We Keep Coding

Is HashSet of a HashSet<T> the right data model here?

I'm managing a data structure that looks like the following
[
{
"name": "Yada yada",
"dataModels": [
{
"entity": "Table",
"columns": [
[
{
"column": "ColumnA",
"value": " \"StringA\""
},
{
"column": "ColumnB",
"value": " \"StringB\""
}
],
...,
[
{
"column": "ColumnA",
"value": " \"StringA\""
},
{
"column": "ColumnB",
"value": " \"StringB\""
}
],
...
]
}
]
}
]
Every object in the columns list is a HashSet of ColumnValues, a class I created and is defined the following way:
public class ColumnValues
{
private string column;
private string value;
public ColumnValues(string col, string val)
{
column = col;
value = val;
}
public string Column
{
get {return column;}
}
public string Value
{
get {return value;}
}
public override bool Equals(object obj)
{
return obj is ColumnValues values &&
column == values.column &&
value == values.value &&
Column == values.Column &&
Value == values.Value;
}
public override int GetHashCode()
{
return HashCode.Combine(column, value, Column, Value);
}
}
So far, so good. However the issue is that in columns, I'm having duplicates of lists. columns is a field in DataModel, another class:
public class DataModel
{
private string entity;
private List<HashSet<ColumnValues>> columns;
private string rule;
public DataModel(string entityName, List<HashSet<ColumnValues>> columnList)
{
entity = entityName;
columns = columnList;
}
public string Entity
{
get { return entity; }
}
public List<HashSet<ColumnValues>> Columns
{
get { return columns; }
set { columns = value; }
}
public string Rule
{
set { rule = value; }
}
}
I'm not understanding why the Set is allowing the existence of duplicates.
I do add that I'm returning the columns list after applying Distinct().ToList():
var dataModels = new List<DataModel>();
GatherColumns(code.Syntax);
dataModels.ForEach(dataModel => dataModel.Columns = dataModel.Columns.Distinct().ToList());
return dataModels;
TIA!
EDIT:
As per Tim's comment, I've provided the override methods Equals and GetHashCode.
I've also decided to define Columns in DataModel as a List of HashSets.
However, unfortunately the result hasn't changed.
EDIT N°2:
I've decided to create a class implementing the IEqualityComparer interface. However, little effect so far:
public class ColumnValuesComparer : IEqualityComparer<HashSet<ColumnValues>>
{
public bool Equals(HashSet<ColumnValues> c1, HashSet<ColumnValues> c2)
{
return c1.Equals(c2);
}
public int GetHashCode(HashSet<ColumnValues> obj)
{
return obj.GetHashCode();
}
}
Used here:
var dataModels = new List<DataModel>();
GatherColumns(code.Syntax);
ColumnValuesComparer comparer = new ColumnValuesComparer();
dataModels.ForEach(dataModel => dataModel.Columns = dataModel.Columns.Distinct(comparer).ToList());
return dataModels;
At this point I'm pretty much stuck, and don't know where to go moving forward.

After a bit of searching a documentation, I fell upon the concept of SetComparer.
They allow me to create objects that evaluate deep nested equality within HashSets.
So, when I try to create my HashSet of HashSets, I must pass that method:
var set = new HashSet<HashSet<ColumnValues>>(HashSet<ColumnValues>.CreateSetComparer());
set.Add(columnValues);

Controlling JSON .NET's reference ID generation

I'd like to be able to control how JSON .NET generates its meta reference IDs such as "$id": "1". Take the following code:
public class Person
{
public string Name { get; set; }
public Person Mother { get; set; }
}
.
var settings = new Newtonsoft.Json.JsonSerializerSettings
{
PreserveReferencesHandling = Newtonsoft.Json.PreserveReferencesHandling.Objects,
Formatting = Newtonsoft.Json.Formatting.Indented
};
Newtonsoft.Json.JsonConvert.DefaultSettings = () => settings;
var person = new Person
{
Name = "bob",
Mother = new Person { Name = "jane" }
};
var personJson = JsonConvert.SerializeObject(person);
var motherJson = JsonConvert.SerializeObject(person.Mother);
The JSON for person looks like this:
{
"$id": "1",
"Name": "bob",
"Mother": {
"$id": "2",
"Name": "jane",
"Mother": null
}
}
However, if I serialize person.Mother directly, the JSON looks like this:
{
"$id": "1",
"Name": "jane",
"Mother": null
}
In the first JSON, Jane is "$id": "2", but serializing Jane directly is "$id": "1". This is the behavior I'd expect under normal conditions as the serializer assigns the IDs in the order it traverses the objects, but I'd really like to override the ID generation so that I could make it a hash of the object reference itself. This way, Jane would generate the same ID per running instance of the program every time regardless if serialized as a member of a parent or serialized individually.
UPDATE
Per sample code in selected answer and recommendation in comment, I have used IReferenceResolver. It turns out that I can't use it, though, but I'll include the code below anyway. The reason why this won't work is because I am trying to bastardize JSON.NET as a quick and dirty cloning tool, so I can't fault it for not suiting my needs. I've since fallen back on my own custom cloning utility, so I no longer need this.
public class ObjectReferenceResolver : Newtonsoft.Json.Serialization.IReferenceResolver
{
readonly Dictionary<object, int> objectDic = new Dictionary<object, int>();
int maxId = 0;
//Called during serialization
public string GetReference(object context, object value)
{
//This method will return the meta $id that you choose. In this example, I am storing
//object references in a dictionary with an incremented ID. If the reference exists, I
//return its ID. Otherwise, I increment the ID and add the reference to the dictionary.
var id = 0;
if (objectDic.ContainsKey(value))
{
id = objectDic[value];
}
else
{
objectDic[value] = maxId++;
}
return id.ToString();
}
//Called during serialization
public bool IsReferenced(object context, object value)
{
//Return whether or not value exists in a reference bank.
//If false, the JSON will return as a full JSON object with "$id": "x"
//If true, the JSON will return "$ref": "x"
return objectDic.ContainsKey(value);
}
//Called during deserialization
public void AddReference(object context, string reference, object value)
{
//This method is called after the deserializer has created a new instance of the
//object. At this time, it's only the initial instance and no properties have been set.
//This method presents a problem because it does not allow you to create the instance or
//retrieve it from a repo and then return it for later use by the reference resolver.
//Therefore, I have to find the existing object by $id, remove it, and then add the new
//object created by the deseralizer. This creates the possibility for two instances of
//the same data object to exist within the running application, so, unfortunately, this
//will not work.
var e = objectDic.First(x => x.Value.ToString() == reference).Key;
objectDic.Remove(e);
objectDic[value] = reference.ParseInt().Value;
}
//Called during deserialization
public object ResolveReference(object context, string reference)
{
//This method retrieves an existing reference by $id and returns it.
var value = objectDic.FirstOrDefault(x => x.Value.ToString() == reference).Key;
return value;
}
}

