0eeb64b4ff
Bumps [Newtonsoft.Json](https://github.com/JamesNK/Newtonsoft.Json) from 12.0.3 to 13.0.1. - [Release notes](https://github.com/JamesNK/Newtonsoft.Json/releases) - [Commits](https://github.com/JamesNK/Newtonsoft.Json/compare/12.0.3...13.0.1) --- updated-dependencies: - dependency-name: Newtonsoft.Json dependency-type: direct:production ... Signed-off-by: dependabot[bot] <support@github.com> |
||
---|---|---|
.github | ||
azure-pipelines | ||
benchmark | ||
doc | ||
sandbox | ||
src | ||
tests | ||
tools | ||
.editorconfig | ||
.gitattributes | ||
.gitignore | ||
CONTRIBUTING.md | ||
CONTRIBUTORS | ||
Directory.Build.props | ||
Directory.Build.rsp | ||
LICENSE | ||
MessagePack.ruleset | ||
MessagePack.sln | ||
README.md | ||
SECURITY.md | ||
azure-pipelines.yml | ||
global.json | ||
graph.xlsx | ||
init.cmd | ||
init.ps1 | ||
installcredprovider.ps1 | ||
nuget.config | ||
opensource.snk | ||
prepare_release.ps1 | ||
stylecop.json | ||
version.json |
README.md
MessagePack for C# (.NET, .NET Core, Unity, Xamarin)
The extremely fast MessagePack serializer for C#. It is 10x faster than MsgPack-Cli and outperforms other C# serializers. MessagePack for C# also ships with built-in support for LZ4 compression - an extremely fast compression algorithm. Performance is important, particularly in applications like games, distributed computing, microservices, or data caches.
MessagePack has a compact binary size and a full set of general purpose expressive data types. Please have a look at the comparison with JSON, protobuf, ZeroFormatter section and learn why MessagePack C# is the fastest.
Table of Contents
- Installation
- Quick Start
- Analyzer
- Built-in supported types
- Object Serialization
- DataContract compatibility
- Serializing readonly/immutable object members (SerializationConstructor)
- Serialization Callback
- Union
- Dynamic (Untyped) Deserialization
- Object Type Serialization
- Typeless
- Security
- Performance
- LZ4 Compression
- Comparison with protobuf, JSON, ZeroFormatter
- Hints to achieve maximum performance when using MessagePack for C#
- Extensions
- Experimental Features
- High-Level API (
MessagePackSerializer
) - Low-Level API (
IMessagePackFormatter<T>
) - Primitive API (
MessagePackWriter
,MessagePackReader
) - Main Extension Point (
IFormatterResolver
) - MessagePackFormatterAttribute
- IgnoreFormatter
- Reserved Extension Types
- Unity support
- AOT Code Generation (support for Unity/Xamarin)
- RPC
- How to build
- Author Info
Installation
This library is distributed via NuGet. Special Unity support is available, too.
We target .NET Standard 2.0 with special optimizations for .NET Core 2.1+, making it compatible with most reasonably recent .NET runtimes such as Core 2.0 and later, Framework 4.6.1 and later, Mono 5.4 and later and Unity 2018.3 and later. The library code is pure C# (with Just-In-Time IL code generation on some platforms).
NuGet packages
To install with NuGet, just install the MessagePack
package:
Install-Package MessagePack
Install the optional C# analyzers package to get warnings about coding mistakes and automatic fix suggestions to save you time:
Install-Package MessagePackAnalyzer
There are also a range of official and third party Extension Packages available (learn more in our extensions section):
Install-Package MessagePack.ReactiveProperty
Install-Package MessagePack.UnityShims
Install-Package MessagePack.AspNetCoreMvcFormatter
Unity
For Unity projects, the [releases][Releases] page provides downloadable .unitypackage
files. When using in Unity IL2CPP or Xamarin AOT environments, please carefully read the pre-code generation section.
Migration notes from v1.x
If you were using MessagePack for C# v1.x, check out the "How to update to our new v2.x version" document.
Quick Start
Define the struct or class to be serialized and annotate it with a [MessagePackObject]
attribute.
Annotate members whose values should be serialized (fields as well as properties) with [Key]
attributes.
[MessagePackObject]
public class MyClass
{
// Key attributes take a serialization index (or string name)
// The values must be unique and versioning has to be considered as well.
// Keys are described in later sections in more detail.
[Key(0)]
public int Age { get; set; }
[Key(1)]
public string FirstName { get; set; }
[Key(2)]
public string LastName { get; set; }
// All fields or properties that should not be serialized must be annotated with [IgnoreMember].
[IgnoreMember]
public string FullName { get { return FirstName + LastName; } }
}
Call MessagePackSerializer.Serialize<T>/Deserialize<T>
to serialize/deserialize your object instance.
You can use the ConvertToJson
method to get a human readable representation of any MessagePack binary blob.
class Program
{
static void Main(string[] args)
{
var mc = new MyClass
{
Age = 99,
FirstName = "hoge",
LastName = "huga",
};
// Call Serialize/Deserialize, that's all.
byte[] bytes = MessagePackSerializer.Serialize(mc);
MyClass mc2 = MessagePackSerializer.Deserialize<MyClass>(bytes);
// You can dump MessagePack binary blobs to human readable json.
// Using indexed keys (as opposed to string keys) will serialize to MessagePack arrays,
// hence property names are not available.
// [99,"hoge","huga"]
var json = MessagePackSerializer.ConvertToJson(bytes);
Console.WriteLine(json);
}
}
By default, a MessagePackObject
annotation is required. This can be made optional; see the Object Serialization section and the Formatter Resolver section for details.
Analyzer
The MessagePackAnalyzer package aids with:
- Automating definitions for your serializable objects.
- Produces compiler warnings upon incorrect attribute use, member accessibility, and more.
If you want to allow a specific custom type (for example, when registering a custom type), put MessagePackAnalyzer.json
at the project root and change the Build Action to AdditionalFiles
.
An example MessagePackAnalyzer.json
:
[ "MyNamespace.FooClass", "MyNameSpace.BarStruct" ]
Built-in supported types
These types can serialize by default:
- Primitives (
int
,string
, etc...),Enum
s,Nullable<>
,Lazy<>
TimeSpan
,DateTime
,DateTimeOffset
Guid
,Uri
,Version
,StringBuilder
BigInteger
,Complex
,Half
Array[]
,Array[,]
,Array[,,]
,Array[,,,]
,ArraySegment<>
,BitArray
KeyValuePair<,>
,Tuple<,...>
,ValueTuple<,...>
ArrayList
,Hashtable
List<>
,LinkedList<>
,Queue<>
,Stack<>
,HashSet<>
,ReadOnlyCollection<>
,SortedList<,>
IList<>
,ICollection<>
,IEnumerable<>
,IReadOnlyCollection<>
,IReadOnlyList<>
Dictionary<,>
,IDictionary<,>
,SortedDictionary<,>
,ILookup<,>
,IGrouping<,>
,ReadOnlyDictionary<,>
,IReadOnlyDictionary<,>
ObservableCollection<>
,ReadOnlyObservableCollection<>
ISet<>
,ConcurrentBag<>
,ConcurrentQueue<>
,ConcurrentStack<>
,ConcurrentDictionary<,>
- Immutable collections (
ImmutableList<>
, etc) - Custom implementations of
ICollection<>
orIDictionary<,>
with a parameterless constructor - Custom implementations of
IList
orIDictionary
with a parameterless constructor
You can add support for custom types, and there are some official/third-party extension packages for:
- ReactiveProperty
- for Unity (
Vector3
,Quaternion
, etc...) - F# (Record, FsList, Discriminated Unions, etc...)
Please see the extensions section.
MessagePack.Nil
is the built-in type representing null/void in MessagePack for C#.
Object Serialization
MessagePack for C# can serialize your own public class
or struct
types. By default, serializable types must be annotated with the [MessagePackObject]
attribute and members with the [Key]
attribute. Keys can be either indexes (int
) or arbitrary strings. If all keys are indexes, arrays are used for serialization, which offers advantages in performance and binary size. Otherwise, MessagePack maps (dictionaries) will be used.
If you use [MessagePackObject(keyAsPropertyName: true)]
, then members do not require explicit Key
attributes, but string keys will be used.
[MessagePackObject]
public class Sample1
{
[Key(0)]
public int Foo { get; set; }
[Key(1)]
public int Bar { get; set; }
}
[MessagePackObject]
public class Sample2
{
[Key("foo")]
public int Foo { get; set; }
[Key("bar")]
public int Bar { get; set; }
}
[MessagePackObject(keyAsPropertyName: true)]
public class Sample3
{
// No need for a Key attribute
public int Foo { get; set; }
// If want to ignore a public member, you can use the IgnoreMember attribute
[IgnoreMember]
public int Bar { get; set; }
}
// [10,20]
Console.WriteLine(MessagePackSerializer.SerializeToJson(new Sample1 { Foo = 10, Bar = 20 }));
// {"foo":10,"bar":20}
Console.WriteLine(MessagePackSerializer.SerializeToJson(new Sample2 { Foo = 10, Bar = 20 }));
// {"Foo":10}
Console.WriteLine(MessagePackSerializer.SerializeToJson(new Sample3 { Foo = 10, Bar = 20 }));
All public instance members (fields as well as properties) will be serialized. If you want to ignore certain public members, annotate the member with a [IgnoreMember]
attribute.
