@callback all_flag_names() :: {:ok, [atom()]} | {:error, any()}

Retrieves all the names of the persisted flags.

@callback all_flags() :: {:ok, [FunWithFlags.Flag.t()]} | {:error, any()}

Retrieves all the persisted flags.

@callback delete(flag_name :: atom()) :: {:ok, FunWithFlags.Flag.t()} | {:error, any()}

Deletes an entire flag, identified by name.

@callback delete(flag_name :: atom(), gate :: FunWithFlags.Gate.t()) ::
  {:ok, FunWithFlags.Flag.t()} | {:error, any()}

Deletes a gate from a flag, identified by name.

@callback get(flag_name :: atom()) :: {:ok, FunWithFlags.Flag.t()} | {:error, any()}

Retrieves a flag by name.

@callback put(flag_name :: atom(), gate :: FunWithFlags.Gate.t()) ::
  {:ok, FunWithFlags.Flag.t()} | {:error, any()}

Persists a gate for a flag, identified by name.

@callback worker_spec() :: Supervisor.child_spec() | nil

A persistent adapter should return either a child specification if it needs any process to be started and supervised, or nil if it does not.

For example, the builtin Redis persistence adapter implements this function by delegating to Redix.child_spec/1 because it needs the Redix processes to work. On the other hand, the builtin Ecto adapter implements this function by returning nil, because the Ecto repo is provided to this package by the host application, and it's assumed that the Ecto process tree is started and supervised somewhere else.

This custom worker_spec/0 function is used instead of the typical child_spec/1 function because this function can return nil if the adapter doesn't need to be supervised, whereas child_spec/1 must return a valid child spec map.