Feature: Add Streaming API

CHANGELOG.md

@@ -9,6 +9,13 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+- Added a new Streaming API that processes data in chunks, reducing peak memory
+  usage when handling large datasets or network streams
+  - Introduced `Msgpack.encode_stream/2` to lazily encode a stream of Elixir
+    terms one by one
+  - Introduced `Msgpack.decode_stream/2` to lazily decode a stream of
+    MessagePack objects, capable of handling data that arrives in multiple
+    chunks
 - Added CI workflow to run tests against supported Elixir versions
 
 ### Changed

README.md

@@ -20,6 +20,9 @@ types.
   limits to mitigate resource exhaustion from malformed or malicious payloads.
 - **Telemetry Integration:** Emits standard `:telemetry` events for integration
   with monitoring tools.
+- **Streaming API:** Process large collections or continuous data streams with
+  low memory overhead using `Msgpack.encode_stream/2` and
+  `Msgpack.decode_stream/2`.
 
 ## Installation
 
@@ -50,6 +53,27 @@ iex> Msgpack.decode!(<<0xC1>>)
 ** (Msgpack.DecodeError) Unknown type prefix: 193. The byte `0xC1` is not a valid MessagePack type marker.
 ```
 
+### Streaming Large Collections
+
+For datasets that may not fit in memory, you can use the streaming API, which
+processes one term at a time.
+
+```elixir
+# Create a lazy stream of terms to be encoded.
+iex> terms = Stream.cycle([1, "elixir", true])
+
+# The output is a lazy stream of encoded binaries.
+iex> encoded_stream = Msgpack.encode_stream(terms)
+
+# The stream is only consumed when you enumerate it.
+iex> encoded_stream |> Stream.take(3) |> Enum.to_list()
+[
+  {:ok, <<1>>},
+  {:ok, <<166, 101, 108, 105, 120, 105, 114>>},
+  {:ok, <<195>>}
+]
+```
+
 ## Full Documentation
 
 For detailed information on all features, options, and functions, see the [full

lib/msgpack.ex

@@ -49,6 +49,8 @@ defmodule Msgpack do
 
   alias Msgpack.Encoder
   alias Msgpack.Decoder
+  alias Msgpack.StreamEncoder
+  alias Msgpack.StreamDecoder
   alias Msgpack.EncodeError
   alias Msgpack.DecodeError
 
@@ -314,4 +316,63 @@ defmodule Msgpack do
         raise DecodeError, reason: reason
     end
   end
+
+  @doc """
+  Encodes a stream of Elixir terms into a stream of MessagePack binaries.
+
+  Each term in the input enumerable is encoded individually. The output stream
+  will contain `{:ok, binary}` tuples for successful encodings or `{:error,
+  reason}` tuples for failures.
+
+  This function delegates to `Msgpack.StreamEncoder.encode/2`.
+
+  ## Options
+
+  Accepts the same options as `Msgpack.encode/2`.
+
+  ## Examples
+
+  ```elixir
+  iex> terms = [1, "elixir", :world]
+  iex> Msgpack.encode_stream(terms, atoms: :string) |> Enum.to_list()
+  [
+    {:ok, <<1>>},
+    {:ok, <<166, 101, 108, 105, 120, 105, 114>>},
+    {:ok, <<165, 119, 111, 114, 108, 100>>}
+  ]
+  ```
+  """
+  @spec encode_stream(Enumerable.t(), StreamEncoder.opts_t()) :: StreamEncoder.t()
+  def encode_stream(enumerable, opts \\ []) do
+    StreamEncoder.encode(enumerable, opts)
+  end
+
+  @doc """
+  Decodes a stream of MessagePack binaries into a stream of Elixir terms.
+
+  This function provides a streaming, lazy interface for decoding, making it
+  suitable for handling large datasets that do not fit into memory.
+
+  It delegates to `Msgpack.StreamDecoder.decode/2`.
+
+  For more detailed information on behavior, see the `Msgpack.StreamDecoder`