Please note that any serializable struct or class must have public accessibility; private and internal structs and classes cannot be serialized!
The default of requiring MessagePackObject
annotations is meant to enforce explicitness and therefore may help write more robust code.
Should you use an indexed (int
) key or a string key?
We recommend using indexed keys for faster serialization and a more compact binary representation than string keys.
However, the additional information in the strings of string keys can be quite useful when debugging.
When classes change or are extended, be careful about versioning. MessagePackSerializer
will initialize members to their default
value if a key does not exist in the serialized binary blob, meaning members using reference types can be initialized to null
.
If you use indexed (int
) keys, the keys should start at 0 and should be sequential. If a later version stops using certain members, you should keep the obsolete members (C# provides an Obsolete
attribute to annotate such members) until all other clients had a chance to update and remove their uses of these members as well. Also, when the values of indexed keys "jump" a lot, leaving gaps in the sequence, it will negatively affect the binary size, as null
placeholders will be inserted into the resulting arrays. However, you shouldn't reuse indexes of removed members to avoid compatibility issues between clients or when trying to deserialize legacy blobs.
Example of index gaps and resulting placeholders:
[MessagePackObject]
public class IntKeySample
{
[Key(3)]
public int A { get; set; }
[Key(10)]
public int B { get; set; }
}
// [null,null,null,0,null,null,null,null,null,null,0]
Console.WriteLine(MessagePackSerializer.SerializeToJson(new IntKeySample()));
If you do not want to explicitly annotate with the MessagePackObject
/Key
attributes and instead want to use MessagePack for C# more like e.g. Json.NET, you can make use of the contractless resolver.
public class ContractlessSample
{
public int MyProperty1 { get; set; }
public int MyProperty2 { get; set; }
}
var data = new ContractlessSample { MyProperty1 = 99, MyProperty2 = 9999 };
var bin = MessagePackSerializer.Serialize(
data,
MessagePack.Resolvers.ContractlessStandardResolver.Options);
// {"MyProperty1":99,"MyProperty2":9999}
Console.WriteLine(MessagePackSerializer.ConvertToJson(bin));
// You can also set ContractlessStandardResolver as the default.
// (Global state; Not recommended when writing library code)
MessagePackSerializer.DefaultOptions = MessagePack.Resolvers.ContractlessStandardResolver.Options;
// Now serializable...
var bin2 = MessagePackSerializer.Serialize(data);
If you want to serialize private members as well, you can use one of the *AllowPrivate
resolvers.
[MessagePackObject]
public class PrivateSample
{
[Key(0)]
int x;
public void SetX(int v)
{
x = v;
}
public int GetX()
{
return x;
}
}
var data = new PrivateSample();
data.SetX(9999);
// You can choose either StandardResolverAllowPrivate
// or ContractlessStandardResolverAllowPrivate
var bin = MessagePackSerializer.Serialize(
data,
MessagePack.Resolvers.DynamicObjectResolverAllowPrivate.Options);
If you want to use MessagePack for C# more like a BinaryFormatter with a typeless serialization API, use the typeless resolver and helpers. Please consult the Typeless section.
Resolvers are the way to add specialized support for custom types to MessagePack for C#. Please refer to the Extension point section.
DataContract compatibility
You can use [DataContract]
annotations instead of [MessagePackObject]
ones. If type is annotated with DataContract
, you can use [DataMember]
annotations instead of [Key]
ones and [IgnoreDataMember]
instead of [IgnoreMember]
.
Then [DataMember(Order = int)]
will behave the same as [Key(int)]
, [DataMember(Name = string)]
the same as [Key(string)]
, and [DataMember]
the same as [Key(nameof(member name)]
.
Using DataContract
, e.g. in shared libraries, makes your classes/structs independent from MessagePack for C# serialization. However, it is not supported by the analyzers nor in code generation by the mpc
tool. Also, features like UnionAttribute
, MessagePackFormatter
, SerializationConstructor
, etc can not be used. Due to this, we recommend that you use the specific MessagePack for C# annotations when possible.
Serializing readonly/immutable object members (SerializationConstructor)
MessagePack for C# supports serialization of readonly/immutable objects/members. For example, this struct can be serialized and deserialized.
[MessagePackObject]
public struct Point
{
[Key(0)]
public readonly int X;
[Key(1)]
public readonly int Y;
public Point(int x, int y)
{
this.X = x;
this.Y = y;
}
}
var data = new Point(99, 9999);
var bin = MessagePackSerializer.Serialize(data);
// Okay to deserialize immutable object
var point = MessagePackSerializer.Deserialize<Point>(bin);
MessagePackSerializer
will choose the constructor with the best matched argument list, using argument indexes index for index keys, or parameter names for string keys. If it cannot determine an appropriate constructor, a MessagePackDynamicObjectResolverException: can't find matched constructor parameter
exception will be thrown.
You can specify which constructor to use manually with a [SerializationConstructor]
annotation.
[MessagePackObject]
public struct Point
{
[Key(0)]
public readonly int X;
[Key(1)]
public readonly int Y;
[SerializationConstructor]
public Point(int x)
{
this.X = x;
this.Y = -1;
}
// If not marked attribute, used this(most matched argument)
public Point(int x, int y)
{
this.X = x;
this.Y = y;
}
}
C# 9 record
types
C# 9.0 record with primary constructor is similar immutable object, also supports serialize/deserialize.
// use key as property name
[MessagePackObject(true)]public record Point(int X, int Y);
// use property: to set KeyAttribute
[MessagePackObject] public record Point([property:Key(0)] int X, [property: Key(1)] int Y);
// Or use explicit properties
[MessagePackObject]
public record Person
{
[Key(0)]
public string FirstName { get; init; }
[Key(1)]
public string LastName { get; init; }
}
C# 9 init
property setter limitations
When using init
property setters in generic classes, a CLR bug prevents our most efficient code generation from invoking the property setter.
As a result, you should avoid using init
on property setters in generic classes when using the public-only DynamicObjectResolver
/StandardResolver
.
When using the DynamicObjectResolverAllowPrivate
/StandardResolverAllowPrivate
resolver the bug does not apply and you may use init
without restriction.
Serialization Callback
Objects implementing the IMessagePackSerializationCallbackReceiver
interface will received OnBeforeSerialize
and OnAfterDeserialize
calls during serialization/deserialization.
[MessagePackObject]
public class SampleCallback : IMessagePackSerializationCallbackReceiver
{
[Key(0)]
public int Key { get; set; }
public void OnBeforeSerialize()
{
Console.WriteLine("OnBefore");
}
public void OnAfterDeserialize()
{
Console.WriteLine("OnAfter");
}
}
Union
MessagePack for C# supports serializing interface-typed and abstract class-typed objects. It behaves like XmlInclude
or ProtoInclude
. In MessagePack for C# these are called Union
. Only interfaces and abstracts classes are allowed to be annotated with Union
attributes. Unique union keys are required.
// Annotate inheritance types
[MessagePack.Union(0, typeof(FooClass))]
[MessagePack.Union(1, typeof(BarClass))]
public interface IUnionSample
{
}
[MessagePackObject]
public class FooClass : IUnionSample
{
[Key(0)]
public int XYZ { get; set; }
}
[MessagePackObject]
public class BarClass : IUnionSample
{
[Key(0)]
public string OPQ { get; set; }
}
// ---
IUnionSample data = new FooClass() { XYZ = 999 };
// Serialize interface-typed object.
var bin = MessagePackSerializer.Serialize(data);
// Deserialize again.
var reData = MessagePackSerializer.Deserialize<IUnionSample>(bin);
// Use with e.g. type-switching in C# 7.0
switch (reData)
{
case FooClass x:
Console.WriteLine(x.XYZ);
break;
case BarClass x:
Console.WriteLine(x.OPQ);
break;
default:
break;
}
Unions are internally serialized to two-element arrays.
IUnionSample data = new BarClass { OPQ = "FooBar" };
var bin = MessagePackSerializer.Serialize(data);
// Union is serialized to two-length array, [key, object]
// [1,["FooBar"]]
Console.WriteLine(MessagePackSerializer.ConvertToJson(bin));
Using Union
with abstract classes works the same way.
[Union(0, typeof(SubUnionType1))]
[Union(1, typeof(SubUnionType2))]
[MessagePackObject]
public abstract class ParentUnionType
{
[Key(0)]
public int MyProperty { get; set; }
}
[MessagePackObject]
public class SubUnionType1 : ParentUnionType
{
[Key(1)]
public int MyProperty1 { get; set; }
}
[MessagePackObject]
public class SubUnionType2 : ParentUnionType
{
[Key(1)]
public int MyProperty2 { get; set; }
}
Please be mindful that you cannot reuse the same keys in derived types that are already present in the parent type, as internally a single flat array or map will be used and thus cannot have duplicate indexes/keys.
