Field inferenceedit

Several places in the Elasticsearch API expect the path to a field from your original source document, as a string value. NEST allows you to use C# expressions to strongly type these field path strings.

These expressions are assigned to a type called Field, and there are several ways to create an instance of one


Using the constructor directly is possible but can get rather involved when resolving from a member access lambda expression

var fieldString = new Field("name");

var fieldProperty = new Field(typeof(Project).GetProperty(nameof(Project.Name)));

Expression<Func<Project, object>> expression = p => p.Name;
var fieldExpression = new Field(expression);


When using the constructor and passing a value for Name, Property or Expression, ComparisonValue is also set on the Field instance; this is used when

  • determining Field equality
  • getting the hash code for a Field instance
var fieldStringWithBoostTwo = new Field("name^2");
var fieldStringWithBoostThree = new Field("name^3");

Expression<Func<Project, object>> expression = p => p.Name;
var fieldExpression = new Field(expression);

var fieldProperty = new Field(typeof(Project).GetProperty(nameof(Project.Name)));



Field Names with Boostedit

When specifying a Field name, the name can include a boost value; NEST will split the name and boost value and set the Boost property; a boost value as part of the string takes precedence over a boost value that may also be passed as the second constructor argument

Field fieldString = "name^2";
var fieldStringConstructor = new Field("name^2");
var fieldStringCreate = new Field("name^2", 3); 


NEST will take the boost from the name

Implicit Conversionedit

As well as using the constructor, you can also implicitly convert string, PropertyInfo and member access lambda expressions to a Field. For expressions however, this is still rather involved as the expression first needs to be assigned to a variable that explicitly specifies the expression delegate type.

Field fieldString = "name";

Field fieldProperty = typeof(Project).GetProperty(nameof(Project.Name));

Expression<Func<Project, object>> expression = p => p.Name;
Field fieldExpression = expression;


Using Nest.Infer methodsedit

To ease creating a Field instance from expressions, there is a static Infer class you can use

This example uses the static import using static Nest.Infer; in the using directives to shorthand Nest.Infer.Field<T>() to simply Field<T>(). Be sure to include this static import if copying any of these examples.

Field fieldString = "name";

but for expressions this is still rather involved

var fieldExpression = Field<Project>(p => p.Name);

this can be even shortened even further using a static import. Now that is much terser than our first example using the constructor!

fieldExpression = Field<Project>(p => p.Name);


You can specify boosts in the field using a string, as well as using Nest.Infer.Field

fieldString = "name^2.1";

fieldExpression = Field<Project>(p => p.Name, 2.1);


Field name casingedit

By default, NEST camelcases all field names to better align with typical JavaScript and JSON conventions

using DefaultFieldNameInferrer() on ConnectionSettings you can change this behavior

var setup = WithConnectionSettings(s => s.DefaultFieldNameInferrer(p => p.ToUpper()));

setup.Expect("NAME").WhenSerializing(Field<Project>(p => p.Name));

A Field constructed from a string however is always passed along verbatim


If you’d like NEST to not change the casing of field names at all, simply pass a Func<string,string> to DefaultFieldNameInferrer that simply returns the input string

setup = WithConnectionSettings(s => s.DefaultFieldNameInferrer(p => p));
setup.Expect("Name").WhenSerializing(Field<Project>(p => p.Name));

Complex field name expressionsedit

You can follow your property expression to any depth. Here we are traversing to the LeadDeveloper FirstName

Expect("leadDeveloper.firstName").WhenSerializing(Field<Project>(p => p.LeadDeveloper.FirstName));

When dealing with collection indexers, the indexer access is ignored allowing you to traverse into properties of collections

Expect("curatedTags").WhenSerializing(Field<Project>(p => p.CuratedTags[0]));

Similarly, LINQ’s .First() method also works

Expect("curatedTags").WhenSerializing(Field<Project>(p => p.CuratedTags.First()));
Expect("curatedTags.added").WhenSerializing(Field<Project>(p => p.CuratedTags[0].Added));
Expect("").WhenSerializing(Field<Project>(p => p.CuratedTags.First().Name));

Remember, these are expressions to access members, and not actual code that will be executed

An indexer on a dictionary is assumed to describe a property name

Expect("metadata.hardcoded").WhenSerializing(Field<Project>(p => p.Metadata["hardcoded"]));
Expect("metadata.hardcoded.created").WhenSerializing(Field<Project>(p => p.Metadata["hardcoded"].Created));

A cool feature here is that NEST will evaluate variables passed to an indexer

var variable = "var";
Expect("metadata.var").WhenSerializing(Field<Project>(p => p.Metadata[variable]));
Expect("metadata.var.created").WhenSerializing(Field<Project>(p => p.Metadata[variable].Created));

If you are using Elasticearch’s multi-fields, which you really should as they allow you to analyze a string in a number of different ways, these "virtual" sub fields do not always map back on to your POCO. By calling .Suffix() on expressions, you describe the sub fields that should be mapped and how they are mapped

    Field<Project>(p => p.LeadDeveloper.FirstName.Suffix("raw")));

    Field<Project>(p => p.CuratedTags[0].Suffix("raw")));

    Field<Project>(p => p.CuratedTags.First().Suffix("raw")));

    Field<Project>(p => p.CuratedTags[0].Added.Suffix("raw")));

    Field<Project>(p => p.Metadata["hardcoded"].Suffix("raw")));

    Field<Project>(p => p.Metadata["hardcoded"].Created.Suffix("raw")));

You can even chain .Suffix() calls to any depth!