As per others have recommended, you need a custom IReferenceResolver:
class PersonNameAsIdResolver : IReferenceResolver
{
public void AddReference(object context, string reference, object value)
{
// This method is called during deserialize for $id
}
public string GetReference(object context, object value)
{
// Returns person name as value of $id
return ((Person)value).Name;
}
public bool IsReferenced(object context, object value)
{
// Returns false, so that $id is used, not $ref.
return false;
}
public object ResolveReference(object context, string reference)
{
// This method is called during deserialize for $ref
return null;
}
}
To use that:
var settings = new Newtonsoft.Json.JsonSerializerSettings
{
PreserveReferencesHandling = Newtonsoft.Json.PreserveReferencesHandling.Objects,
Formatting = Newtonsoft.Json.Formatting.Indented
};
settings.ReferenceResolverProvider = ()=> new PersonNameAsIdResolver();
UPDATE
Answer to the OP's update
AddReference is called while an object is being populated, so it has been too late to replace the object. To be able to find and populate desired object, you need a JsonConverter, which is called before the reference resolver:
class PersonJsonConverter : JsonConverter
{
private readonly PersonNameAsIdResolver _idResolver;
public PersonJsonConverter(PersonNameAsIdResolver idResolver)
{
_idResolver = idResolver;
}
public override bool CanConvert(Type objectType)
=> objectType == typeof(Person);
// Can't write. There's nothing to changing for writing scenario.
public override bool CanWrite => false;
public override object ReadJson(JsonReader reader, Type objectType,
object existingValue, JsonSerializer serializer)
{
var token = JToken.ReadFrom(reader);
if (token.Type == JTokenType.Null)
{
return null;
}
var obj = (JObject)token;
// The code below calls the resolver to find the existing instance.
// This can stop JSON.NET creating a new instance.
Person instance = null;
var #id = obj["$id"].Value<string>();
if (#id != null)
{
instance = (Person)_idResolver.ResolveReference(this, #id);
}
else
{
var #ref = obj["$ref"]?.Value<string>();
if (#ref != null)
{
instance = (Person)_idResolver.ResolveReference(this, #ref);
}
}
// Assuming can't resolve, create a new instance.
if (instance == null)
{
instance = new Person();
}
// This will populate existing Person object if found
serializer.Populate(obj.CreateReader(), instance);
return instance;
}
public override void WriteJson(JsonWriter writer, object value,
JsonSerializer serializer)
{
throw new NotSupportedException();
}
}
And the default serialization settings should look like:
var settings = new Newtonsoft.Json.JsonSerializerSettings
{
PreserveReferencesHandling = Newtonsoft.Json.PreserveReferencesHandling.Objects,
Formatting = Newtonsoft.Json.Formatting.Indented
};
var idResolver = new PersonNameAsIdResolver();
settings.Converters.Add(new PersonJsonConverter(idResolver));
settings.ReferenceResolverProvider = () => idResolver;

A possible solution would be the following:
Replace PreserveReferncesHandling from Objects to None:
PreserveReferencesHandling = Newtonsoft.Json.PreserveReferencesHandling.None
Add an Id property in Person class:
public class Person
{
public string Id { get; set; }
public string Name { get; set; }
public Person Mother { get; set; }
}
The complete solution is as follows:
using System;
using Newtonsoft.Json;
namespace ControlJsonId
{
class Program
{
static void Main(string[] args)
{
Console.WriteLine("INICIO");
var settings = new Newtonsoft.Json.JsonSerializerSettings
{
PreserveReferencesHandling = Newtonsoft.Json.PreserveReferencesHandling.None,
Formatting = Newtonsoft.Json.Formatting.Indented,
};
Newtonsoft.Json.JsonConvert.DefaultSettings = () => settings;
var person = new Person
{
Id = Guid.NewGuid().ToString(),
Name = "bob",
Mother = new Person { Id = string.Empty, Name = "jane" }
};
var personJson = JsonConvert.SerializeObject(person);
Console.WriteLine(personJson);
var motherJson = JsonConvert.SerializeObject(person.Mother);
Console.WriteLine(motherJson);
Console.WriteLine("FIN");
Console.ReadKey();
}
}
public class Person
{
public string Id { get; set; }
public string Name { get; set; }
public Person Mother { get; set; }
}
}
The result is:
INICIO
{
"Id": "76d6b5f0-2be8-4d1d-aafe-fe1b4b7d6ae1",
"Name": "bob",
"Mother": {
"Id": "",
"Name": "jane",
"Mother": null
}
}
{
"Id": "",
"Name": "jane",
"Mother": null
}
FIN

How do I include subclasses in Swagger API documentation/ OpenAPI specification using Swashbuckle?

I have an Asp.Net web API 5.2 project in c# and generating documentation with Swashbuckle.
I have model that contain inheritance something like having an Animal property from an Animal abstract class and Dog and Cat classes that derive from it.
Swashbuckle only shows the schema for the Animal class so I tried to play with ISchemaFilter (that what they suggest too) but I couldn't make it work and also I cannot find a proper example.
Anybody can help?

