View Source Unifex.Specs.DSL (Unifex v1.2.0)

Module exporting macros that can be used to define Unifex specs.

Example of such specs is provided below:

module Membrane.Element.Mad.Decoder.Native

type position :: :manager | :developer | :intern | :product_owner

type personal_data :: %PersonalData{
  age: int,
  experience: int,
  name: string
}

spec get_salary(person :: personal_data, person_position :: position) ::
       {:ok :: label, salary :: int}
       {:error :: label, :unemployed :: label}

spec create() :: {:ok :: label, state}

spec decode_frame(payload, offset :: int, state) ::
       {:ok :: label, {payload, bytes_used :: long, sample_rate :: long, channels :: int}}
       | {:error :: label, :buflen :: label}
       | {:error :: label, :malformed :: label}
       | {:error :: label, {:recoverable :: label, bytes_to_skip :: int}}

dirty :cpu, decode_frame: 3

sends {:example_msg :: label, number :: int}

According to this specification, module Membrane.Element.Mad.Decoder.Native should contain 2 functions: create/0 and decode_frame/3 (which is a cpu-bound dirty NIF). The module should use Unifex.Loader to provide access to these functions. What is more, messages of the form {:example_msg, integer} can be sent from the native code to erlang processes.

Following types definitions will be generated and will be available from native code

#ifdef __cplusplus
enum Position { MANAGER, DEVELOPER, INTERN, PRODUCT_OWNER };
#else
enum Position_t { MANAGER, DEVELOPER, INTERN, PRODUCT_OWNER };
typedef enum Position_t Position;
#endif

#ifdef __cplusplus
struct personal_data {
  int age;
  int experience;
  char *name;
};
#else
struct personal_data_t {
  int age;
  int experience;
  char *name;
};
typedef struct personal_data_t personal_data;
#endif

The generated boilerplate would require implementation of the following functions:

UNIFEX_TERM get_salary(UnifexEnv *env, personal_data person, Position person_position);
UNIFEX_TERM create(UnifexEnv* env);
UNIFEX_TERM decode_frame(UnifexEnv* env, UnifexPayload * payload, int offset, State* state);

Also the following functions that should be called to return results will be generated:

UNIFEX_TERM get_salary_result_ok(UnifexEnv *env, int salary);
UNIFEX_TERM get_salary_result_error_unemployed(UnifexEnv *env);
UNIFEX_TERM create_result_ok(UnifexEnv* env, State* state);
UNIFEX_TERM decode_frame_result_ok(UnifexEnv* env, UnifexPayload * payload,
                                   long bytes_used, long sample_rate, int channels);
UNIFEX_TERM decode_frame_result_error_buflen(UnifexEnv* env);
UNIFEX_TERM decode_frame_result_error_malformed(UnifexEnv* env);
UNIFEX_TERM decode_frame_result_error_recoverable(UnifexEnv* env, int bytes_to_skip);

See docs for appropriate macros for more details.

Summary

Functions

@arg

Defines native function documentation. Documentation will be visible in the generated docs.

callback(hook, fun \\ nil)

Defines names of callbacks invoked on specified hook.

dirty(type, funs)

Macro used for marking functions as dirty, i.e. performing long cpu-bound or io-bound operations.

interface(interface)

Specifies interface, for example NIF or CNode.

module(module)

Defines module that exports native functions to Elixir world.

sends(spec)

Defines terms that can be sent from the native code to elixir processes.

spec(spec)

Defines native function specification.

state_type(state_type)

Defines the state type, required for using Unifex.CodeGenerator.BaseTypes.State.

type(struct)

Contains user custom type specification

Functions

@arg

(macro)

Defines native function documentation. Documentation will be visible in the generated docs.

The documentation should be in the form of

  @doc """
  Documentation...
  """
  spec function_name...

callback(hook, fun \\ nil)

(macro)

Defines names of callbacks invoked on specified hook.

The available hooks are:

:load - invoked when the library is loaded. Callback must have the following typing:
int on_load(UnifexEnv *env, void ** priv_data)
The callback receives an env and a pointer to private data that is initialized with NULL and can be set to whatever should be passed to other callbacks. If callback returns anything else than 0, the library fails to load.
:upgrade - called when the library is loaded while there is old code for this module with a native library loaded. Compared to :load, it also receives old_priv_data:
int on_upgrade(UnifexEnv* env, void** priv_data, void** old_priv_data)
Both old and new private data can be modified If this callback is not defined, the module code cannot be hot-swapped. Non-zero return value also prevents code upgrade.
:unload - called when the code for module is unloaded. It has the following declaration:
void on_unload(UnifexEnv *env, void * priv_data)

dirty(type, funs)

(macro)

Macro used for marking functions as dirty, i.e. performing long cpu-bound or io-bound operations.

The macro should be used the following way:

dirty type, function1: function1_arity, ...

when type is one of:

:cpu - marks function as CPU-bound (maps to the ERL_NIF_DIRTY_JOB_CPU_BOUND erlang flag)
:io - marks function as IO-bound (maps to the ERL_NIF_DIRTY_JOB_IO_BOUND erlang flag)

interface(interface)

(macro)

Specifies interface, for example NIF or CNode.

It should be a module name (or list of module names) that will be prepended with Unifex.CodeGenerators. If no interface is specified, it is automatically detected basing on Bundlex.Project specification.

module(module)

(macro)

Defines module that exports native functions to Elixir world.

The module needs to be defined manually, but it can use Unifex.Loader to have functions declared with spec/1 automatically defined.

sends(spec)

(macro)

Defines terms that can be sent from the native code to elixir processes.

Creates native function that can be invoked to send specified data. Name of the function starts with send_ and is constructed from labels.

spec(spec)

(macro)

Defines native function specification.

The specification should be in the form of

spec function_name(parameter1 :: parameter1_type, some_type, parameter3 :: parameter3_type, ...) ::
    {:label1 :: label, {result_value1 :: result_value1_type, some_type2, ...}}
    | {:label2 :: label, other_result_value2 :: other_result_value2_type}

Parameters

Specs for parameters can either take the form of parameter1 :: parameter1_type which will generate parameter with name parameter1 of type parameter1_type The other form - just a name, like some_type - will generate parameter some_type of type some_type.

Custom types can be added by creating modules Unifex.CodeGenerator.BaseTypes.Type that implement Unifex.CodeGenerator.BaseType behaviour. Then, they can by used in specs as type.

Each generated function gets additional UnifexEnv* env as the first parameter implicitly.

Returned values

Specs for returned values contain a special type - label. An atom of type label will be put literally in returned values by the special function generated for each spec. Names of the generated functions start with Elixir function name (e.g. create) followed by _result_ part and then all the labels joined with _.

Example

The example is provided in the moduledoc of this module.

state_type(state_type)

(macro)

Defines the state type, required for using Unifex.CodeGenerator.BaseTypes.State.

type(struct)

(macro)

Contains user custom type specification

Currently supported types are enums and structs

The struct specification should be in the form of

type my_struct :: %My.Struct{
  field1: type1,
  field2: type2,
  ...
}

The enum specification should be in the form of

type my_enum :: :option_one | :option_two | :option_three | ...

Struct or enums specified in such way can be used in like any other supported type, E.g.

spec my_function(in_enum :: my_enum) :: {:ok :: label, out_struct :: my_struct}

Elixir definition of %My.Struct{} should contain every field listed in specification and is not generated by Unifex