Dynamic (Untyped) Deserialization
When calling MessagePackSerializer.Deserialize<object>
or MessagePackSerializer.Deserialize<dynamic>
, any values present in the blob will be converted to primitive values, i.e. bool
, char
, sbyte
, byte
, short
, int
, long
, ushort
, uint
, ulong
, float
, double
, DateTime
, string
, byte[]
, object[]
, IDictionary<object, object>
.
// Sample blob.
var model = new DynamicModel { Name = "foobar", Items = new[] { 1, 10, 100, 1000 } };
var blob = MessagePackSerializer.Serialize(model, ContractlessStandardResolver.Options);
// Dynamic ("untyped")
var dynamicModel = MessagePackSerializer.Deserialize<dynamic>(blob, ContractlessStandardResolver.Options);
// You can access the data using array/dictionary indexers, as shown above
Console.WriteLine(dynamicModel["Name"]); // foobar
Console.WriteLine(dynamicModel["Items"][2]); // 100
Exploring object trees using the dictionary indexer syntax is the fastest option for untyped deserialization, but it is tedious to read and write. Where performance is not as important as code readability, consider deserializing with ExpandoObject.
Object Type Serialization
StandardResolver
and ContractlessStandardResolver
can serialize object
/anonymous typed objects.
var objects = new object[] { 1, "aaa", new ObjectFieldType { Anything = 9999 } };
var bin = MessagePackSerializer.Serialize(objects);
// [1,"aaa",[9999]]
Console.WriteLine(MessagePackSerializer.ConvertToJson(bin));
// Support anonymous Type Serialize
var anonType = new { Foo = 100, Bar = "foobar" };
var bin2 = MessagePackSerializer.Serialize(anonType, MessagePack.Resolvers.ContractlessStandardResolver.Options);
// {"Foo":100,"Bar":"foobar"}
Console.WriteLine(MessagePackSerializer.ConvertToJson(bin2));
Unity supports is limited.
When deserializing, the behavior will be the same as Dynamic (Untyped) Deserialization.
Typeless
The typeless API is similar to BinaryFormatter
, as it will embed type information into the blobs, so no types need to be specified explicitly when calling the API.
object mc = new Sandbox.MyClass()
{
Age = 10,
FirstName = "hoge",
LastName = "huga"
};
// Serialize with the typeless API
var blob = MessagePackSerializer.Typeless.Serialize(mc);
// Blob has embedded type-assembly information.
// ["Sandbox.MyClass, Sandbox",10,"hoge","huga"]
Console.WriteLine(MessagePackSerializer.ConvertToJson(bin));
// You can deserialize to MyClass again with the typeless API
// Note that no type has to be specified explicitly in the Deserialize call
// as type information is embedded in the binary blob
var objModel = MessagePackSerializer.Typeless.Deserialize(bin) as MyClass;
Type information is represented by the MessagePack ext
format, type code 100
.
MessagePackSerializer.Typeless
is a shortcut of Serialize/Deserialize<object>(TypelessContractlessStandardResolver.Instance)
.
If you want to configure it as the default resolver, you can use MessagePackSerializer.Typeless.RegisterDefaultResolver
.
TypelessFormatter
can used standalone or combined with other resolvers.
// Replaced `object` uses the typeless resolver
var resolver = MessagePack.Resolvers.CompositeResolver.Create(
new[] { MessagePack.Formatters.TypelessFormatter.Instance },
new[] { MessagePack.Resolvers.StandardResolver.Instance });
public class Foo
{
// use Typeless(this field only)
[MessagePackFormatter(typeof(TypelessFormatter))]
public object Bar;
}
If a type's name is changed later, you can no longer deserialize old blobs. But you can specify a fallback name in such cases, providing a TypelessFormatter.BindToType
function of your own.
MessagePack.Formatters.TypelessFormatter.BindToType = typeName =>
{
if (typeName.StartsWith("SomeNamespace"))
{
typeName = typeName.Replace("SomeNamespace", "AnotherNamespace");
}
return Type.GetType(typeName, false);
};
Security
Deserializing data from an untrusted source can introduce security vulnerabilities in your application.
Depending on the settings used during deserialization, untrusted data may be able to execute arbitrary code or cause a denial of service attack.
Untrusted data might come from over the network from an untrusted source (e.g. any and every networked client) or can be tampered with by an intermediary when transmitted over an unauthenticated connection, or from a local storage that might have been tampered with, or many other sources. MessagePack for C# does not provide any means to authenticate data or make it tamper-resistant. Please use an appropriate method of authenticating data before deserialization - such as a MAC
.
Please be very mindful of these attack scenarios; many projects and companies, and serialization library users in general, have been bitten by untrusted user data deserialization in the past.
When deserializing untrusted data, put MessagePack into a more secure mode by configuring your MessagePackSerializerOptions.Security
property:
var options = MessagePackSerializerOptions.Standard
.WithSecurity(MessagePackSecurity.UntrustedData);
// Pass the options explicitly for the greatest control.
T object = MessagePackSerializer.Deserialize<T>(data, options);
// Or set the security level as the default.
MessagePackSerializer.DefaultOptions = options;
You should also avoid the Typeless serializer/formatters/resolvers for untrusted data as that opens the door for the untrusted data to potentially deserialize unanticipated types that can compromise security.
The UntrustedData
mode merely hardens against some common attacks, but is no fully secure solution in itself.
Performance
Benchmarks comparing MessagePack For C# to other serializers were run on Windows 10 Pro x64 Intel Core i7-6700K 4.00GHz, 32GB RAM
. Benchmark code is available here - and their version info.
ZeroFormatter and FlatBuffers have infinitely fast deserializers, so ignore their deserialization performance.
MessagePack for C# uses many techniques to improve performance.
- The serializer uses
IBufferWriter<byte>
rather thanSystem.IO.Stream
to reduce memory overhead. - Buffers are rented from pools to reduce allocations, keeping throughput high through reduced GC pressure.
- Don't create intermediate utility instances (
*Writer/*Reader
,*Context
, etc...) - Utilize dynamic code generation and JIT to avoid boxing value types. Use AOT generation on platforms that prohibit JITs.
- Cached generated formatters on static generic fields (don't use dictionary-cache because dictionary lookup is overhead). See Resolvers
- Heavily tuned dynamic IL code generation and JIT to avoid boxing value types. See DynamicObjectTypeBuilder. Use AOT generation on platforms that prohibit JIT.
- Call the Primitive API directly when IL code generation determines target types to be primitive.
- Reduce branching of variable length formats when IL code generation knows the target type (integer/string) ranges
- Don't use the
IEnumerable<T>
abstraction to iterate over collections when possible, see: CollectionFormatterBase and derived collection formatters - Use pre-generated lookup tables to reduce checks of mgpack type constraints, see: MessagePackBinary
- Uses optimized type key dictionary for non-generic methods, see: ThreadsafeTypeKeyHashTable
- Avoid string key decoding for lookup maps (string key and use automata based name lookup with inlined IL code generation, see: AutomataDictionary
- To encode string keys, use pre-generated member name bytes and fixed sized byte array copies in IL, see: UnsafeMemory.cs
Before creating this library, I implemented a fast serializer with ZeroFormatter#Performance. This is a further evolved implementation. MessagePack for C# is always fast and optimized for all types (primitive, small struct, large object, any collections).
Deserialization Performance for different options
Performance varies depending on the options used. This is a micro benchmark with BenchmarkDotNet. The target object has 9 members (MyProperty1
~ MyProperty9
), values are zero.
Method | Mean | Error | Scaled | Gen 0 | Allocated |
---|---|---|---|---|---|
M IntKey | 72.67 ns | NA | 1.00 | 0.0132 | 56 B |
M StringKey | 217.95 ns | NA | 3.00 | 0.0131 | 56 B |
M Typeless_IntKey | 176.71 ns | NA | 2.43 | 0.0131 | 56 B |
M Typeless_StringKey | 378.64 ns | NA | 5.21 | 0.0129 | 56 B |
MsgPackCliMap | 1,355.26 ns | NA | 18.65 | 0.1431 | 608 B |
MsgPackCliArray | 455.28 ns | NA | 6.26 | 0.0415 | 176 B |
ProtobufNet | 265.85 ns | NA | 3.66 | 0.0319 | 136 B |
Hyperion | 366.47 ns | NA | 5.04 | 0.0949 | 400 B |
JsonNetString | 2,783.39 ns | NA | 38.30 | 0.6790 | 2864 B |
JsonNetStreamReader | 3,297.90 ns | NA | 45.38 | 1.4267 | 6000 B |
JilString | 553.65 ns | NA | 7.62 | 0.0362 | 152 B |
JilStreamReader | 1,408.46 ns | NA | 19.38 | 0.8450 | 3552 B |
ÌntKey
, StringKey
, Typeless_IntKey
, Typeless_StringKey
are MessagePack for C# options. All MessagePack for C# options achieve zero memory allocations in the deserialization process. JsonNetString
/JilString
is deserialized from strings. JsonNetStreamReader
/JilStreamReader
is deserialized from UTF-8 byte arrays using StreamReader
. Deserialization is normally read from Stream. Thus, it will be restored from byte arrays (or Stream) instead of strings.