    Field<Project>(p => p.CuratedTags.First().Name.Suffix("raw").Suffix("evendeeper")));

Variables passed to suffix will be evaluated as well

var suffix = "unanalyzed";
    Field<Project>(p => p.Metadata[variable].Suffix(suffix)));

    Field<Project>(p => p.Metadata[variable].Created.Suffix(suffix)));

Suffixes can also be appended to expressions using .AppendSuffix(). This is useful in cases where you want to apply the same suffix to a list of fields.

Here we have a list of expressions

var expressions = new List<Expression<Func<Project, object>>>
    p => p.Name,
    p => p.Description,
    p => p.CuratedTags.First().Name,
    p => p.LeadDeveloper.FirstName,
    p => p.Metadata["hardcoded"]

and we want to append the suffix "raw" to each

var fieldExpressions =
    expressions.Select<Expression<Func<Project, object>>, Field>(e => e.AppendSuffix("raw")).ToList();


or we might even want to chain multiple .AppendSuffix() calls

var multiSuffixFieldExpressions =
    expressions.Select<Expression<Func<Project, object>>, Field>(e => e.AppendSuffix("raw").AppendSuffix("evendeeper")).ToList();


Member Expressions onlyedit

The expression passed to Field should only be a MemberExpression

var fieldExpression = Field<Project>(p => p.Name + 2);

Attribute based namingedit

Using NEST’s property attributes you can specify a new name for the properties

public class BuiltIn
    [Text(Name = "naam")]
    public string Name { get; set; }

Expect("naam").WhenSerializing(Field<BuiltIn>(p => p.Name));

DataMember attributesedit

If a property has a System.Runtime.Serialization.DataMemberAttribute applied, this can be used to resolve a field value for a property

public class DataMember
    [DataMember(Name = "nameFromDataMember")]
    public string Name { get; set; }

Expect("nameFromDataMember").WhenSerializing(Field<DataMember>(p => p.Name));

Serializer specific attributesedit

NEST can also use a serializer specific attribute to resolve a field value for a property. In this example, the JsonNetSerializer is hooked up as the custom serializer for the client and we use the JsonPropertyAttribute to resolve a field

public class SerializerSpecific
    [PropertyName("nameInJson"), JsonProperty("nameInJson")]
    public string Name { get; set; }

Expect("nameInJson").WhenSerializing(Field<SerializerSpecific>(p => p.Name));

If both a NEST property attribute and a serializer specific attribute are present on a property, NEST attributes take precedence

public class Both
    [Text(Name = "naam")]
    [PropertyName("nameInJson"), DataMember(Name = "nameInJson")]
    public string Name { get; set; }

Expect("naam").WhenSerializing(Field<Both>(p => p.Name));
        naam = "Martijn Laarman"
    }).WhenSerializing(new Both { Name = "Martijn Laarman" });

Field Inference Cachingedit

Resolution of field names is cached per ConnectionSettings instance. To demonstrate, take the following simple POCOs

class A { public C C { get; set; } }

class B { public C C { get; set; } }

class C
    public string Name { get; set; }

var client = TestClient.Default;

var fieldNameOnA = client.Infer.Field(Field<A>(p => p.C.Name));
var fieldNameOnB = client.Infer.Field(Field<B>(p => p.C.Name));

Here we have two similarly shaped expressions, one coming from A and one from B that will resolve to the same field name, as expected


now we create a new connection settings with a re-map for C on class A to "d" now when we resolve the field path for property C on A, it will be different than for property C on B

var newConnectionSettings = new TestConnectionSettings()
    .DefaultMappingFor<A>(m => m
        .PropertyName(p => p.C, "d")

var newClient = new ElasticClient(newConnectionSettings);

fieldNameOnA = newClient.Infer.Field(Field<A>(p => p.C.Name));
fieldNameOnB = newClient.Infer.Field(Field<B>(p => p.C.Name));


however we didn’t break inference on the first client instance using its separate connection settings

fieldNameOnA = client.Infer.Field(Field<A>(p => p.C.Name));
fieldNameOnB = client.Infer.Field(Field<B>(p => p.C.Name));


Inference Precedenceedit

To wrap up, the precedence in which field names are inferred is:

  1. A naming of the property on ConnectionSettings using .PropertyName()
  2. A NEST PropertyNameAttribute
  3. Ask the serializer if the property has a verbatim value, e.g. it has a JsonPropertyAttribute if using JsonNetSerializer
  4. See if the MemberInfo has a DataMemberAttribute applied
  5. Pass the MemberInfo to the DefaultFieldNameInferrer, which by default will camel case the Name property