+  module documentation.
+
+  ## Options
+
+  Accepts the same options as `Msgpack.decode/2`.
+
+  ## Examples
+
+  ```elixir
+  iex> objects = [1, "elixir", true]
+  iex> stream = Enum.map(objects, &Msgpack.encode!/1)
+  iex> Msgpack.decode_stream(stream) |> Enum.to_list()
+  [1, "elixir", true]
+  ```
+  """
+  @spec decode_stream(Enumerable.t(binary()), StreamDecoder.opts_t()) :: StreamDecoder.t()
+  def decode_stream(enumerable, opts \\ []) do
+    StreamDecoder.decode(enumerable, opts)
+  end
 end

lib/msgpack/decoder.ex

@@ -3,21 +3,14 @@ defmodule Msgpack.Decoder do
   Handles the logic of decoding a MessagePack binary into an Elixir term.
   """
 
-  @default_max_depth 100
-  @default_max_byte_size 10_000_000 # 10MB
-
-  # The number of gregorian seconds from year 0 to the Unix epoch. This is a constant.
-  @epoch_offset :calendar.datetime_to_gregorian_seconds({{1970, 1, 1}, {0, 0, 0}})
+  alias Msgpack.Decoder.Internal
 
   @spec decode(binary(), keyword()) :: {:ok, term()} | {:error, term()}
   def decode(binary, opts \\ []) do
-    merged_opts =
-      opts
-      |> Keyword.put_new(:max_depth, @default_max_depth)
-      |> Keyword.put_new(:max_byte_size, @default_max_byte_size)
+    merged_opts = Keyword.merge(default_opts(), opts)
 
     try do
-      case do_decode(binary, merged_opts) do
+      case Internal.decode(binary, merged_opts) do
         {:ok, {term, <<>>}} ->
           {:ok, term}
 
@@ -33,241 +26,13 @@ defmodule Msgpack.Decoder do
     end
   end
 
-  # ==== Nil ====
-  defp do_decode(<<0xC0, rest::binary>>, _opts), do: {:ok, {nil, rest}}
-
-  # ==== Boolean ====
-  defp do_decode(<<0xC3, rest::binary>>, _opts), do: {:ok, {true, rest}}
-  defp do_decode(<<0xC2, rest::binary>>, _opts), do: {:ok, {false, rest}}
-
-  # ==== Integers ====
-  # ==== Positive Fixint ====
-  defp do_decode(<<int::8, rest::binary>>, _opts) when int < 128 do
-    {:ok, {int, rest}}
-  end
-
-  # ==== Negative Fixint ====
-  defp do_decode(<<int::signed-8, rest::binary>>, _opts) when int >= -32 and int < 0 do
-    {:ok, {int, rest}}
-  end
-
-  # ==== Unsigned Integers ====
-  defp do_decode(<<0xCC, int::8, rest::binary>>, _opts), do: {:ok, {int, rest}}
-  defp do_decode(<<0xCD, int::16, rest::binary>>, _opts), do: {:ok, {int, rest}}
-  defp do_decode(<<0xCE, int::32, rest::binary>>, _opts), do: {:ok, {int, rest}}
-  defp do_decode(<<0xCF, int::64, rest::binary>>, _opts), do: {:ok, {int, rest}}
-
-  # ==== Signed Integers ====
-  defp do_decode(<<0xD0, int::signed-8, rest::binary>>, _opts), do: {:ok, {int, rest}}
-  defp do_decode(<<0xD1, int::signed-16, rest::binary>>, _opts), do: {:ok, {int, rest}}
-  defp do_decode(<<0xD2, int::signed-32, rest::binary>>, _opts), do: {:ok, {int, rest}}
-  defp do_decode(<<0xD3, int::signed-64, rest::binary>>, _opts), do: {:ok, {int, rest}}
-
-  # ==== Floats ====
-  defp do_decode(<<0xCA, float::float-32, rest::binary>>, _opts), do: {:ok, {float, rest}}
-  defp do_decode(<<0xCB, float::float-64, rest::binary>>, _opts), do: {:ok, {float, rest}}
-
-  # ==== Strings ====
-  defp do_decode(<<prefix, rest::binary>>, opts) when prefix >= 0xA0 and prefix <= 0xBF do
-    size = prefix - 0xA0
-    decode_string(rest, size, opts)
-  end
-
-  defp do_decode(<<0xD9, size::8, rest::binary>>, opts), do: decode_string(rest, size, opts)