MessagePack for C# IntKey
is the fastest. StringKey
is slower than IntKey
because matching the character string of property names is required. IntKey
works by reading the array length, then for (array length) { binary decode }
. StringKey
works by reading map length, for (map length) { decode key, lookup key, binary decode }
, so it requires an additional two steps (decoding of keys and lookups of keys).
String key is often a useful, contractless, simple replacement of JSON, interoperability with other languages, and more robust versioning. MessagePack for C# is also optimized for string keys as much a possible. First of all, it does not decode UTF-8 byte arrays to full string for matching with the member name; instead it will look up the byte arrays as it is (to avoid decoding costs and extra memory allocations).
And It will try to match each long type
(per 8 character, if it is not enough, pad with 0) using automata and inline it when generating IL code.
This also avoids calculating the hash code of byte arrays, and the comparison can be made several times faster using the long type.
This is the sample of decompiled generated deserializer code, decompiled using ILSpy.
If the number of nodes is large, searches will use an embedded binary search.
Extra note, this is serialization benchmark result.
Method | Mean | Error | Scaled | Gen 0 | Allocated |
---|---|---|---|---|---|
IntKey | 84.11 ns | NA | 1.00 | 0.0094 | 40 B |
StringKey | 126.75 ns | NA | 1.51 | 0.0341 | 144 B |
Typeless_IntKey | 183.31 ns | NA | 2.18 | 0.0265 | 112 B |
Typeless_StringKey | 193.95 ns | NA | 2.31 | 0.0513 | 216 B |
MsgPackCliMap | 967.68 ns | NA | 11.51 | 0.1297 | 552 B |
MsgPackCliArray | 284.20 ns | NA | 3.38 | 0.1006 | 424 B |
ProtobufNet | 176.43 ns | NA | 2.10 | 0.0665 | 280 B |
Hyperion | 280.14 ns | NA | 3.33 | 0.1674 | 704 B |
ZeroFormatter | 149.95 ns | NA | 1.78 | 0.1009 | 424 B |
JsonNetString | 1,432.55 ns | NA | 17.03 | 0.4616 | 1944 B |
JsonNetStreamWriter | 1,775.72 ns | NA | 21.11 | 1.5526 | 6522 B |
JilString | 547.51 ns | NA | 6.51 | 0.3481 | 1464 B |
JilStreamWriter | 778.78 ns | NA | 9.26 | 1.4448 | 6066 B |
Of course, IntKey
is fastest but StringKey
also performs reasonably well.
String interning
The msgpack format does not provide for reusing strings in the data stream.
This naturally leads the deserializer to create a new string
object for every string encountered,
even if it is equal to another string previously encountered.
When deserializing data that may contain the same strings repeatedly it can be worthwhile to have the deserializer take a little extra time to check whether it has seen a given string before and reuse it if it has.
To enable string interning on all string values, use a resolver that specifies StringInterningFormatter
before any of the standard ones, like this:
var options = MessagePackSerializerOptions.Standard.WithResolver(
CompositeResolver.Create(
new IMessagePackFormatter[] { new StringInterningFormatter() },
new IFormatterResolver[] { StandardResolver.Instance }));
MessagePackSerializer.Deserialize<ClassOfStrings>(data, options);
If you know which fields of a particular type are likely to contain duplicate strings,
you can apply the string interning formatter to just those fields so the deserializer only pays
for the interned string check where it matters most.
Note that this technique requires a [MessagePackObject]
or [DataContract]
class.
[MessagePackObject]
public class ClassOfStrings
{
[Key(0)]
[MessagePackFormatter(typeof(StringInterningFormatter))]
public string InternedString { get; set; }
[Key(1)]
public string OrdinaryString { get; set; }
}
If you are writing your own formatter for some type that contains strings,
you can call on the StringInterningFormatter
directly from your formatter as well for the strings.
LZ4 Compression
MessagePack is a fast and compact format but it is not compression. LZ4 is an extremely fast compression algorithm, and using it MessagePack for C# can achieve extremely fast performance as well as extremely compact binary sizes!
MessagePack for C# has built-in LZ4 support. You can activate it using a modified options object and passing it into an API like this:
var lz4Options = MessagePackSerializerOptions.Standard.WithCompression(MessagePackCompression.Lz4BlockArray);
MessagePackSerializer.Serialize(obj, lz4Options);
MessagePackCompression
has two modes, Lz4Block
and Lz4BlockArray
. Neither is a simple binary LZ4 compression, but a special compression integrated into the serialization pipeline, using MessagePack ext
code (Lz4BlockArray (98)
or Lz4Block (99)
). Therefore, it is not readily compatible with compression offered in other languages.
Lz4Block
compresses an entire MessagePack sequence as a single LZ4 block. This is the simple compression that achieves best compression ratio, at the cost of copying the entire sequence when necessary to get contiguous memory.
Lz4BlockArray
compresses an entire MessagePack sequence as a array of LZ4 blocks. Compressed/decompressed blocks are chunked and thus do not enter the GC's Large-Object-Heap, but the compression ratio is slightly worse.
We recommend to use Lz4BlockArray
as the default when using compression.
For compatibility with MessagePack v1.x, use Lz4Block
.
Regardless of which LZ4 option is set at the deserialization, both methods can be deserialized. For example, when the Lz4BlockArray
option was used, binary data using either Lz4Block
and Lz4BlockArray
can be deserialized. Neither can be decompressed and hence deserialized when the compression option is set to None
.
Attributions
LZ4 compression support is using Milosz Krajewski's lz4net code with some modifications.
Comparison with protobuf, JSON, ZeroFormatter
protobuf-net is major, widely used binary-format library on .NET. I love protobuf-net and respect their great work. But when you use protobuf-net as a general purpose serialization format, you may encounter an annoying issue.
[ProtoContract]
public class Parent
{
[ProtoMember(1)]
public int Primitive { get; set; }
[ProtoMember(2)]
public Child Prop { get; set; }
[ProtoMember(3)]
public int[] Array { get; set; }
}
[ProtoContract]
public class Child
{
[ProtoMember(1)]
public int Number { get; set; }
}
using (var ms = new MemoryStream())
{
// serialize null.
ProtoBuf.Serializer.Serialize<Parent>(ms, null);
ms.Position = 0;
var result = ProtoBuf.Serializer.Deserialize<Parent>(ms);
Console.WriteLine(result != null); // True, not null. but all property are zero formatted.
Console.WriteLine(result.Primitive); // 0
Console.WriteLine(result.Prop); // null
Console.WriteLine(result.Array); // null
}
using (var ms = new MemoryStream())
{
// serialize empty array.
ProtoBuf.Serializer.Serialize<Parent>(ms, new Parent { Array = System.Array.Empty<int>() });
ms.Position = 0;
var result = ProtoBuf.Serializer.Deserialize<Parent>(ms);
Console.WriteLine(result.Array == null); // True, null!
}
protobuf(-net) cannot handle null and empty collection correctly, because protobuf has no null
representation (see this SO answer from a protobuf-net author).
MessagePack's type system can correctly serialize the entire C# type system. This is a strong reason to recommend MessagePack over protobuf.
Protocol Buffers have good IDL and gRPC support. If you want to use IDL, I recommend Google.Protobuf over MessagePack.
JSON is good general-purpose format. It is simple, human-readable and thoroughly-enough specified. Utf8Json - which I created as well - adopts same architecture as MessagePack for C# and avoids encoding/decoding costs as much as possible just like this library does. If you want to know more about binary vs text formats, see Utf8Json/which serializer should be used.
ZeroFormatter is similar as FlatBuffers but specialized to C#, and special in that regard. Deserialization is infinitely fast but the produced binary size is larger. And ZeroFormatter's caching algorithm requires additional memory.
For many common uses, MessagePack for C# would be a better fit.
Hints to achieve maximum performance when using MessagePack for C#
MessagePack for C# prioritizes maximum performance by default. However, there are also some options that sacrifice performance for convenience.
Use indexed keys instead of string keys (Contractless)
The Deserialization Performance for different options section shows the results of indexed keys (IntKey
) vs string keys (StringKey
) performance. Indexed keys serialize the object graph as a MessagePack array. String keys serializes the object graph as a MessagePack map.
For example this type is serialized to
[MessagePackObject]
public class Person
{
[Key(0)] or [Key("name")]
public string Name { get; set;}
[Key(1)] or [Key("age")]
public int Age { get; set;}
}
new Person { Name = "foobar", Age = 999 }
IntKey
:["foobar", 999]
StringKey
:{"name:"foobar","age":999}
.
IntKey
is always fast in both serialization and deserialization because it does not have to handle and lookup key names, and always has the smaller binary size.
StringKey
is often a useful, contractless, simple replacement for JSON, interoperability with other languages with MessagePack support, and less error prone versioning. But to achieve maximum performance, use IntKey
.