It seems Swashbuckle doesn't implement polymorphism correctly and I understand the point of view of the author about subclasses as parameters (if an action expects an Animal class and behaves differently if you call it with a dog object or a cat object, then you should have 2 different actions...) but as return types I believe that it is correct to return Animal and the objects could be Dog or Cat types.
So to describe my API and produce a proper JSON schema in line with correct guidelines (be aware of the way I describe the disciminator, if you have your own discriminator you may need to change that part in particular), I use document and schema filters as follows:
SwaggerDocsConfig configuration;
.....
configuration.DocumentFilter<PolymorphismDocumentFilter<YourBaseClass>>();
configuration.SchemaFilter<PolymorphismSchemaFilter<YourBaseClass>>();
.....
public class PolymorphismSchemaFilter<T> : ISchemaFilter
{
private readonly Lazy<HashSet<Type>> derivedTypes = new Lazy<HashSet<Type>>(Init);
private static HashSet<Type> Init()
{
var abstractType = typeof(T);
var dTypes = abstractType.Assembly
.GetTypes()
.Where(x => abstractType != x && abstractType.IsAssignableFrom(x));
var result = new HashSet<Type>();
foreach (var item in dTypes)
result.Add(item);
return result;
}
public void Apply(Schema schema, SchemaRegistry schemaRegistry, Type type)
{
if (!derivedTypes.Value.Contains(type)) return;
var clonedSchema = new Schema
{
properties = schema.properties,
type = schema.type,
required = schema.required
};
//schemaRegistry.Definitions[typeof(T).Name]; does not work correctly in SwashBuckle
var parentSchema = new Schema { #ref = "#/definitions/" + typeof(T).Name };
schema.allOf = new List<Schema> { parentSchema, clonedSchema };
//reset properties for they are included in allOf, should be null but code does not handle it
schema.properties = new Dictionary<string, Schema>();
}
}
public class PolymorphismDocumentFilter<T> : IDocumentFilter
{
public void Apply(SwaggerDocument swaggerDoc, SchemaRegistry schemaRegistry, System.Web.Http.Description.IApiExplorer apiExplorer)
{
RegisterSubClasses(schemaRegistry, typeof(T));
}
private static void RegisterSubClasses(SchemaRegistry schemaRegistry, Type abstractType)
{
const string discriminatorName = "discriminator";
var parentSchema = schemaRegistry.Definitions[SchemaIdProvider.GetSchemaId(abstractType)];
//set up a discriminator property (it must be required)
parentSchema.discriminator = discriminatorName;
parentSchema.required = new List<string> { discriminatorName };
if (!parentSchema.properties.ContainsKey(discriminatorName))
parentSchema.properties.Add(discriminatorName, new Schema { type = "string" });
//register all subclasses
var derivedTypes = abstractType.Assembly
.GetTypes()
.Where(x => abstractType != x && abstractType.IsAssignableFrom(x));
foreach (var item in derivedTypes)
schemaRegistry.GetOrRegister(item);
}
}
What the previous code implements is specified here, in the section "Models with Polymorphism Support. It basically produces something like the following:
{
"definitions": {
"Pet": {
"type": "object",
"discriminator": "petType",
"properties": {
"name": {
"type": "string"
},
"petType": {
"type": "string"
}
},
"required": [
"name",
"petType"
]
},
"Cat": {
"description": "A representation of a cat",
"allOf": [
{
"$ref": "#/definitions/Pet"
},
{
"type": "object",
"properties": {
"huntingSkill": {
"type": "string",
"description": "The measured skill for hunting",
"default": "lazy",
"enum": [
"clueless",
"lazy",
"adventurous",
"aggressive"
]
}
},
"required": [
"huntingSkill"
]
}
]
},
"Dog": {
"description": "A representation of a dog",
"allOf": [
{
"$ref": "#/definitions/Pet"
},
{
"type": "object",
"properties": {
"packSize": {
"type": "integer",
"format": "int32",
"description": "the size of the pack the dog is from",
"default": 0,
"minimum": 0
}
},
"required": [
"packSize"
]
}
]
}
}
}

To follow on from Paulo's great answer, if you're using Swagger 2.0, you'll need to modify the classes as shown:
public class PolymorphismSchemaFilter<T> : ISchemaFilter
{
private readonly Lazy<HashSet<Type>> derivedTypes = new Lazy<HashSet<Type>>(Init);
private static HashSet<Type> Init()
{
var abstractType = typeof(T);
var dTypes = abstractType.Assembly
.GetTypes()
.Where(x => abstractType != x && abstractType.IsAssignableFrom(x));
var result = new HashSet<Type>();
foreach (var item in dTypes)
result.Add(item);
return result;
}
public void Apply(Schema model, SchemaFilterContext context)
{
if (!derivedTypes.Value.Contains(context.SystemType)) return;
var clonedSchema = new Schema
{
Properties = model.Properties,
Type = model.Type,
Required = model.Required
};
//schemaRegistry.Definitions[typeof(T).Name]; does not work correctly in SwashBuckle
var parentSchema = new Schema { Ref = "#/definitions/" + typeof(T).Name };
model.AllOf = new List<Schema> { parentSchema, clonedSchema };
//reset properties for they are included in allOf, should be null but code does not handle it
model.Properties = new Dictionary<string, Schema>();
}
}
public class PolymorphismDocumentFilter<T> : IDocumentFilter
{
private static void RegisterSubClasses(ISchemaRegistry schemaRegistry, Type abstractType)
{
const string discriminatorName = "discriminator";
var parentSchema = schemaRegistry.Definitions[abstractType.Name];
//set up a discriminator property (it must be required)
parentSchema.Discriminator = discriminatorName;
parentSchema.Required = new List<string> { discriminatorName };
if (!parentSchema.Properties.ContainsKey(discriminatorName))
parentSchema.Properties.Add(discriminatorName, new Schema { Type = "string" });
//register all subclasses
var derivedTypes = abstractType.Assembly
.GetTypes()
.Where(x => abstractType != x && abstractType.IsAssignableFrom(x));
foreach (var item in derivedTypes)
schemaRegistry.GetOrRegister(item);
}
public void Apply(SwaggerDocument swaggerDoc, DocumentFilterContext context)
{
RegisterSubClasses(context.SchemaRegistry, typeof(T));
}
}

As of this merge into Swashbuckle.AspNetCore, you can get basic support for polymorphic schemas by using:
services.AddSwaggerGen(c =>
{
c.GeneratePolymorphicSchemas();
}
You can also express your derived types via attributes present in the Annotations library:
[SwaggerSubTypes(typeof(SubClass), Discriminator = "value")]
This article goes into further detail as to how you can deserialize derived types using Newtonsoft.

