EmitMapper,AutoMapper,NLiteMapper和手工映射性能大比拼
在大比拼之前先讲一个小插曲，我这个人以前比较低调，做了很多好东西仅仅在公司内的朋友圈项目圈内分享，很少在博客园内进行分享，后来在dudu 老大的文章博客园现代化建设——AutoMapper有感便推荐一下OOMapper 组件,于是乎接连写了几篇入门性的介绍使用文章：
在园友Repository 兄的NLiteMapper与EmitMapper性能简单比较中了解到NLiteMapper与EmitMapper的性能巨大差距，于是乎进行了两天的性能优化，同时总结了优化过程：一次性能优化最佳实践。在这里非常感谢Repository 兄的测试，也非常感谢他把OOMapper纠正为NLiteMapper,否则NLiteMapper的性能是非常低下的，同时感谢dudu,感谢博客园给大家一个平台，在这个平台使我学到了很多很多......
不说废话进入主题。
准备工作：
- 软硬件环境：VS2008，.net3.5, xp 双核
- 测试组件（都是最新Release版本）：AutoMapper.dll(v2.0), EmitMappe.dll (V1.0),NLite.dll(V1.0)
性能测试工具：老赵的CodeTimer
测试接口代码：
[Contract]
public interface IObjectToObjectMapper
{
// 初始化映射器
void Initialize();
// 执行映射
void Map();
}
为了输出更友好的结果定义一下测试元数据代码：
// 测试映射器元数据
public interface IMapperMetadata
{
// 目录
string Category { get ; }
// 名称
string Name { get ; }
string Descrption { get ; }
}
// 映射器元数据注解
[AttributeUsage(AttributeTargets.Class, AllowMultiple = false )]
[MetadataAttributeAttribute]
public class MapperAttribute : ComponentAttribute
{
public string Category { get ; set ; }
public string Name { get ; set ; }
public string Descrption { get ; set ; }
}
利用NLite的Mini容器书写测试框架代码如下：测试次数10万次
class Program
{
[InjectMany]
private Lazy < IObjectToObjectMapper,IMapperMetadata > [] Mappers;
// 初始化映射器，并做一次映射操作
void Init()
{
foreach (var item in Mappers)
{
item.Value.Initialize();
item.Value.Map();
}
}
// 进行测试
void Run()
{
foreach (var item in Mappers)
CodeTimer.Time(item.Metadata.Category + " -> " + item.Metadata.Name , 100000 ,() => item.Value.Map());
}
static void Main( string [] args)
{
ServiceRegistry.RegisteryFromAssemblyOf < Program > ();
var host = new Program();
ServiceRegistry.Compose(host);
host.Init();
host.Run();
Console.Read();
}
}
这样完成了测试框架的搭建，现在就开始书写测试代码了。
定义测试数据：
public class ModelObject
{
public DateTime BaseDate { get ; set ; }
public ModelSubObject Sub { get ; set ; }
public ModelSubObject Sub2 { get ; set ; }
public ModelSubObject SubWithExtraName { get ; set ; }
}
public class ModelSubObject
{
public string ProperName { get ; set ; }
public ModelSubSubObject SubSub { get ; set ; }
}
public class ModelSubSubObject
{
public string IAmACoolProperty { get ; set ; }
}
public class ModelDto
{
public DateTime BaseDate { get ; set ; }
public string SubProperName { get ; set ; }
public string Sub2ProperName { get ; set ; }
public string SubWithExtraNameProperName { get ; set ; }
public string SubSubSubIAmACoolProperty { get ; set ; }
}
定义测试基类：
public abstract class MapperBase : IObjectToObjectMapper
{
protected ModelObject _source;
protected ModelDto _target;
protected virtual void OnInitialize() { }
public void Initialize()
{
OnInitialize();
_source = new ModelObject
{
BaseDate = new DateTime( 2007 , 4 , 5 ),
Sub = new ModelSubObject
{
ProperName = " Some name " ,
SubSub = new ModelSubSubObject
{
IAmACoolProperty = " Cool daddy-o "
}
},
Sub2 = new ModelSubObject
{
ProperName = " Sub 2 name "
},
SubWithExtraName = new ModelSubObject
{
ProperName = " Some other name "
},
};
}
public abstract void Map();
}
手工映射代码：
[Mapper(Category = " Flattening.Class " , Name = " Manual " )]