Create own custom composite resolver
CompositeResolver.Create
is an easy way to create composite resolvers. But formatter lookups have some overhead. If you create a custom resolver (or use StaticCompositeResolver.Instance
), you can avoid this overhead.
public class MyApplicationResolver : IFormatterResolver
{
public static readonly IFormatterResolver Instance = new MyApplicationResolver();
// configure your custom resolvers.
private static readonly IFormatterResolver[] Resolvers = new IFormatterResolver[]
{
};
private MyApplicationResolver() { }
public IMessagePackFormatter<T> GetFormatter<T>()
{
return Cache<T>.Formatter;
}
private static class Cache<T>
{
public static IMessagePackFormatter<T> Formatter;
static Cache()
{
// configure your custom formatters.
if (typeof(T) == typeof(XXX))
{
Formatter = new ICustomFormatter();
return;
}
foreach (var resolver in Resolvers)
{
var f = resolver.GetFormatter<T>();
if (f != null)
{
Formatter = f;
return;
}
}
}
}
}
NOTE: If you are creating a library, recommend using the above custom resolver instead of
CompositeResolver.Create
. Also, libraries must not useStaticCompositeResolver
- as it is global state - to avoid compatibility issues.
Use native resolvers
By default, MessagePack for C# serializes GUID as string. This is much slower than the native .NET format GUID. The same applies to Decimal. If your application makes heavy use of GUID or Decimal and you don't have to worry about interoperability with other languages, you can replace them with the native serializers NativeGuidResolver
and NativeDecimalResolver
respectively.
Also, DateTime
is serialized using the MessagePack timestamp format. By using the NativeDateTimeResolver
, it is possible to maintain Kind and perform faster serialization.
Be careful when copying buffers
MessagePackSerializer.Serialize
returns byte[]
in default. The final byte[]
is copied from an internal buffer pool. That is an extra cost. You can use IBufferWriter<T>
or the Stream
API to write to buffers directly. If you want to use a buffer pool outside of the serializer, you should implement custom IBufferWriter<byte>
or use an existing one such as Sequence<T>
from the Nerdbank.Streams package.
During deserialization, MessagePackSerializer.Deserialize(ReadOnlyMemory<byte> buffer)
is better than the Deserialize(Stream stream)
overload. This is because the Stream API version starts by reading the data, generating a ReadOnlySequence<byte>
, and only then starts the deserialization.
Choosing compression
Compression is generally effective when there is duplicate data. In MessagePack, arrays containing objects using string keys (Contractless) can be compressed efficiently because compression can be applied to many duplicate property names. Indexed keys compression is not as effectively compressed as string keys, but indexed keys are smaller in the first place.
This is some example benchmark performance data;
Serializer | Mean | DataSize |
---|---|---|
IntKey | 2.941 us | 469.00 B |
IntKey(Lz4) | 3.449 us | 451.00 B |
StringKey | 4.340 us | 1023.00 B |
StringKey(Lz4) | 5.469 us | 868.00 B |
IntKey(Lz4)
is not as effectively compressed, but performance is still somewhat degraded. On the other hand, StringKey
can be expected to have a sufficient effect on the binary size. However, this is just an example. Compression can be quite effective depending on the data, too, or have little effect other than slowing down your program. There are also cases in which well-compressible data exists in the values (such as long strings, e.g. containing HTML data with many repeated HTML tags). It is important to verify the actual effects of compression on a case by case basis.
Extensions
MessagePack for C# has extension points that enable you to provide optimal serialization support for custom types. There are official extension support packages.
Install-Package MessagePack.ReactiveProperty
Install-Package MessagePack.UnityShims
Install-Package MessagePack.AspNetCoreMvcFormatter
The MessagePack.ReactiveProperty
package adds support for types of the ReactiveProperty library. It adds ReactiveProperty<>
, IReactiveProperty<>
, IReadOnlyReactiveProperty<>
, ReactiveCollection<>
, Unit
serialization support. It is useful for save viewmodel state.
The MessagePack.UnityShims
package provides shims for Unity's standard structs (Vector2
, Vector3
, Vector4
, Quaternion
, Color
, Bounds
, Rect
, AnimationCurve
, Keyframe
, Matrix4x4
, Gradient
, Color32
, RectOffset
, LayerMask
, Vector2Int
, Vector3Int
, RangeInt
, RectInt
, BoundsInt
) and corresponding formatters. It can enable proper communication between servers and Unity clients.
After installation, extension packages must be enabled, by creating composite resolvers. Here is an example showing how to enable all extensions.
// Set extensions to default resolver.
var resolver = MessagePack.Resolvers.CompositeResolver.Create(
// enable extension packages first
ReactivePropertyResolver.Instance,
MessagePack.Unity.Extension.UnityBlitResolver.Instance,
MessagePack.Unity.UnityResolver.Instance,
// finally use standard (default) resolver
StandardResolver.Instance
);
var options = MessagePackSerializerOptions.Standard.WithResolver(resolver);
// Pass options every time or set as default
MessagePackSerializer.DefaultOptions = options;
For configuration details, see: Extension Point section.
The MessagePack.AspNetCoreMvcFormatter
is add-on for ASP.NET Core MVC's serialization to boost up performance. This is configuration sample.
public void ConfigureServices(IServiceCollection services)
{
services.AddMvc().AddMvcOptions(option =>
{
option.OutputFormatters.Clear();
option.OutputFormatters.Add(new MessagePackOutputFormatter(ContractlessStandardResolver.Options));
option.InputFormatters.Clear();
option.InputFormatters.Add(new MessagePackInputFormatter(ContractlessStandardResolver.Options));
});
}
Other authors are creating extension packages, too.
- MagicOnion - gRPC based HTTP/2 RPC Streaming Framework
- MasterMemory - Embedded Readonly In-Memory Document Database
You can make your own extension serializers or integrate with frameworks. Let's create and share!
- MessagePack.FSharpExtensions - supports F# list, set, map, unit, option, discriminated union
- MessagePack.NodaTime - Support for NodaTime types to MessagePack C#
- WebApiContrib.Core.Formatter.MessagePack - supports ASP.NET Core MVC (details in blog post)
- MessagePack.MediaTypeFormatter - MessagePack MediaTypeFormatter
Experimental Features
MessagePack for C# has experimental features which provides you with very performant formatters. There is an official package.
Install-Package MessagePack.Experimental
For detailed information, see: Experimental.md
API
High-Level API (MessagePackSerializer
)
The MessagePackSerializer
class is the entry point of MessagePack for C#. Static methods make up the main API of MessagePack for C#.
API | Description |
---|---|
Serialize<T> |
Serializes an object graph to a MessagePack binary blob. Async variant for Stream available. Non-generic overloads available. |
Deserialize<T> |
Deserializes a MessagePack binary to an object graph. Async variant for Stream available. Non-generic overloads available. |
SerializeToJson |
Serialize a MessagePack-compatible object graph to JSON instead of MessagePack. Useful for debugging. |
ConvertToJson |
Convert MessagePack binary to JSON. Useful for debugging. |
ConvertFromJson |
Convert JSON to a MessagePack binary. |
The MessagePackSerializer.Typeless
class offers most of the same APIs as above, but removes all type arguments from the API, forcing serialization to include the full type name of the root object. It uses the TypelessContractlessStandardResolver
. Consider the result to be a .NET-specific MessagePack binary that isn't readily compatible with MessagePack deserializers in other runtimes.
MessagePack for C# fundamentally serializes using IBufferWriter<byte>
and deserializes using ReadOnlySequence<byte>
or Memory<byte>
. Method overloads are provided to conveniently use it with common buffer types and the .NET Stream
class, but some of these convenience overloads require copying buffers once and therefore have a certain overhead.
The high-level API uses a memory pool internally to avoid unnecessary memory allocation. If result size is under 64K, it allocates GC memory only for the return bytes.
Each serialize/deserialize method takes an optional MessagePackSerializerOptions
parameter which can be used to specify a custom IFormatterResolver
to use or to activate LZ4 compression support.
Multiple MessagePack structures on a single Stream
To deserialize a Stream
that contains multiple consecutive MessagePack data structures,
you can use the MessagePackStreamReader
class to efficiently identify the ReadOnlySequence<byte>
for each data structure and deserialize it. For example:
static async Task<List<T>> DeserializeListFromStreamAsync<T>(Stream stream, CancellationToken cancellationToken)
{
var dataStructures = new List<T>();
using (var streamReader = new MessagePackStreamReader(stream))
{
while (await streamReader.ReadAsync(cancellationToken) is ReadOnlySequence<byte> msgpack)
{
dataStructures.Add(MessagePackSerializer.Deserialize<T>(msgpack, cancellationToken: cancellationToken));
}
}
return dataStructures;
}
Low-Level API (IMessagePackFormatter<T>
)
The IMessagePackFormatter<T>
interface is responsible for serializing a unique type. For example Int32Formatter : IMessagePackFormatter<Int32>
represents Int32 MessagePack serializer.
public interface IMessagePackFormatter<T>
{
void Serialize(ref MessagePackWriter writer, T value, MessagePackSerializerOptions options);
T Deserialize(ref MessagePackReader reader, MessagePackSerializerOptions options);
}
Many built-in formatters exists under MessagePack.Formatters
. Your custom types are usually automatically supported with the built-in type resolvers that generate new IMessagePackFormatter<T>
types on-the-fly using dynamic code generation. See our AOT code generation support for platforms that do not support this.
