into Spark. We had various features to add to it, and the spirit of nimble options is to be as lightweight as possible. With that in mind, we were advised to vendor it. We would like to thank the authors of NimbleOptions for their excellent work, and their blessing to transplant their work into Spark.

Spark.Options allows developers to create schemas using a pre-defined set of options and types. The main benefits are:

• A single unified way to define simple static options
• Config validation against schemas
• Automatic doc generation
• More types over what is provided by NimbleOptions
• Compile time validators that are highly optimized and produce structs. See Spark.Options.Validator.
• Shared logic between Spark DSLs and options lists.

Schema Options

These are the options supported in a schema. They are what defines the validation for the items in the given schema.

• :type - The type of the option item. The default value is :any.

• :required (boolean/0) - Defines if the option item is required. The default value is false.

• :default (term/0) - The default value for the option item if that option is not specified. This value is validated according to the given :type. This means that you cannot have, for example, type: :integer and use default: "a string".

• :keys (keyword/0) - Available for types :keyword_list, :non_empty_keyword_list, and :map, it defines which set of keys are accepted for the option item. The value of the :keys option is a schema itself. For example: keys: [foo: [type: :atom]]. Use :* as the key to allow multiple arbitrary keys and specify their schema: keys: [*: [type: :integer]].

• :deprecated (String.t/0) - Defines a message to indicate that the option item is deprecated. The message will be displayed as a warning when passing the item.

• :private? (boolean/0) - Defines an option as private, used with Spark.Options.Validator The default value is false.

• :hide (one or a list of atom/0) - A list of keys that should be hidden when generating documentation

• :as (atom/0) - A name to remap the option to when used in DSLs. Not supported in regular option parsing

• :snippet (String.t/0) - A snippet to use when autocompleting DSLs. Not supported in regular option parsing

• :links (term/0) - A keyword list of links to include in DSL documentation for the option item.

• :doc (String.t/0 or false) - The documentation for the option item.

• :subsection (String.t/0) - The title of separate subsection of the options' documentation

• :type_doc (String.t/0 or false) - The type doc to use in the documentation for the option item. If false, no type documentation is added to the item. If it's a string, it can be anything. For example, you can use "a list of PIDs", or you can use a typespec reference that ExDoc can link to the type definition, such as "`t:binary/0`". You can use Markdown in this documentation. If the :type_doc option is not present, Spark.Options tries to produce a type documentation automatically if it can do it unambiguously. For example, if type: :integer, Spark.Options will use integer/0 as the auto-generated type doc.

• :type_spec (Macro.t/0) - The quoted spec to use in the typespec for the option item. You should use this when the auto-generated spec is not specific enough. For example, if you are performing custom validation on an option (with the {:custom, ...} type), then the generated type spec for that option will always be term/0, but you can use this option to customize that. The value for this option must be a quoted Elixir term. For example, if you have an :exception option that is validated with a {:custom, ...} type (based on is_exception/1), you can override the type spec for that option to be quote(do: Exception.t()). Available since v1.1.0.

Types

• :any - Any type.

• :keyword_list - A keyword list.

• :non_empty_keyword_list - A non-empty keyword list.

• {:keyword_list, schema} - A keyword list matching the given options schema.

• :non_empty_keyword_list - A non-empty keyword list.

• {:non_empty_keyword_list, schema} - A non-empty keyword list matching the given options schema.

• :map - A map consisting of :atom keys. Shorthand for {:map, :atom, :any}. Keys can be specified using the keys option.

• {:map, key_type, value_type} - A map consisting of key_type keys and value_type values.

• :atom - An atom.

• :string - A string.

• :boolean - A boolean.

• :integer - An integer.

• :non_neg_integer - A non-negative integer.

• :pos_integer - A positive integer.

• :float - A float.

• :timeout - A non-negative integer or the atom :infinity.

• :pid - A PID (process identifier).

• :reference - A reference (see reference/0).

• nil - The value nil itself. Available since v1.0.0.

• :mfa - A named function in the format {module, function, arity} where arity is a list of arguments. For example, {MyModule, :my_fun, [arg1, arg2]}.

• :mod_arg - A module along with arguments, such as {MyModule, arguments}. Usually used for process initialization using start_link and similar. The second element of the tuple can be any term.

• :fun - Any function.

• {:fun, arity} - Any function with the specified arity.

• {:fun, args_types} - A function with the specified arguments.

• {:fun, args_types, return_type} - A function with the specified arguments and return type.

• {:in, choices} or {:one_of, choices} - A value that is a member of one of the choices. choices should be a list of terms or a Range. The value is an element in said list of terms, that is, value in choices is true.

• {:struct, struct_name} - An instance of the struct type given.

• :struct - An instance of any struct

• {:tagged_tuple, tag, inner_type} - maps to {tag, type}

• {:spark_behaviour, behaviour} - expects a module that implements the given behaviour, and can be specified with options, i.e mod or {mod, [opt: :val]}

• {:spark_behaviour, behaviour, builtin_module} - Same as the above, but also accepts a builtin_module. The builtin_module is used to provide additional options for the elixir_sense plugin.

• {:spark_function_behaviour, behaviour, {function_mod, arity}} - expects a module that implements the given behaviour, and can be specified with options, i.e mod or {mod, [opt: :val]}, that also has a special module that supports being provided an anonymous function or MFA as the :fun option.

• {:spark_function_behaviour, behaviour, builtin_module, {function_mod, arity}} - Same as the above, but also accepts a builtin_module. The builtin_module is used to provide additional options for the elixir_sense plugin.