The following example class will demonstrate this precedence

private class Precedence
    [Text(Name = "renamedIgnoresNest")]
    public string RenamedOnConnectionSettings { get; set; } 

    [Text(Name = "nestAtt")]
    public string NestAttribute { get; set; } 

    public string NestProperty { get; set; } 

    [DataMember(Name ="jsonProp")]
    public string JsonProperty { get; set; } 

    public string AskSerializer { get; set; } 

    [DataMember(Name = "data")]
    public string DataMember { get; set; }

    public string DefaultFieldNameInferrer { get; set; } 

Even though this property has various attributes applied we provide an override on ConnectionSettings later that takes precedence.

Has a TextAttribute, PropertyNameAttribute and a JsonPropertyAttribute - the TextAttribute takes precedence.

Has both a PropertyNameAttribute and a JsonPropertyAttribute - the PropertyNameAttribute takes precedence.

JsonPropertyAttribute takes precedence.

This property we are going to hard code in our custom serializer to resolve to ask.

We are going to register a DefaultFieldNameInferrer on ConnectionSettings that will uppercase all properties.

We’ll create a custom IPropertyMappingProvider that renames any property named AskSerializer to ask. and hook it up when creating the Connection Settings in the following section.

private class CustomPropertyMappingProvider : PropertyMappingProvider
    public override IPropertyMapping CreatePropertyMapping(MemberInfo memberInfo)
        return memberInfo.Name == nameof(Precedence.AskSerializer)
            ? new PropertyMapping { Name = "ask" }
            : base.CreatePropertyMapping(memberInfo);

Now, when we create the Connection Settings to use to configure the client, we’ll add

  • a default mapping for the Precedence type
  • our CustomPropertyMappingProvider
  • a delegate to perform default field name inference
var usingSettings = WithConnectionSettings(s => s

    .DefaultMappingFor<Precedence>(m => m
        .PropertyName(p => p.RenamedOnConnectionSettings, "renamed") 
    .DefaultFieldNameInferrer(p => p.ToUpperInvariant()) 
).WithPropertyMappingProvider(new CustomPropertyMappingProvider()); 

usingSettings.Expect("renamed").ForField(Field<Precedence>(p => p.RenamedOnConnectionSettings));
usingSettings.Expect("nestAtt").ForField(Field<Precedence>(p => p.NestAttribute));
usingSettings.Expect("nestProp").ForField(Field<Precedence>(p => p.NestProperty));
usingSettings.Expect("jsonProp").ForField(Field<Precedence>(p => p.JsonProperty));
usingSettings.Expect("ask").ForField(Field<Precedence>(p => p.AskSerializer));
usingSettings.Expect("data").ForField(Field<Precedence>(p => p.DataMember));
usingSettings.Expect("DEFAULTFIELDNAMEINFERRER").ForField(Field<Precedence>(p => p.DefaultFieldNameInferrer));

Rename on the mapping for the Precedence type

Default inference for a field, if no other rules apply or are specified for a given field

Hook up the custom IPropertyMappingProvider

The same naming rules also apply when indexing a document

usingSettings.Expect(new []
}).AsPropertiesOf(new Precedence
    RenamedOnConnectionSettings = "renamed on connection settings",
    NestAttribute = "using a nest attribute",
    NestProperty = "using a nest property",
    JsonProperty = "the default serializer resolves json property attributes",
    AskSerializer = "serializer fiddled with this one",
    DefaultFieldNameInferrer = "shouting much?",
    DataMember = "using a DataMember attribute"

Overriding inherited field inferenceedit

Properties inherited from a base type can be ignored and renamed using DefaultMappingFor<T> for a given type, on Connection Settings.

To demonstrate, the IgnoreMe property on Parent can be ignored on the Child type, and the Description property renamed, using DefaultMappingFor<Child>(...)

public class Parent
    public int Id { get; set; }
    public string Description { get; set; }
    public string IgnoreMe { get; set; }

public class Child : Parent { }

var usingSettings = WithConnectionSettings(s => s
        .DefaultMappingFor<Child>(m => m
            .PropertyName(p => p.Description, "desc")
            .Ignore(p => p.IgnoreMe)
usingSettings.Expect(new []
    }).AsPropertiesOf(new Child
        Id = 1,
        Description = "this property will be renamed for Child",
        IgnoreMe = "this property will be ignored (won't be serialized) for Child",

public class SourceModel
    public GeoModel Geo { get; set; }

public class GeoModel
    [DataMember(Name = "country_iso_code")]
    public string CountryIsoCode { get; set; }

var usingSettings = WithConnectionSettings(s => s)

usingSettings.Expect(new []
    }).AsPropertiesOf(new GeoModel { CountryIsoCode = "nl" });
usingSettings.Expect(new []
    }).AsPropertiesOf(new SourceModel { Geo = new GeoModel { CountryIsoCode = "nl" } });