However, some types - especially those provided by third party libraries or the runtime itself - cannot be appropriately annotated, and contractless serialization would produce inefficient or even wrong results.
To take more control over the serialization of such custom types, write your own IMessagePackFormatter<T>
implementation.
Here is an example of such a custom formatter implementation. Note its use of the primitive API that is described in the next section.
/// <summary>Serializes a <see cref="FileInfo" /> by its full path as a string.</summary>
public class FileInfoFormatter : IMessagePackFormatter<FileInfo>
{
public void Serialize(
ref MessagePackWriter writer, FileInfo value, MessagePackSerializerOptions options)
{
if (value == null)
{
writer.WriteNil();
return;
}
writer.WriteString(value.FullName);
}
public FileInfo Deserialize(
ref MessagePackReader reader, MessagePackSerializerOptions options)
{
if (reader.TryReadNil())
{
return null;
}
options.Security.DepthStep(ref reader);
var path = reader.ReadString();
reader.Depth--;
return new FileInfo(path);
}
}
The DepthStep
and Depth--
statements provide a level of security while deserializing untrusted data
that might otherwise be able to execute a denial of service attack by sending MessagePack data that would
deserialize into a very deep object graph leading to a StackOverflowException
that would crash the process.
This pair of statements should surround the bulk of any IMessagePackFormatter<T>.Deserialize
method.
Important: A message pack formatter must read or write exactly one data structure. In the above example we just read/write a string. If you have more than one element to write out, you must precede it with a map or array header. You must read the entire map/array when deserializing. For example:
public class MySpecialObjectFormatter : IMessagePackFormatter<MySpecialObject>
{
public void Serialize(
ref MessagePackWriter writer, MySpecialObject value, MessagePackSerializerOptions options)
{
if (value == null)
{
writer.WriteNil();
return;
}
writer.WriteArrayHeader(2);
writer.WriteString(value.FullName);
writer.WriteString(value.Age);
}
public MySpecialObject Deserialize(
ref MessagePackReader reader, MessagePackSerializerOptions options)
{
if (reader.TryReadNil())
{
return null;
}
options.Security.DepthStep(ref reader);
string fullName = null;
int age = 0;
// Loop over *all* array elements independently of how many we expect,
// since if we're serializing an older/newer version of this object it might
// vary in number of elements that were serialized, but the contract of the formatter
// is that exactly one data structure must be read, regardless.
// Alternatively, we could check that the size of the array/map is what we expect
// and throw if it is not.
int count = reader.ReadArrayHeader();
for (int i = 0; i < count; i++)
{
switch (i)
{
case 0:
fullName = reader.ReadString();
break;
case 1:
age = reader.ReadInt32();
break;
default:
reader.Skip();
break;
}
}
reader.Depth--;
return new MySpecialObject(fullName, age);
}
}
Your custom formatters must be discoverable via some IFormatterResolver
. Learn more in our resolvers section.
You can see many other samples from builtin formatters.
Primitive API (MessagePackWriter
, MessagePackReader
)
The MessagePackWriter
and MessagePackReader
structs make up the lowest-level API. They read and write the primitives types defined in the MessagePack specification.
MessagePackReader
A MessagePackReader
can efficiently read from ReadOnlyMemory<byte>
or ReadOnlySequence<byte>
without any allocations, except to allocate a new string
as required by the ReadString()
method. All other methods return either value structs or ReadOnlySequence<byte>
slices for extensions/arrays.
Reading directly from ReadOnlySequence<byte>
means the reader can directly consume some modern high performance APIs such as PipeReader
.
Method | Description |
---|---|
Skip |
Advances the reader's position past the current value. If the value is complex (e.g. map, array) the entire structure is skipped. |
Read* |
Read and return a value whose type is named by the method name from the current reader position. Throws if the expected type does not match the actual type. When reading numbers, the type need not match the binary-specified type exactly. The numeric value will be coerced into the desired type or throw if the integer type is too small for a large value. |
TryReadNil |
Advances beyond the current value if the current value is nil and returns true ; otherwise leaves the reader's position unchanged and returns false . |
ReadBytes |
Returns a slice of the input sequence representing the contents of a byte[] , and advances the reader. |
ReadStringSequence |
Returns a slice of the input sequence representing the contents of a string without decoding it, and advances the reader. |
Clone |
Creates a new MessagePackReader with the specified input sequence and the same settings as the original reader. |
CreatePeekReader |
Creates a new reader with the same position as this one, allowing the caller to "read ahead" without impacting the original reader's position. |
NextCode |
Reads the low-level MessagePack byte that describes the type of the next value. Does not advance the reader. See MessagePack format of first byte. Its static class has ToMessagePackType and ToFormatName utility methods. MessagePackRange means Min-Max fix range of MessagePack format. |
NextMessagePackType |
Describes the NextCode value as a higher level category. Does not advance the reader. See MessagePack spec of source types. |
(others) | Other methods and properties as described by the .xml doc comment file and Intellisense. |
The MessagePackReader
is capable of automatically interpreting both the old and new MessagePack spec.
MessagePackWriter
A MessagePackWriter
writes to a given instance of IBufferWriter<byte>
. Several common implementations of this exist, allowing zero allocations and minimal buffer copies while writing directly to several I/O APIs including PipeWriter
.
The MessagePackWriter
writes the new MessagePack spec by default, but can write MessagePack compatible with the old spec by setting the OldSpec
property to true
.
Method | Description |
---|---|
Clone |
Creates a new MessagePackWriter with the specified underlying IBufferWriter<byte> and the same settings as the original writer. |
Flush |
Writes any buffered bytes to the underlying IBufferWriter<byte> . |
WriteNil |
Writes the MessagePack equivalent of .NET's null value. |
Write |
Writes any MessagePack primitive value in the most compact form possible. Has overloads for every primitive type defined by the MessagePack spec. |
Write*IntType* |
Writes an integer value in exactly the MessagePack type specified, even if a more compact format exists. |
WriteMapHeader |
Introduces a map by specifying the number of key=value pairs it contains. |
WriteArrayHeader |
Introduces an array by specifying the number of elements it contains. |
WriteExtensionFormat |
Writes the full content of an extension value including length, type code and content. |
WriteExtensionFormatHeader |
Writes just the header (length and type code) of an extension value. |
WriteRaw |
Copies the specified bytes directly to the underlying IBufferWriter<byte> without any validation. |
(others) | Other methods and properties as described by the .xml doc comment file and Intellisense. |
DateTime
is serialized to MessagePack Timestamp format, it serialize/deserialize UTC and loses Kind
info and requires that MessagePackWriter.OldSpec == false
.
If you use the NativeDateTimeResolver
, DateTime
values will be serialized using .NET's native Int64
representation, which preserves Kind
info but may not be interoperable with non-.NET platforms.
Main Extension Point (IFormatterResolver
)
An IFormatterResolver
is storage of typed serializers. The MessagePackSerializer
API accepts a MessagePackSerializerOptions
object which specifies the IFormatterResolver
to use, allowing customization of the serialization of complex types.
Resolver Name | Description |
---|---|
BuiltinResolver | Builtin primitive and standard classes resolver. It includes primitive(int, bool, string...) and there nullable, array and list. and some extra builtin types(Guid , Uri , BigInteger , etc...). |
StandardResolver | Composited resolver. It resolves in the following order builtin -> attribute -> dynamic enum -> dynamic generic -> dynamic union -> dynamic object -> dynamic object fallback . This is the default of MessagePackSerializer. |
ContractlessStandardResolver | Composited StandardResolver (except dynamic object fallback) -> DynamicContractlessObjectResolver -> DynamicObjectTypeFallbackResolver . It enables contractless serialization. |
StandardResolverAllowPrivate | Same as StandardResolver but allow serialize/deserialize private members. |
ContractlessStandardResolverAllowPrivate | Same as ContractlessStandardResolver but allow serialize/deserialize private members. |
PrimitiveObjectResolver | MessagePack primitive object resolver. It is used fallback in object type and supports bool , char , sbyte , byte , short , int , long , ushort , uint , ulong , float , double , DateTime , string , byte[] , ICollection , IDictionary . |
DynamicObjectTypeFallbackResolver | Serialize is used type in from object type, deserialize is used PrimitiveObjectResolver. |
AttributeFormatterResolver | Get formatter from [MessagePackFormatter] attribute. |
CompositeResolver | Composes several resolvers and/or formatters together in an ordered list, allowing reuse and overriding of behaviors of existing resolvers and formatters. |
NativeDateTimeResolver | Serialize by .NET native DateTime binary format. It keeps DateTime.Kind that loses by standard(MessagePack timestamp) format. |
NativeGuidResolver | Serialize by .NET native Guid binary representation. It is faster than standard(string) representation. |
NativeDecimalResolver | Serialize by .NET native decimal binary representation. It is faster than standard(string) representation. |
DynamicEnumResolver | Resolver of enum and there nullable, serialize there underlying type. It uses dynamic code generation to avoid boxing and boostup performance serialize there name. |
DynamicEnumAsStringResolver | Resolver of enum and there nullable. It uses reflection call for resolve nullable at first time. |
DynamicGenericResolver | Resolver of generic type(Tuple<> , List<> , Dictionary<,> , Array , etc). It uses reflection call for resolve generic argument at first time. |
DynamicUnionResolver | Resolver of interface marked by UnionAttribute. It uses dynamic code generation to create dynamic formatter. |
DynamicObjectResolver | Resolver of class and struct made by MessagePackObjectAttribute. It uses dynamic code generation to create dynamic formatter. |
DynamicContractlessObjectResolver | Resolver of all classes and structs. It does not needs MessagePackObjectAttribute and serialized key as string(same as marked [MessagePackObject(true)] ). |
DynamicObjectResolverAllowPrivate | Same as DynamicObjectResolver but allow serialize/deserialize private members. |
DynamicContractlessObjectResolverAllowPrivate | Same as DynamicContractlessObjectResolver but allow serialize/deserialize private members. |
TypelessObjectResolver | Used for object , embed .NET type in binary by ext(100) format so no need to pass type in deserialization. |
TypelessContractlessStandardResolver | Composited resolver. It resolves in the following order nativedatetime -> builtin -> attribute -> dynamic enum -> dynamic generic -> dynamic union -> dynamic object -> dynamiccontractless -> typeless . This is the default of MessagePackSerializer.Typeless |
Each instance of MessagePackSerializer
accepts only a single resolver. Most object graphs will need more than one for serialization, so composing a single resolver made up of several is often required, and can be done with the CompositeResolver
as shown below:
// Do this once and store it for reuse.