-  defp do_decode(<<0xDA, size::16, rest::binary>>, opts), do: decode_string(rest, size, opts)
-  defp do_decode(<<0xDB, size::32, rest::binary>>, opts), do: decode_string(rest, size, opts)
-
-  # ==== Raw Binary ====
-  defp do_decode(<<0xC4, size::8, rest::binary>>, opts), do: decode_binary(rest, size, opts)
-  defp do_decode(<<0xC5, size::16, rest::binary>>, opts), do: decode_binary(rest, size, opts)
-  defp do_decode(<<0xC6, size::32, rest::binary>>, opts), do: decode_binary(rest, size, opts)
-
-  # ==== Arrays ====
-  defp do_decode(<<prefix, rest::binary>>, opts) when prefix >= 0x90 and prefix <= 0x9F do
-    size = prefix - 0x90
-    decode_array(rest, size, opts)
-  end
-
-  defp do_decode(<<0xDC, size::16, rest::binary>>, opts), do: decode_array(rest, size, opts)
-  defp do_decode(<<0xDD, size::32, rest::binary>>, opts), do: decode_array(rest, size, opts)
-
-  # ==== Maps ====
-  defp do_decode(<<prefix, rest::binary>>, opts) when prefix >= 0x80 and prefix <= 0x8F do
-    size = prefix - 0x80
-    decode_map(rest, size, opts)
-  end
-
-  defp do_decode(<<0xDE, size::16, rest::binary>>, opts), do: decode_map(rest, size, opts)
-  defp do_decode(<<0xDF, size::32, rest::binary>>, opts), do: decode_map(rest, size, opts)
-
-  # ==== Extensions & Timestamps ====
-  # ==== Fixext ====
-  defp do_decode(<<0xD4, type::signed-8, data::binary-size(1), rest::binary>>, opts),
-    do: decode_ext(type, data, rest, opts)
-
-  defp do_decode(<<0xD5, type::signed-8, data::binary-size(2), rest::binary>>, opts),
-    do: decode_ext(type, data, rest, opts)
-
-  defp do_decode(<<0xD6, type::signed-8, data::binary-size(4), rest::binary>>, opts),
-    do: decode_ext(type, data, rest, opts)
-
-  defp do_decode(<<0xD7, type::signed-8, data::binary-size(8), rest::binary>>, opts),
-    do: decode_ext(type, data, rest, opts)
-
-  defp do_decode(<<0xD8, type::signed-8, data::binary-size(16), rest::binary>>, opts),
-    do: decode_ext(type, data, rest, opts)
-
-  # ==== Ext ====
-  defp do_decode(<<0xC7, len::8, type::signed-8, data::binary-size(len), rest::binary>>, opts),
-    do: decode_ext(type, data, rest, opts)
-
-  defp do_decode(<<0xC8, len::16, type::signed-8, data::binary-size(len), rest::binary>>, opts),
-    do: decode_ext(type, data, rest, opts)
-
-  defp do_decode(<<0xC9, len::32, type::signed-8, data::binary-size(len), rest::binary>>, opts),
-    do: decode_ext(type, data, rest, opts)
-
-  # ==== Unknown types ====
-  defp do_decode(<<prefix, _rest::binary>>, _opts) do
-    {:error, {:unknown_prefix, prefix}}
-  end
-
-  defp do_decode(<<>>, _opts) do
-    {:error, :unexpected_eof}
-  end
-
-  # ==== Helpers ====
-  defp decode_string(binary, size, opts) do
-    if max_size = opts[:max_byte_size], do: check_byte_size(size, max_size)
-
-    case binary do
-      <<string::binary-size(size), rest::binary>> ->
-        {:ok, {string, rest}}
-
-      _ ->
-        {:error, :unexpected_eof}
-    end
-  end
-
-  defp decode_binary(binary, size, opts) do
-    if max_size = opts[:max_byte_size], do: check_byte_size(size, max_size)
-
-    case binary do
-      <<bin::binary-size(size), rest::binary>> ->
-        {:ok, {bin, rest}}
-
-      _ ->
-        {:error, :unexpected_eof}
-    end
-  end
-
-  defp decode_array(binary, size, opts) do
-    depth = opts[:depth] || 0
-
-    check_depth(depth, opts[:max_depth])
-    check_byte_size(size, opts[:max_byte_size])
-