I'd like to follow up on Craig's answer.
If you use NSwag to generate TypeScript definitions from the Swagger API documentation generated with Swashbuckle (3.x at the time of writing) using the method explained in Paulo's answer and further enhanced in Craig's answer you will probably face the following problems:
Generated TypeScript definitions will have duplicate properties even though the generated classes will extend the base classes. Consider the following C# classes:
public abstract class BaseClass
{
public string BaseProperty { get; set; }
}
public class ChildClass : BaseClass
{
public string ChildProperty { get; set; }
}
When using the aforementioned answers, the resulting TypeScript definition of IBaseClass and IChildClass interfaces would look like this:
export interface IBaseClass {
baseProperty : string | undefined;
}
export interface IChildClass extends IBaseClass {
baseProperty : string | undefined;
childProperty: string | undefined;
}
As you can see, the baseProperty is incorrectly defined in both base and child classes. To solve this, we can modify the Apply method of the PolymorphismSchemaFilter<T> class to include only owned properties to the schema, i.e. to exclude the inherited properties from the current types schema. Here is an example:
public void Apply(Schema model, SchemaFilterContext context)
{
...
// Prepare a dictionary of inherited properties
var inheritedProperties = context.SystemType.GetProperties()
.Where(x => x.DeclaringType != context.SystemType)
.ToDictionary(x => x.Name, StringComparer.OrdinalIgnoreCase);
var clonedSchema = new Schema
{
// Exclude inherited properties. If not excluded,
// they would have appeared twice in nswag-generated typescript definition
Properties =
model.Properties.Where(x => !inheritedProperties.ContainsKey(x.Key))
.ToDictionary(x => x.Key, x => x.Value),
Type = model.Type,
Required = model.Required
};
...
}
Generated TypeScript definitions will not reference properties from any existing intermediate abstract classes. Consider the following C# classes:
public abstract class SuperClass
{
public string SuperProperty { get; set; }
}
public abstract class IntermediateClass : SuperClass
{
public string IntermediateProperty { get; set; }
}
public class ChildClass : BaseClass
{
public string ChildProperty { get; set; }
}
In this case, the generated TypeScript definitions would look like this:
export interface ISuperClass {
superProperty: string | undefined;
}
export interface IIntermediateClass extends ISuperClass {
intermediateProperty : string | undefined;
}
export interface IChildClass extends ISuperClass {
childProperty: string | undefined;
}
Notice how the generated IChildClass interface extends ISuperClass directly, ignoring the IIntermediateClass interface, effectively leaving any instance of IChildClass without the intermediateProperty property.
We can use the following code to solve this problem:
public void Apply(Schema model, SchemaFilterContext context)
{
...
// Use the BaseType name for parentSchema instead of typeof(T),
// because we could have more classes in the hierarchy
var parentSchema = new Schema
{
Ref = "#/definitions/" + (context.SystemType.BaseType?.Name ?? typeof(T).Name)
};
...
}
This will ensure that the child class correctly references the intermediate class.
In conclusion, the final code would then look like this:
public void Apply(Schema model, SchemaFilterContext context)
{
if (!derivedTypes.Value.Contains(context.SystemType))
{
return;
}
// Prepare a dictionary of inherited properties
var inheritedProperties = context.SystemType.GetProperties()
.Where(x => x.DeclaringType != context.SystemType)
.ToDictionary(x => x.Name, StringComparer.OrdinalIgnoreCase);
var clonedSchema = new Schema
{
// Exclude inherited properties. If not excluded,
// they would have appeared twice in nswag-generated typescript definition
Properties =
model.Properties.Where(x => !inheritedProperties.ContainsKey(x.Key))
.ToDictionary(x => x.Key, x => x.Value),
Type = model.Type,
Required = model.Required
};
// Use the BaseType name for parentSchema instead of typeof(T),
// because we could have more abstract classes in the hierarchy
var parentSchema = new Schema
{
Ref = "#/definitions/" + (context.SystemType.BaseType?.Name ?? typeof(T).Name)
};
model.AllOf = new List<Schema> { parentSchema, clonedSchema };
// reset properties for they are included in allOf, should be null but code does not handle it
model.Properties = new Dictionary<string, Schema>();
}

We recently upgraded to .NET Core 3.1 and Swashbuckle.AspNetCore 5.0
And the API is somewhat changed.
In case somebody needs this filter here is the code with minimal changes to get similar behavior:
public class PolymorphismDocumentFilter<T> : IDocumentFilter
{
public void Apply(OpenApiDocument swaggerDoc, DocumentFilterContext context)
{
RegisterSubClasses(context.SchemaRepository, context.SchemaGenerator, typeof(T));
}
private static void RegisterSubClasses(SchemaRepository schemaRegistry, ISchemaGenerator schemaGenerator, Type abstractType)
{
const string discriminatorName = "$type";
OpenApiSchema parentSchema = null;
if (schemaRegistry.TryGetIdFor(abstractType, out string parentSchemaId))
parentSchema = schemaRegistry.Schemas[parentSchemaId];
else
parentSchema = schemaRegistry.GetOrAdd(abstractType, parentSchemaId, () => new OpenApiSchema());
// set up a discriminator property (it must be required)
parentSchema.Discriminator = new OpenApiDiscriminator() { PropertyName = discriminatorName };
parentSchema.Required = new HashSet<string> { discriminatorName };
if (parentSchema.Properties == null)
parentSchema.Properties = new Dictionary<string, OpenApiSchema>();
if (!parentSchema.Properties.ContainsKey(discriminatorName))
parentSchema.Properties.Add(discriminatorName, new OpenApiSchema() { Type = "string", Default = new OpenApiString(abstractType.FullName) });
// register all subclasses
var derivedTypes = abstractType.GetTypeInfo().Assembly.GetTypes()
.Where(x => abstractType != x && abstractType.IsAssignableFrom(x));
foreach (var item in derivedTypes)
schemaGenerator.GenerateSchema(item, schemaRegistry);
}
}
public class PolymorphismSchemaFilter<T> : ISchemaFilter
{
private readonly Lazy<HashSet<Type>> derivedTypes = new Lazy<HashSet<Type>>(Init);
public void Apply(OpenApiSchema schema, SchemaFilterContext context)
{
if (!derivedTypes.Value.Contains(context.Type)) return;
Type type = context.Type;
var clonedSchema = new OpenApiSchema
{
Properties = schema.Properties,
Type = schema.Type,
Required = schema.Required
};
// schemaRegistry.Definitions[typeof(T).Name]; does not work correctly in Swashbuckle.AspNetCore
var parentSchema = new OpenApiSchema
{
Reference = new OpenApiReference() { ExternalResource = "#/definitions/" + typeof(T).Name }
};
var assemblyName = Assembly.GetAssembly(type).GetName();
schema.Discriminator = new OpenApiDiscriminator() { PropertyName = "$type" };
// This is required if you use Microsoft's AutoRest client to generate the JavaScript/TypeScript models
schema.Extensions.Add("x-ms-discriminator-value", new OpenApiObject() { ["name"] = new OpenApiString($"{type.FullName}, {assemblyName.Name}") });
schema.AllOf = new List<OpenApiSchema> { parentSchema, clonedSchema };
// reset properties for they are included in allOf, should be null but code does not handle it
schema.Properties = new Dictionary<string, OpenApiSchema>();
}
private static HashSet<Type> Init()
{
var abstractType = typeof(T);
var dTypes = abstractType.GetTypeInfo().Assembly
.GetTypes()
.Where(x => abstractType != x && abstractType.IsAssignableFrom(x));
var result = new HashSet<Type>();
foreach (var item in dTypes)
result.Add(item);
return result;
}
}
I did not inspect the result fully, but it seems that it gives the same behavior.
Also note that you need to import these namespaces:
using Microsoft.OpenApi.Models;
using Microsoft.OpenApi.Any;
using System.Reflection;
using Swashbuckle.AspNetCore.SwaggerGen;