var resolver = MessagePack.Resolvers.CompositeResolver.Create(
// resolver custom types first
ReactivePropertyResolver.Instance,
MessagePack.Unity.Extension.UnityBlitResolver.Instance,
MessagePack.Unity.UnityResolver.Instance,
// finally use standard resolver
StandardResolver.Instance
);
var options = MessagePackSerializerOptions.Standard.WithResolver(resolver);
// Each time you serialize/deserialize, specify the options:
byte[] msgpackBytes = MessagePackSerializer.Serialize(myObject, options);
T myObject2 = MessagePackSerializer.Deserialize<MyObject>(msgpackBytes, options);
A resolver can be set as default with MessagePackSerializer.DefaultOptions = options
, but WARNING:
When developing an application where you control all MessagePack-related code it may be safe to rely on this mutable static to control behavior.
For all other libraries or multi-purpose applications that use MessagePackSerializer
you should explicitly specify the MessagePackSerializerOptions
to use with each method invocation to guarantee your code behaves as you expect even when sharing an AppDomain
or process with other MessagePack users that may change this static property.
Here is sample of use DynamicEnumAsStringResolver
with DynamicContractlessObjectResolver
(It is Json.NET-like lightweight setting.)
// composite same as StandardResolver
var resolver = MessagePack.Resolvers.CompositeResolver.Create(
MessagePack.Resolvers.BuiltinResolver.Instance,
MessagePack.Resolvers.AttributeFormatterResolver.Instance,
// replace enum resolver
MessagePack.Resolvers.DynamicEnumAsStringResolver.Instance,
MessagePack.Resolvers.DynamicGenericResolver.Instance,
MessagePack.Resolvers.DynamicUnionResolver.Instance,
MessagePack.Resolvers.DynamicObjectResolver.Instance,
MessagePack.Resolvers.PrimitiveObjectResolver.Instance,
// final fallback(last priority)
MessagePack.Resolvers.DynamicContractlessObjectResolver.Instance
);
If you want to make an extension package, you should write both a formatter and resolver for easier consumption. Here is sample of a resolver:
public class SampleCustomResolver : IFormatterResolver
{
// Resolver should be singleton.
public static readonly IFormatterResolver Instance = new SampleCustomResolver();
private SampleCustomResolver()
{
}
// GetFormatter<T>'s get cost should be minimized so use type cache.
public IMessagePackFormatter<T> GetFormatter<T>()
{
return FormatterCache<T>.Formatter;
}
private static class FormatterCache<T>
{
public static readonly IMessagePackFormatter<T> Formatter;
// generic's static constructor should be minimized for reduce type generation size!
// use outer helper method.
static FormatterCache()
{
Formatter = (IMessagePackFormatter<T>)SampleCustomResolverGetFormatterHelper.GetFormatter(typeof(T));
}
}
}
internal static class SampleCustomResolverGetFormatterHelper
{
// If type is concrete type, use type-formatter map
static readonly Dictionary<Type, object> formatterMap = new Dictionary<Type, object>()
{
{typeof(FileInfo), new FileInfoFormatter()}
// add more your own custom serializers.
};
internal static object GetFormatter(Type t)
{
object formatter;
if (formatterMap.TryGetValue(t, out formatter))
{
return formatter;
}
// If type can not get, must return null for fallback mechanism.
return null;
}
}
MessagePackFormatterAttribute
MessagePackFormatterAttribute is a lightweight extension point of class, struct, interface, enum and property/field. This is like Json.NET's JsonConverterAttribute. For example, serialize private field, serialize x10 formatter.
[MessagePackFormatter(typeof(CustomObjectFormatter))]
public class CustomObject
{
string internalId;
public CustomObject()
{
this.internalId = Guid.NewGuid().ToString();
}
// serialize/deserialize internal field.
class CustomObjectFormatter : IMessagePackFormatter<CustomObject>
{
public void Serialize(ref MessagePackWriter writer, CustomObject value, MessagePackSerializerOptions options)
{
options.Resolver.GetFormatterWithVerify<string>().Serialize(ref writer, value.internalId, options);
}
public CustomObject Deserialize(ref MessagePackReader reader, MessagePackSerializerOptions options)
{
var id = options.Resolver.GetFormatterWithVerify<string>().Deserialize(ref reader, options);
return new CustomObject { internalId = id };
}
}
}
// per field, member
public class Int_x10Formatter : IMessagePackFormatter<int>
{
public int Deserialize(ref MessagePackReader reader, MessagePackSerializerOptions options)
{
return reader.ReadInt32() * 10;
}
public void Serialize(ref MessagePackWriter writer, int value, MessagePackSerializerOptions options)
{
writer.WriteInt32(value * 10);
}
}
[MessagePackObject]
public class MyClass
{
// You can attach custom formatter per member.
[Key(0)]
[MessagePackFormatter(typeof(Int_x10Formatter))]
public int MyProperty1 { get; set; }
}
Formatter is retrieved by AttributeFormatterResolver
, it is included in StandardResolver
.
IgnoreFormatter
IgnoreFormatter<T>
is lightweight extension point of class and struct. If there exists types that can't be serialized, you can register IgnoreFormatter<T>
that serializes those to nil/null.
// CompositeResolver can set custom formatter.
var resolver = MessagePack.Resolvers.CompositeResolver.Create(
new IMessagePackFormatter[]
{
// for example, register reflection infos (can not serialize)
new IgnoreFormatter<MethodBase>(),
new IgnoreFormatter<MethodInfo>(),
new IgnoreFormatter<PropertyInfo>(),
new IgnoreFormatter<FieldInfo>()
},
new IFormatterResolver[]
{
ContractlessStandardResolver.Instance
});
Reserved Extension Types
MessagePack for C# already used some MessagePack extension type codes, be careful to avoid using the same ext code for other purposes.
Range | Reserved for |
---|
-128, -1
30, 120) | Reserved for this library's use to support common types in .NET
This leaves the following ranges for your use:
- \[0, 30)
- \[120, 127]
Within the *reserved* ranges, this library defines or implements extensions that use these type codes:
| Code | Type | Use by |
| ---- | ---- | --- |
| -1 | DateTime | MessagePack-spec reserved for timestamp |
| 30 | Vector2[] | for Unity, UnsafeBlitFormatter |
| 31 | Vector3[] | for Unity, UnsafeBlitFormatter |
| 32 | Vector4[] | for Unity, UnsafeBlitFormatter |
| 33 | Quaternion[] | for Unity, UnsafeBlitFormatter |
| 34 | Color[] | for Unity, UnsafeBlitFormatter |
| 35 | Bounds[] | for Unity, UnsafeBlitFormatter |
| 36 | Rect[] | for Unity, UnsafeBlitFormatter |
| 37 | Int[] | for Unity, UnsafeBlitFormatter |
| 38 | Float[] | for Unity, UnsafeBlitFormatter |
| 39 | Double[] | for Unity, UnsafeBlitFormatter |
| 98 | All | MessagePackCompression.Lz4BlockArray |
| 99 | All | MessagePackCompression.Lz4Block |
| 100 | object | TypelessFormatter |
## Unity support
Unity lowest supported version is `2018.3`, API Compatibility Level supports both `.NET 4.x` and `.NET Standard 2.0`.
You can install the `unitypackage` from the [releases][Releases] page.
If your build targets .NET Framework 4.x and runs on mono, you can use it as is.
But if your build targets IL2CPP, you can not use `Dynamic***Resolver`, so it is required to use pre-code generation. Please see [pre-code generation section](#aot).
MessagePack for C# includes some additional `System.*.dll` libraries that originally provides in NuGet. They are located under `Plugins`. If other packages use these libraries (e.g. Unity Collections package using `System.Runtime.CompilerServices.Unsafe.dll`), to avoid conflicts, please delete the DLL under `Plugins`.