• {:behaviour, behaviour} - expects a module that implements a given behaviour.

• {:protocol, protocol} - expects a value for which the protocol is implemented.

• {:impl, protocol} - expects a module for which the protocol is implemented.

• {:spark, dsl_module} - expects a module that is a Spark.Dsl

• {:mfa_or_fun, arity} - expects a function or MFA of a corresponding arity.

• {:spark_type, module, builtin_function} - a behaviour that defines builtin_function/0 that returns a list of atoms that map to built in variations of that thing.

• {:spark_type, module, builtin_function, templates} - same as the above, but includes additional templates for elixir_sense autocomplete

• :literal -> any literal value. Maps to :any, but is used for documentation.

• {:literal, value} -> exactly the value specified.

• :quoted -> retains the quoted value of the code provided to the option

• {:wrap_list, type} -> Allows a single value or a list of values.

• {:custom, mod, fun, args} - A custom type. The related value must be validated by mod.fun(values, ...args). The function should return {:ok, value} or {:error, message}.

• {:or, subtypes} - A value that matches one of the given subtypes. The value is matched against the subtypes in the order specified in the list of subtypes. If one of the subtypes matches and updates (casts) the given value, the updated value is used. For example: {:or, [:string, :boolean, {:fun, 2}]}. If one of the subtypes is a keyword list or map, you won't be able to pass :keys directly. For this reason, :keyword_list, :non_empty_keyword_list, and :map are special cased and can be used as subtypes with {:keyword_list, keys}, {:non_empty_keyword_list, keys} or {:map, keys}. For example, a type such as {:or, [:boolean, keyword_list: [enabled: [type: :boolean]]]} would match either a boolean or a keyword list with the :enabled boolean option in it.

• {:list, subtype} - A list where all elements match subtype. subtype can be any of the accepted types listed here. Empty lists are allowed. The resulting validated list contains the validated (and possibly updated) elements, each as returned after validation through subtype. For example, if subtype is a custom validator function that returns an updated value, then that updated value is used in the resulting list. Validation fails at the first element that is invalid according to subtype. If subtype is a keyword list or map, you won't be able to pass :keys directly. For this reason, :keyword_list, :non_empty_keyword_list, and :map are special cased and can be used as the subtype by using {:keyword_list, keys}, {:non_empty_keyword_list, keys} or {:keyword_list, keys}. For example, a type such as {:list, {:keyword_list, enabled: [type: :boolean]}} would a list of keyword lists, where each keyword list in the list could have the :enabled boolean option in it.

• {:tuple, list_of_subtypes} - A tuple as described by tuple_of_subtypes. list_of_subtypes must be a list with the same length as the expected tuple. Each of the list's elements must be a subtype that should match the given element in that same position. For example, to describe 3-element tuples with an atom, a string, and a list of integers you would use the type {:tuple, [:atom, :string, {:list, :integer}]}. Available since v0.4.1.

Example

iex> schema = [
...>   producer: [
...>     type: :non_empty_keyword_list,
...>     required: true,
...>     keys: [
...>       module: [required: true, type: :mod_arg],
...>       concurrency: [
...>         type: :pos_integer,
...>       ]
...>     ]
...>   ]
...> ]
...>
...> config = [
...>   producer: [
...>     concurrency: 1,
...>   ]
...> ]
...>
...> {:error, %Spark.Options.ValidationError{} = error} = Spark.Options.validate(config, schema)
...> Exception.message(error)
"required :module option not found, received options: [:concurrency] (in options [:producer])"

Nested Option Items

Spark.Options allows option items to be nested so you can recursively validate any item down the options tree.

Example

iex> schema = [
...>   producer: [
...>     required: true,
...>     type: :non_empty_keyword_list,
...>     keys: [
...>       rate_limiting: [
...>         type: :non_empty_keyword_list,
...>         keys: [
...>           interval: [required: true, type: :pos_integer]
...>         ]
...>       ]
...>     ]
...>   ]
...> ]
...>
...> config = [
...>   producer: [
...>     rate_limiting: [
...>       interval: :oops!
...>     ]
...>   ]
...> ]
...>
...> {:error, %Spark.Options.ValidationError{} = error} = Spark.Options.validate(config, schema)
...> Exception.message(error)
"invalid value for :interval option: expected positive integer, got: :oops! (in options [:producer, :rate_limiting])"

Validating Schemas

Each time validate/2 is called, the given schema itself will be validated before validating the options.

In most applications the schema will never change but validating options will be done repeatedly.

To avoid the extra cost of validating the schema, it is possible to validate the schema once, and then use that valid schema directly. This is done by using the new!/1 function first, and then passing the returned schema to validate/2.

Create the Schema at Compile Time

If your option schema doesn't include any runtime-only terms in it (such as anonymous functions), you can call new!/1 to validate the schema and returned a compiled schema at compile time. This is an efficient way to avoid doing any unnecessary work at runtime. See the example below for more information.

Example

iex> raw_schema = [
...>   hostname: [
...>     required: true,
...>     type: :string
...>   ]
...> ]
...>
...> schema = Spark.Options.new!(raw_schema)
...> Spark.Options.validate([hostname: "elixir-lang.org"], schema)
{:ok, hostname: "elixir-lang.org"}

Calling new!/1 from a function that receives options will still validate the schema each time that function is called. Declaring the schema as a module attribute is supported:

@options_schema Spark.Options.new!([...])

This schema will be validated at compile time. Calling docs/1 on that schema is also supported.