This works as of version 5.6.3:
services.AddSwaggerGen(options =>
{
options.UseOneOfForPolymorphism();
options.SelectDiscriminatorNameUsing(_ => "type");
});

Pass complex parameters to [Theory]

Xunit has a nice feature: you can create one test with a Theory attribute and put data in InlineData attributes, and xUnit will generate many tests, and test them all.
I want to have something like this, but the parameters to my method are not 'simple data' (like string, int, double), but a list of my class:
public static void WriteReportsToMemoryStream(
IEnumerable<MyCustomClass> listReport,
MemoryStream ms,
StreamWriter writer) { ... }

There are many xxxxData attributes in XUnit. Check out for example the MemberData attribute.
You can implement a property that returns IEnumerable<object[]>. Each object[] that this method generates will be then "unpacked" as a parameters for a single call to your [Theory] method.
See i.e. these examples from here
Here are some examples, just for a quick glance.
MemberData Example: just here at hand
public class StringTests2
{
[Theory, MemberData(nameof(SplitCountData))]
public void SplitCount(string input, int expectedCount)
{
var actualCount = input.Split(' ').Count();
Assert.Equal(expectedCount, actualCount);
}
public static IEnumerable<object[]> SplitCountData =>
new List<object[]>
{
new object[] { "xUnit", 1 },
new object[] { "is fun", 2 },
new object[] { "to test with", 3 }
};
}
XUnit < 2.0: Another option is ClassData, which works the same, but allows to easily share the 'generators' between tests in different classes/namespaces, and also separates the 'data generators' from the actual test methods.
ClassData Example
public class StringTests3
{
[Theory, ClassData(typeof(IndexOfData))]
public void IndexOf(string input, char letter, int expected)
{
var actual = input.IndexOf(letter);
Assert.Equal(expected, actual);
}
}
public class IndexOfData : IEnumerable<object[]>
{
private readonly List<object[]> _data = new List<object[]>
{
new object[] { "hello world", 'w', 6 },
new object[] { "goodnight moon", 'w', -1 }
};
public IEnumerator<object[]> GetEnumerator()
{ return _data.GetEnumerator(); }
IEnumerator IEnumerable.GetEnumerator()
{ return GetEnumerator(); }
}
XUnit >= 2.0: Instead of ClassData, now there's an 'overload' of [MemberData] that allows to use static members from other classes. Examples below have been updated to use it, since XUnit < 2.x is pretty ancient now.
Another option is ClassData, which works the same, but allows to easily share the 'generators' between tests in different classes/namespaces, and also separates the 'data generators' from the actual test methods.
MemberData Example: look there to another type
public class StringTests3
{
[Theory, MemberData(nameof(IndexOfData.SplitCountData), MemberType = typeof(IndexOfData))]
public void IndexOf(string input, char letter, int expected)
{
var actual = input.IndexOf(letter);
Assert.Equal(expected, actual);
}
}
public class IndexOfData : IEnumerable<object[]>
{
public static IEnumerable<object[]> SplitCountData =>
new List<object[]>
{
new object[] { "hello world", 'w', 6 },
new object[] { "goodnight moon", 'w', -1 }
};
}
Disclaimer :)
Last time checked #20210903 with dotnetfiddle.net on C# 5.0 and xunit 2.4.1 .. and failed. I couldn't mix-in a test-runner into that fiddle. But at least it compiled fine. Note that this was originally written years ago, things changed a little. I fixed them according to my hunch and comments. So.. it may contain inobvious typos, otherwise obvious bugs that would instantly pop up at runtime, and traces of milk & nuts.

Suppose that we have a complex Car class that has a Manufacturer class:
public class Car
{
public int Id { get; set; }
public long Price { get; set; }
public Manufacturer Manufacturer { get; set; }
}
public class Manufacturer
{
public string Name { get; set; }
public string Country { get; set; }
}
We're going to fill and pass the Car class to a Theory test.
So create a 'CarClassData' class that returns an instance of the Car class like below:
public class CarClassData : IEnumerable<object[]>
{
public IEnumerator<object[]> GetEnumerator()
{
yield return new object[] {
new Car
{
Id=1,
Price=36000000,
Manufacturer = new Manufacturer
{
Country="country",
Name="name"
}
}
};
}
IEnumerator IEnumerable.GetEnumerator() => GetEnumerator();
}
It's time for creating a test method(CarTest) and define the car as a parameter:
[Theory]
[ClassData(typeof(CarClassData))]
public void CarTest(Car car)
{
var output = car;
var result = _myRepository.BuyCar(car);
}
**If you're going to pass a list of car objects to Theory then change the CarClassData as follow:
public class CarClassData : IEnumerable<object[]>
{
public IEnumerator<object[]> GetEnumerator()
{
yield return new object[] {
new List<Car>()
{
new Car
{
Id=1,
Price=36000000,
Manufacturer = new Manufacturer
{
Country="Iran",
Name="arya"
}
},
new Car
{
Id=2,
Price=45000,
Manufacturer = new Manufacturer
{
Country="Torbat",
Name="kurosh"
}
}
}
};
}
IEnumerator IEnumerable.GetEnumerator() => GetEnumerator();
}
And the theory will be:
[Theory]
[ClassData(typeof(CarClassData))]
public void CarTest(List<Car> cars)
{
var output = cars;
}
Good Luck