Currently `CompositeResolver.Create` does not work on IL2CPP, so it is recommended to use `StaticCompositeResolver.Instance.Register` instead.
In Unity, MessagePackSerializer can serialize `Vector2`, `Vector3`, `Vector4`, `Quaternion`, `Color`, `Bounds`, `Rect`, `AnimationCurve`, `Keyframe`, `Matrix4x4`, `Gradient`, `Color32`, `RectOffset`, `LayerMask`, `Vector2Int`, `Vector3Int`, `RangeInt`, `RectInt`, `BoundsInt` and their nullable, array and list types with the built-in extension `UnityResolver`. It is included in StandardResolver by default.
MessagePack for C# has an additional unsafe extension. `UnsafeBlitResolver` is special resolver for extremely fast but unsafe serialization/deserialization of struct arrays.
![image](https://cloud.githubusercontent.com/assets/46207/23837633/76589924-07ce-11e7-8b26-e50eab548938.png)
x20 faster Vector3[] serialization than native JsonUtility. If use `UnsafeBlitResolver`, serialization uses a special format (ext:typecode 30~39) for `Vector2[]`, `Vector3[]`, `Quaternion[]`, `Color[]`, `Bounds[]`, `Rect[]`. If use `UnityBlitWithPrimitiveArrayResolver`, it supports `int[]`, `float[]`, `double[]` too. This special feature is useful for serializing Mesh (many `Vector3[]`) or many transform positions.
If you want to use unsafe resolver, register `UnityBlitResolver` or `UnityBlitWithPrimitiveArrayResolver`.
Here is sample of configuration.
```csharp
StaticCompositeResolver.Instance.Register(
MessagePack.Unity.UnityResolver.Instance,
MessagePack.Unity.Extension.UnityBlitWithPrimitiveArrayResolver.Instance,
MessagePack.Resolvers.StandardResolver.Instance
);
var options = MessagePackSerializerOptions.Standard.WithResolver(StaticCompositeResolver.Instance);
MessagePackSerializer.DefaultOptions = options;
```
The `MessagePack.UnityShims` NuGet package is for .NET server-side serialization support to communicate with Unity. It includes shims for Vector3 etc and the Safe/Unsafe serialization extension.
If you want to share a class between Unity and a server, you can use `SharedProject` or `Reference as Link` or a glob reference (with `LinkBase`), etc. Anyway, you need to share at source-code level. This is a sample project structure using a glob reference (recommended).
- ServerProject(.NET 4.6/.NET Core/.NET Standard)
- \[`<Compile Include="..\UnityProject\Assets\Scripts\Shared\**\*.cs" LinkBase="Shared" />`\]
- \[MessagePack\]
- \[MessagePack.UnityShims\]
- UnityProject
- \[Concrete SharedCodes\]
- \[MessagePack\](not dll/NuGet, use MessagePack.Unity.unitypackage's sourcecode)
## <a name="aot"></a>AOT Code Generation (support for Unity/Xamarin)
By default, MessagePack for C# serializes custom objects by [generating IL](https://msdn.microsoft.com/en-us/library/system.reflection.emit.ilgenerator.aspx) on the fly at runtime to create custom, highly tuned formatters for each type. This code generation has a minor upfront performance cost.
Because strict-AOT environments such as Xamarin and Unity IL2CPP forbid runtime code generation, MessagePack provides a way for you to run a code generator ahead of time as well.
> Note: When using Unity, dynamic code generation only works when targeting .NET Framework 4.x + mono runtime.
For all other Unity targets, AOT is required.
If you want to avoid the upfront dynamic generation cost or you need to run on Xamarin or Unity, you need AOT code generation. `mpc` (MessagePackCompiler) is the code generator of MessagePack for C#. mpc uses [Roslyn](https://github.com/dotnet/roslyn) to analyze source code.
First of all, mpc requires [.NET Core 3 Runtime](https://dotnet.microsoft.com/download). The easiest way to acquire and run mpc is as a dotnet tool.
```
dotnet tool install --global MessagePack.Generator
```
Installing it as a local tool allows you to include the tools and versions that you use in your source control system. Run these commands in the root of your repo:
```
dotnet new tool-manifest
dotnet tool install MessagePack.Generator
```
Check in your `.config\dotnet-tools.json` file. On another machine you can "restore" your tool using the `dotnet tool restore` command.
Once you have the tool installed, simply invoke using `dotnet mpc` within your repo:
```
dotnet mpc --help
```
Alternatively, you can download mpc from the [releases][Releases] page, that includes platform native binaries (that don't require a separate dotnet runtime).
```
Usage: mpc [options...]
Options:
-i, -input <String> Input path to MSBuild project file or the directory containing Unity source files. (Required)
-o, -output <String> Output file path(.cs) or directory(multiple generate file). (Required)
-c, -conditionalSymbol <String> Conditional compiler symbols, split with ','. (Default: null)
-r, -resolverName <String> Set resolver name. (Default: GeneratedResolver)
-n, -namespace <String> Set namespace root name. (Default: MessagePack)
-m, -useMapMode <Boolean> Force use map mode serialization. (Default: False)
-ms, -multipleIfDirectiveOutputSymbols <String> Generate #if-- files by symbols, split with ','. (Default: null)
```
`mpc` targets C# code with `[MessagePackObject]` or `[Union]` annotations.
```cmd
// Simple Sample:
dotnet mpc -i "..\src\Sandbox.Shared.csproj" -o "MessagePackGenerated.cs"
// Use force map simulate DynamicContractlessObjectResolver
dotnet mpc -i "..\src\Sandbox.Shared.csproj" -o "MessagePackGenerated.cs" -m
```
By default, `mpc` generates the resolver as `MessagePack.Resolvers.GeneratedResolver` and formatters as`MessagePack.Formatters.*`.
Here is the full sample code to register a generated resolver in Unity.
```csharp
using MessagePack;
using MessagePack.Resolvers;
using UnityEngine;
public class Startup
{
static bool serializerRegistered = false;
[RuntimeInitializeOnLoadMethod(RuntimeInitializeLoadType.BeforeSceneLoad)]
static void Initialize()
{
if (!serializerRegistered)
{
StaticCompositeResolver.Instance.Register(
MessagePack.Resolvers.GeneratedResolver.Instance,
MessagePack.Resolvers.StandardResolver.Instance
);
var option = MessagePackSerializerOptions.Standard.WithResolver(StaticCompositeResolver.Instance);
MessagePackSerializer.DefaultOptions = option;
serializerRegistered = true;
}
}
#if UNITY_EDITOR
[UnityEditor.InitializeOnLoadMethod]
static void EditorInitialize()
{
Initialize();
}
#endif
}
```
In Unity, you can use MessagePack CodeGen windows at `Windows -> MessagePack -> CodeGenerator`.
![](https://user-images.githubusercontent.com/46207/69414381-f14da400-0d55-11ea-9f8d-9af448d347dc.png)
Install the .NET Core runtime, install mpc (as a .NET Core Tool as described above), and execute `dotnet mpc`. Currently this tool is experimental so please tell me your opinion.
In Xamarin, you can install the [the `MessagePack.MSBuild.Tasks` NuGet package](doc/msbuildtask.md) into your projects to pre-compile fast serialization code and run in environments where JIT compilation is not allowed.
## RPC
MessagePack advocated [MessagePack RPC](https://github.com/msgpack-rpc/msgpack-rpc), but work on it has stopped and it is not widely used.
### MagicOnion
I've created a gRPC based MessagePack HTTP/2 RPC streaming framework called [MagicOnion](https://github.com/Cysharp/MagicOnion). gRPC usually communicates with Protocol Buffers using IDL. But MagicOnion uses MessagePack for C# and does not need IDL. When communicating C# to C#, schemaless (or rather C# classes as schema) is better than using IDL.
### StreamJsonRpc
The StreamJsonRpc library is based on [JSON-RPC](https://www.jsonrpc.org/) and includes [a pluggable formatter architecture](https://github.com/microsoft/vs-streamjsonrpc/blob/master/doc/extensibility.md#alternative-formatters) and as of v2.3 includes [MessagePack support](https://github.com/microsoft/vs-streamjsonrpc/blob/master/doc/extensibility.md#message-formatterss).
## How to build
See our [contributor's guide](CONTRIBUTING.md).
## Author Info
Yoshifumi Kawai (a.k.a. neuecc) is a software developer in Japan.
He is the Director/CTO at Grani, Inc.
Grani is a mobile game developer company in Japan and well known for using C#.
He is awarding Microsoft MVP for Visual C# since 2011.
He is known as the creator of [UniRx](https://github.com/neuecc/UniRx/) (Reactive Extensions for Unity)
* Blog: [https://medium.com/@neuecc](https://medium.com/@neuecc) (English)
* Blog: [http://neue.cc/](http://neue.cc/) (Japanese)
* Twitter: [https://twitter.com/neuecc](https://twitter.com/neuecc) (Japanese)
[Releases]: https://github.com/neuecc/MessagePack-CSharp/releases