public class ManualMapper : MapperBase
{
public override void Map()
{
var destination = new ModelDto
{
BaseDate = _source.BaseDate,
Sub2ProperName = _source.Sub2.ProperName,
SubProperName = _source.Sub.ProperName,
SubSubSubIAmACoolProperty = _source.Sub.SubSub.IAmACoolProperty,
SubWithExtraNameProperName = _source.SubWithExtraName.ProperName
};
}
}
AutoMapper 映射代码：
[Mapper(Category = " Flattening.Class " , Name = " AutoMapper " )]
public class AutoMapperWrapper : MapperBase
{
protected override void OnInitialize()
{
AutoMapper.Mapper.Initialize(cfg =>
{
cfg.CreateMap < ModelObject, ModelDto > ();
});
AutoMapper.Mapper.AssertConfigurationIsValid();
}
public override void Map()
{
_target = AutoMapper.Mapper.Map < ModelObject, ModelDto > (_source);
}
}
EmitMapper映射代码：
[Mapper(Category = " Flattening.Class " , Name = " EmitMapper " )]
public class EmitMapperWrapper : MapperBase
{
ObjectsMapper < ModelObject, ModelDto > mapper;
protected override void OnInitialize()
{
mapper = ObjectMapperManager.DefaultInstance.GetMapper < ModelObject, ModelDto > ( new FlatteringConfig());
}
protected override ModelDto MapImp()
{
return mapper.Map(Source);
}
}
EmitMapper映射器默认不支持Flatter 映射，如果支持需要写自定义配置:
class FlatteringConfig : DefaultMapConfig
{
protected Func<string, string, bool> nestedMembersMatcher;
public FlatteringConfig()
{
nestedMembersMatcher = (m1, m2) => m1.StartsWith(m2);
}
public override IMappingOperation[] GetMappingOperations(Type from, Type to)
{
var destinationMembers = GetDestinationMemebers(to);
var sourceMembers = GetSourceMemebers(from);
var result = new List<IMappingOperation>();
foreach (var dest in destinationMembers)
{
var matchedChain = GetMatchedChain(dest.Name, sourceMembers).ToArray();
if (matchedChain == null || matchedChain.Length == 0)
{
continue;
}
result.Add(
new ReadWriteSimple
{
Source = new MemberDescriptor(matchedChain),
Destination = new MemberDescriptor(new[] { dest })
}
);
}
return result.ToArray();
}
public DefaultMapConfig MatchNestedMembers(Func<string, string, bool> nestedMembersMatcher)
{
this.nestedMembersMatcher = nestedMembersMatcher;
return this;
}
private List<MemberInfo> GetMatchedChain(string destName, List<MemberInfo> sourceMembers)
{
var matches = sourceMembers.Where(s => MatchMembers(destName, s.Name) || nestedMembersMatcher(destName, s.Name));
int len = 0;
MemberInfo match = null;
foreach (var m in matches)
{
if (m.Name.Length > len)
{
len = m.Name.Length;
match = m;
}
}
if (match == null)
{
return null;
}
var result = new List<MemberInfo> { match };
if (!MatchMembers(destName, match.Name))
{
result.AddRange(
GetMatchedChain(destName.Substring(match.Name.Length), GetDestinationMemebers(match))
);
}
return result;
}
private static List<MemberInfo> GetSourceMemebers(Type t)
{
return GetMemebers(t)
.Where(
m =>
m.MemberType == MemberTypes.Field ||
m.MemberType == MemberTypes.Property ||
m.MemberType == MemberTypes.Method
)
.ToList();
}
private static List<MemberInfo> GetDestinationMemebers(MemberInfo mi)
{
Type t;
if (mi.MemberType == MemberTypes.Field)