To update #Quetzalcoatl's answer: The attribute [PropertyData] has been superseded by [MemberData] which takes as argument the string name of any static method, field, or property that returns an IEnumerable<object[]>. (I find it particularly nice to have an iterator method that can actually calculate test cases one at a time, yielding them up as they're computed.)
Each element in the sequence returned by the enumerator is an object[] and each array must be the same length and that length must be the number of arguments to your test case (annotated with the attribute [MemberData] and each element must have the same type as the corresponding method parameter. (Or maybe they can be convertible types, I don't know.)
(See release notes for xUnit.net March 2014 and the actual patch with example code.)

Creating anonymous object arrays is not the easiest way to construct the data so I used this pattern in my project.
First define some reusable, shared classes:
//http://stackoverflow.com/questions/22093843
public interface ITheoryDatum
{
object[] ToParameterArray();
}
public abstract class TheoryDatum : ITheoryDatum
{
public abstract object[] ToParameterArray();
public static ITheoryDatum Factory<TSystemUnderTest, TExpectedOutput>(TSystemUnderTest sut, TExpectedOutput expectedOutput, string description)
{
var datum= new TheoryDatum<TSystemUnderTest, TExpectedOutput>();
datum.SystemUnderTest = sut;
datum.Description = description;
datum.ExpectedOutput = expectedOutput;
return datum;
}
}
public class TheoryDatum<TSystemUnderTest, TExpectedOutput> : TheoryDatum
{
public TSystemUnderTest SystemUnderTest { get; set; }
public string Description { get; set; }
public TExpectedOutput ExpectedOutput { get; set; }
public override object[] ToParameterArray()
{
var output = new object[3];
output[0] = SystemUnderTest;
output[1] = ExpectedOutput;
output[2] = Description;
return output;
}
}
Now your individual test and member data is easier to write and cleaner...
public class IngredientTests : TestBase
{
[Theory]
[MemberData(nameof(IsValidData))]
public void IsValid(Ingredient ingredient, bool expectedResult, string testDescription)
{
Assert.True(ingredient.IsValid == expectedResult, testDescription);
}
public static IEnumerable<object[]> IsValidData
{
get
{
var food = new Food();
var quantity = new Quantity();
var data= new List<ITheoryDatum>();
data.Add(TheoryDatum.Factory(new Ingredient { Food = food } , false, "Quantity missing"));
data.Add(TheoryDatum.Factory(new Ingredient { Quantity = quantity } , false, "Food missing"));
data.Add(TheoryDatum.Factory(new Ingredient { Quantity = quantity, Food = food } , true, "Valid" ));
return data.ConvertAll(d => d.ToParameterArray());
}
}
}
The string Description property is to throw yourself a bone when one of your many test cases fail.

You can try this way:
public class TestClass {
bool isSaturday(DateTime dt)
{
string day = dt.DayOfWeek.ToString();
return (day == "Saturday");
}
[Theory]
[MemberData("IsSaturdayIndex", MemberType = typeof(TestCase))]
public void test(int i)
{
// parse test case
var input = TestCase.IsSaturdayTestCase[i];
DateTime dt = (DateTime)input[0];
bool expected = (bool)input[1];
// test
bool result = isSaturday(dt);
result.Should().Be(expected);
}
}
Create another class to hold the test data:
public class TestCase
{
public static readonly List<object[]> IsSaturdayTestCase = new List<object[]>
{
new object[]{new DateTime(2016,1,23),true},
new object[]{new DateTime(2016,1,24),false}
};
public static IEnumerable<object[]> IsSaturdayIndex
{
get
{
List<object[]> tmp = new List<object[]>();
for (int i = 0; i < IsSaturdayTestCase.Count; i++)
tmp.Add(new object[] { i });
return tmp;
}
}
}

For my needs I just wanted to run a series of 'test users' through some tests - but [ClassData] etc. seemed overkill for what I needed (because the list of items was localized to each test).
So I did the following, with an array inside the test - indexed from the outside:
[Theory]
[InlineData(0)]
[InlineData(1)]
[InlineData(2)]
[InlineData(3)]
public async Task Account_ExistingUser_CorrectPassword(int userIndex)
{
// DIFFERENT INPUT DATA (static fake users on class)
var user = new[]
{
EXISTING_USER_NO_MAPPING,
EXISTING_USER_MAPPING_TO_DIFFERENT_EXISTING_USER,
EXISTING_USER_MAPPING_TO_SAME_USER,
NEW_USER
} [userIndex];
var response = await Analyze(new CreateOrLoginMsgIn
{
Username = user.Username,
Password = user.Password
});
// expected result (using ExpectedObjects)
new CreateOrLoginResult
{
AccessGrantedTo = user.Username
}.ToExpectedObject().ShouldEqual(response);
}
This achieved my goal, while keeping the intent of the test clear. You just need to keep the indexes in sync but that's all.
Looks nice in the results, it's collapsable and you can rerun a specific instance if you get an error:

This is how I solved your problem, I had the same scenario. So inline with custom objects and a different number of objects on each run.
[Theory]
[ClassData(typeof(DeviceTelemetryTestData))]
public async Task ProcessDeviceTelemetries_TypicalDeserialization_NoErrorAsync(params DeviceTelemetry[] expected)
{
// Arrange
var timeStamp = DateTimeOffset.UtcNow;
mockInflux.Setup(x => x.ExportTelemetryToDb(It.IsAny<List<DeviceTelemetry>>())).ReturnsAsync("Success");
// Act
var actual = await MessageProcessingTelemetry.ProcessTelemetry(JsonConvert.SerializeObject(expected), mockInflux.Object);
// Assert
mockInflux.Verify(x => x.ExportTelemetryToDb(It.IsAny<List<DeviceTelemetry>>()), Times.Once);
Assert.Equal("Success", actual);
}
So this is my unit test, notice the params parameter. This allow to send a different number of object. And now my DeviceTelemetryTestData class :
public class DeviceTelemetryTestData : IEnumerable<object[]>
{
public IEnumerator<object[]> GetEnumerator()
{
yield return new object[] { new DeviceTelemetry { DeviceId = "asd" }, new DeviceTelemetry { DeviceId = "qwe" } };
yield return new object[] { new DeviceTelemetry { DeviceId = "asd" }, new DeviceTelemetry { DeviceId = "qwe" } };
yield return new object[] { new DeviceTelemetry { DeviceId = "asd" }, new DeviceTelemetry { DeviceId = "qwe" } };
yield return new object[] { new DeviceTelemetry { DeviceId = "asd" }, new DeviceTelemetry { DeviceId = "qwe" } };
}
IEnumerator IEnumerable.GetEnumerator() => GetEnumerator();
}
Hope it helps !

Despite this has already been answered I just want to add an improvement here.
The restriction of passing objects in InlineData attribute is not a limitiation of xUnit itself but C# attributes.
See this compiler error: Compiler Error CS0182

xUnit.Sdk provides you with DataAttribute class that you could inherit and override its GetData method and use it to pass whatever you feel like you want..
I usually use it alongside DataTestBuilders pattern and build something like that..
public class ValidComplexObjectDataSource : DataAttribute
{
public override IEnumerable<object[]> GetData(MethodInfo testMethod)
{
yield return new object[] {
ComplexObjectBuilder
.BasicComplexObject()
.Build()
};
yield return new object[] {
ComplexObjectBuilder
.BasicComplexObject()
.WithoutSomeAttribute()
.Build()
};
// ... list all test cases you want to pass to your method
}
}
This ComplexObjectBuilder could be whatever your object is, highly recommend checking builder pattern
[Theory]
[Trait("Validation", "CreateXYZCommand")]
[ValidComplexObjectDataSource]
public void CreateXYZCommandValidator_WithValidInput_ShouldPassAllValidations(CreateComplexObjectInput createComplexObjectInput)
{
var command = new CreateXYZCommand(createComplexObjectInput);
var result = _validator.TestValidate(command);
result.ShouldNotHaveAnyValidationErrors();
}
I only demonstrated it with a single object, you have an array of objects you can yield.
yield return new object[] {
ComplexObject_1,
ComplexObject_2,
string_attribute,
int_attribute
};
and have these as arguments to your test cases.

You can utilize TheoryData for complex types like classes.
[Theory, MemberData(nameof(CustomClassTests))]
public async Task myTestName(MyCustomClass customClassTestData) { ... }
public record MyCustomClass { ... }
public static TheoryData<MyCustomClass> CustomClassTests {
get {
return new() {
new MyCustomClass{ ... },
new MyCustomClass{ ... },
...
};
}
}

I guess you mistaken here. What xUnit Theory attribute actually means: You want to test this function by sending special/random values as parameters that this function-under-test receives. That means that what you define as the next attribute, such as: InlineData, PropertyData, ClassData, etc.. will be the source for those parameters. That means that you should construct the source object to provide those parameters. In your case I guess you should use ClassData object as source. Also - please note that ClassData inherits from: IEnumerable<> - that means each time another set of generated parameters will be used as incoming parameters for function-under-test until IEnumerable<> produces values.
Example here: Tom DuPont .NET
Example may be incorrect - I didn't use xUnit for a long time

Where's the best place for validation... constructor or leave to client to call?