-    new_opts = Keyword.put(opts, :depth, depth + 1)
-
-    decode_many(binary, size, [], new_opts)
-  end
-
-  defp decode_map(binary, size, opts) do
-    depth = opts[:depth] || 0
-
-    check_depth(depth, opts[:max_depth])
-    check_byte_size(size * 2, opts[:max_byte_size])
-
-    new_opts = Keyword.put(opts, :depth, depth + 1)
-
-    with {:ok, {kv_pairs, rest}} <- decode_many(binary, size * 2, [], new_opts) do
-      map =
-        Enum.chunk_every(kv_pairs, 2)
-        |> Enum.map(&List.to_tuple/1)
-        |> Enum.into(%{})
-
-      {:ok, {map, rest}}
-    end
-  end
-
-  # Recursively decodes `count` terms from the binary
-  defp decode_many(binary, 0, acc, _opts) do
-    {:ok, {Enum.reverse(acc), binary}}
-  end
-
-  defp decode_many(binary, count, acc, opts) do
-    case do_decode(binary, opts) do
-      {:ok, {term, rest}} ->
-        decode_many(rest, count - 1, [term | acc], opts)
-
-      {:error, reason} ->
-        {:error, reason}
-    end
-  end
-
-  defp decode_ext(-1, data, rest, _opts) do
-    {:ok, {decode_timestamp(data), rest}}
-  end
-
-  defp decode_ext(type, data, rest, _opts) do
-    {:ok, {%Msgpack.Ext{type: type, data: data}, rest}}
-  end
-
-  # timestamp 32: 4 bytes (32-bit unsigned integer seconds)
-  defp decode_timestamp(<<unix_seconds::unsigned-32>>) do
-    gregorian_seconds = unix_seconds + @epoch_offset
-    erlang_datetime = :calendar.gregorian_seconds_to_datetime(gregorian_seconds)
-    NaiveDateTime.from_erl!(erlang_datetime)
-  end
-
-  # timestamp 64: 8 bytes (30-bit nanoseconds + 34-bit seconds)
-  defp decode_timestamp(<<data::unsigned-64>>) do
-    nanoseconds = :erlang.bsr(data, 34)
-
-    if nanoseconds > 999_999_999 do
-      throw({:error, :invalid_timestamp})
-    else
-      unix_seconds = :erlang.band(data, 0x00000003_FFFFFFFF)
-      gregorian_seconds = unix_seconds + @epoch_offset
-      erlang_datetime = :calendar.gregorian_seconds_to_datetime(gregorian_seconds)
-      base_datetime = NaiveDateTime.from_erl!(erlang_datetime)
-
-      if nanoseconds > 0 do
-        microseconds = div(nanoseconds, 1000)
-        %{base_datetime | microsecond: {microseconds, 6}}
-      else
-        base_datetime
-      end
-    end
-  end
-
-  # timestamp 96: 12 bytes (32-bit nanoseconds + 64-bit seconds)
-  defp decode_timestamp(<<nanoseconds::unsigned-32, unix_seconds::signed-64>>) do
-    if nanoseconds > 999_999_999 do
-      throw({:error, :invalid_timestamp})
-    else
-      gregorian_seconds = unix_seconds + @epoch_offset
-      erlang_datetime = :calendar.gregorian_seconds_to_datetime(gregorian_seconds)
-      base_datetime = NaiveDateTime.from_erl!(erlang_datetime)
-
-      if nanoseconds > 0 do
-        microseconds = div(nanoseconds, 1000)
-        %{base_datetime | microsecond: {microseconds, 6}}
-      else
-        base_datetime
-      end
-    end
-  end
-
-  defp check_byte_size(size, max_size) when size > max_size do
-    throw({:error, {:max_byte_size_exceeded, max_size}})
-  end
-
-  defp check_byte_size(_size, _max_size), do: :ok
-
-  defp check_depth(depth, max_depth) when depth >= max_depth do
-    throw({:error, {:max_depth_reached, max_depth}})
+  @doc """
+  Returns a keyword list of the default options for the decoder.
+  """
+  def default_opts() do
+    [
+      max_depth: 100,
+      max_byte_size: 10_000_000 # 10MB
+    ]
   end
-
-  defp check_depth(_depth, _max_depth), do: :ok
 end