{
t = mi.DeclaringType.GetField(mi.Name).FieldType;
}
else
{
t = mi.DeclaringType.GetProperty(mi.Name).PropertyType;
}
return GetDestinationMemebers(t);
}
private static List<MemberInfo> GetDestinationMemebers(Type t)
{
return GetMemebers(t).Where(m => m.MemberType == MemberTypes.Field || m.MemberType == MemberTypes.Property).ToList();
}
private static List<MemberInfo> GetMemebers(Type t)
{
BindingFlags bindingFlags = BindingFlags.Instance | BindingFlags.Public;
return t.GetMembers(bindingFlags).ToList();
}
}
NLiteMapper映射代码：
[Mapper(Category = " Flattening.Class " , Name = " NLiteMapper " )]
public class NLiteMaperWrapper : MapperBase
{
private NLite.Mapping.IMapper < ModelObject, ModelDto > mapper;
protected override void OnInitialize()
{
base .OnInitialize();
mapper = NLite.Mapper.CreateMapper < ModelObject, ModelDto > ();
}
public override void Map()
{
_target = mapper.Map(_source);
}
}
Ok，完成代码用Release编译，然后再输出bin中找到exe文件，连续执行三次，下面是三次执行结果的截图：
------ Test started: Assembly: NLite.Test.dll ------ Flattening.Class->AutoMapper Time Elapsed: 1,112ms CPU Cycles: 6,718,750 Gen 0: 173 Gen 1: 1 Gen 2: 0 Flattening.Class->NLiteMapper Time Elapsed: 68ms CPU Cycles: 781,250 Gen 0: 4 Gen 1: 1 Gen 2: 0 Flattening.Class->EmitMapper Time Elapsed: 23ms CPU Cycles: 156,250 Gen 0: 3 Gen 1: 0 Gen 2: 0 Flattening.Class->Manual Time Elapsed: 8ms CPU Cycles: 0 Gen 0: 3 Gen 1: 1 Gen 2: 0
------ Test started: Assembly: NLite.Test.dll ------ Flattening.Class->AutoMapper Time Elapsed: 1,701ms CPU Cycles: 10,468,750 Gen 0: 173 Gen 1: 0 Gen 2: 0 Flattening.Class->NLiteMapper Time Elapsed: 69ms CPU Cycles: 781,250 Gen 0: 4 Gen 1: 1 Gen 2: 0 Flattening.Class->EmitMapper Time Elapsed: 22ms CPU Cycles: 0 Gen 0: 3 Gen 1: 0 Gen 2: 0 Flattening.Class->Manual Time Elapsed: 10ms CPU Cycles: 312,500 Gen 0: 3 Gen 1: 1 Gen 2: 0 1 passed, 0 failed, 0 skipped, took 2.98 seconds (NUnit 2.5.5).
------ Test started: Assembly: NLite.Test.dll ------ Flattening.Class->AutoMapper Time Elapsed: 1,205ms CPU Cycles: 10,156,250 Gen 0: 177 Gen 1: 0 Gen 2: 0 Flattening.Class->NLiteMapper Time Elapsed: 66ms CPU Cycles: 781,250 Gen 0: 4 Gen 1: 0 Gen 2: 0 Flattening.Class->EmitMapper Time Elapsed: 18ms CPU Cycles: 312,500 Gen 0: 3 Gen 1: 0 Gen 2: 0 Flattening.Class->Manual Time Elapsed: 9ms CPU Cycles: 0 Gen 0: 3 Gen 1: 0 Gen 2: 0 1 passed, 0 failed, 0 skipped, took 2.56 seconds (NUnit 2.5.5).
通过测试结果可以看出:
- 手工映射速度最快
- EmitMapper第二（大约比手工慢了2-6倍，）
- NLiteMapper第三（大约比EmitMapper慢了3倍）
- 最后是AutoMapper（大约比手工慢了200倍）
内存开销结果：
- 手工映射 Gen 0： 3
- EmitMapper Gen 0：3
- NLiteMapper Gen 0： 4
- AutoMapper Gen 0：173
总结 ：无论从性能和内存EmitMapper都接近于手工，NLiteMapper次之，AutoMapper最后。NLiteMapper，EmitMapper，AutoMapper都是通过Emit的方式进行Get和Set的，为什么性能差别如此之大，设想如果NLiteMapper不进行优化的话（NLiteMapper一直是通过Emit方式进行的），那么NLiteMapper肯定是高高垫背的（NLiteMapper比EmitMapper慢了15000倍）。。。。。。
这次测试结果不代表整体结果，仅仅代表Class->Class（包括级联） 的映射性能，欢迎大家对这几种OO映射器进行性能比较。最后附上整个测试代码：测试代码。
备注：EmitMapper的测试代码修改过，添加了FlatteringConfig class 这样测试就公平了。
附上的源代码是老代码，最新代码：http://nlite.codeplex.com/SourceControl/changeset/view/76359#1528885