This is really a generic (and probably a more subjective too) question. I have some classes where I use an interface to define a standard approach to validating the object state. When I did this, I got to scratching my head... is it best to
1.) allow the constructor (or initializing method) to silently filter out the errant information automatically or...
2.) allow the client to instantiate the object however and let the client also call the interface's IsValid property or Validate() method before moving forward?
Basically one approach is silent but could be misleading in that the client may not be aware that certain pieces of information were filtered away due to it not meeting the validation criteria. The other approach then would be more straight forward, but also adds a step or two? What's typical here?
Okay, after a long day of trying to keep up with some other things, I finally did come up with an example. Please for me for it as it's not ideal and by no means something wonderful, but hopefully should serve well enough to get the point across. My current project is just too complicated to put something simple out for this, so I made something up... and trust me... totally made up.
Alright, the objects in the example are this:
Client: representing client-side code (Console App btw)
IValidationInfo: This is the actual interface I'm using in my current project. It allows me to create a validation framework for the "back-end" objects not necessarily intended for the Client to use since the business logic could be complicated enough. This also allowed me to separate validation code and call as-needed for the business logic.
OrderManager: This is an object the client-side code can use to manage their orders. It's client-friendly so-to-speak.
OrderSpecification: This is an object the client-side code can use to request an order. But if the business logic doesn't work out, an exception can be raised (or if necessary the order not added and exceptions ignored...) In my real-world example I actually have an object that's not quite so black-and-white as to which side of this fence it goes... thus my original question when I realized I could push validation request (calling IsValid or Validate()) to the cilent.
CustomerDescription: represents customers to which I've classified (pretending to have been read from a DB.
Product: Represents a particular product which is classified also.
OrderDescription: Represents the official order request.The business rule is that the Customer cannot order anything to which they've not been classified (I know.. that's not very real-world, but it gave me something to work with...)
Ok... I just realized I can't attach a file here, so here's the code. I apologize for it's lengthy appearance. That was the best I could do to create a client-friendly front-end and business logic back-end using my Validation interface:
public class Client
{
static OrderManager orderMgr = new OrderManager();
static void Main(string[] args)
{
//Request a new order
//Note: Only the OrderManager and OrderSpecification are used by the Client as to keep the
// Client from having to know and understand the framework beyond that point.
OrderSpecification orderSpec = new OrderSpecification("Customer1", new Product(IndustryCategory.FoodServices, "Vending Items"));
orderMgr.SubmitOrderRequest(orderSpec);
Console.WriteLine("The OrderManager has {0} items for {1} customers.", orderMgr.ProductCount, orderMgr.CustomerCount);
//Now add a second item proving that the business logic to add for an existing customer works
Console.WriteLine("Adding another valid item for the same customer.");
orderSpec = new OrderSpecification("Customer1", new Product(IndustryCategory.FoodServices, "Sodas"));
orderMgr.SubmitOrderRequest(orderSpec);
Console.WriteLine("The OrderManager now has {0} items for {1} customers.", orderMgr.ProductCount, orderMgr.CustomerCount);
Console.WriteLine("Adding a new valid order for a new customer.");
orderSpec = new OrderSpecification("Customer2", new Product(IndustryCategory.Residential, "Magazines"));
orderMgr.SubmitOrderRequest(orderSpec);
Console.WriteLine("The OrderManager now has {0} items for {1} customers.", orderMgr.ProductCount, orderMgr.CustomerCount);
Console.WriteLine("Adding a invalid one will not work because the customer is not set up to receive these kinds of items. Should get an exception with message...");
try
{
orderSpec = new OrderSpecification("Customer3", new Product(IndustryCategory.Residential, "Magazines"));
orderMgr.SubmitOrderRequest(orderSpec);
}
catch (Exception ex)
{
Console.WriteLine(ex.Message);
}
Console.ReadLine();
}
}
public interface IValidationInfo
{
string[] ValidationItems { get; }
bool IsValid { get; }
void Validate();
List<string> GetValidationErrors();
string GetValidationError(string itemName);
}
public class OrderManager
{
private List<OrderDescription> _orders = new List<OrderDescription>();
public List<OrderDescription> Orders
{
get { return new List<OrderDescription>(_orders); }
private set { _orders = value; }
}
public int ProductCount
{
get
{
int itemCount = 0;
this.Orders.ForEach(o => itemCount += o.Products.Count);
return itemCount;
}
}
public int CustomerCount
{
get
{
//since there's only one customer per order, just return the number of orders
return this.Orders.Count;
}
}
public void SubmitOrderRequest(OrderSpecification orderSpec)
{
if (orderSpec.IsValid)
{
List<OrderDescription> orders = this.Orders;
//Since the particular customer may already have an order, we might as well add to an existing
OrderDescription existingOrder = orders.FirstOrDefault(o => string.Compare(orderSpec.Order.Customer.Name, o.Customer.Name, true) == 0) as OrderDescription;
if (existingOrder != null)
{
List<Product> existingProducts = orderSpec.Order.Products;
orderSpec.Order.Products.ForEach(p => existingOrder.AddProduct(p));
}
else
{
orders.Add(orderSpec.Order);
}
this.Orders = orders;
}
else
orderSpec.Validate(); //Let the OrderSpecification pass the business logic validation down the chain
}
}
public enum IndustryCategory
{
Residential,
Textile,
FoodServices,
Something
}
public class OrderSpecification : IValidationInfo
{
public OrderDescription Order { get; private set; }
public OrderSpecification(string customerName, Product product)
{
//Should use a method in the class to search and retrieve Customer... pretending here
CustomerDescription customer = null;
switch (customerName)
{
case "Customer1":
customer = new CustomerDescription() { Name = customerName, Category = IndustryCategory.FoodServices };
break;
case "Customer2":
customer = new CustomerDescription() { Name = customerName, Category = IndustryCategory.Residential };
break;
case "Customer3":
customer = new CustomerDescription() { Name = customerName, Category = IndustryCategory.Textile };
break;
}
//Create an OrderDescription to potentially represent the order... valid or not since this is
//a specification being used to request the order
this.Order = new OrderDescription(new List<Product>() { product }, customer);
}
#region IValidationInfo Members
private readonly string[] _validationItems =
{
"OrderDescription"
};
public string[] ValidationItems
{
get { return _validationItems; }
}
public bool IsValid
{
get
{
List<string> validationErrors = GetValidationErrors();
if (validationErrors != null && validationErrors.Count > 0)
return false;
else
return true;
}
}
public void Validate()
{
List<string> errorMessages = GetValidationErrors();
if (errorMessages != null && errorMessages.Count > 0)
{
StringBuilder errorMessageReported = new StringBuilder();
errorMessages.ForEach(em => errorMessageReported.AppendLine(em));
throw new Exception(errorMessageReported.ToString());
}
}
public List<string> GetValidationErrors()
{
List<string> errorMessages = new List<string>();
foreach (string item in this.ValidationItems)
{
string errorMessage = GetValidationError(item);
if (!string.IsNullOrEmpty(errorMessage))
errorMessages.Add(errorMessage);
}
return errorMessages;
}
public string GetValidationError(string itemName)
{
switch (itemName)
{
case "OrderDescription":
return ValidateOrderDescription();
default:
return "Invalid item name.";
}
}
#endregion
private string ValidateOrderDescription()
{
string errorMessage = string.Empty;
if (this.Order == null)
errorMessage = "Order was not instantiated.";
else
{
if (!this.Order.IsValid)
{
List<string> orderErrors = this.Order.GetValidationErrors();
orderErrors.ForEach(ce => errorMessage += "\n" + ce);
}
}
return errorMessage;
}
}
public class CustomerDescription : IValidationInfo
{
public string Name { get; set; }
public string Street { get; set; }
public string City { get; set; }
public string State { get; set; }
public int ZipCode { get; set; }
public IndustryCategory Category { get; set; }
#region IValidationInfo Members
private readonly string[] _validationItems =
{
"Name",
"Street",
"City",
"State",
"ZipCode",
"Category"
};
public string[] ValidationItems
{
get { return _validationItems; }
}
public bool IsValid
{
get
{
List<string> validationErrors = GetValidationErrors();
if (validationErrors != null && validationErrors.Count > 0)
return false;
else
return true;
}
}
public void Validate()
{
List<string> errorMessages = GetValidationErrors();
if (errorMessages != null && errorMessages.Count > 0)
{
StringBuilder errorMessageReported = new StringBuilder();
errorMessages.ForEach(em => errorMessageReported.AppendLine(em));
throw new Exception(errorMessageReported.ToString());
}
}
public List<string> GetValidationErrors()
{
List<string> errorMessages = new List<string>();
foreach (string item in this.ValidationItems)
{
string errorMessage = GetValidationError(item);
if (!string.IsNullOrEmpty(errorMessage))
errorMessages.Add(errorMessage);
}
return errorMessages;
}
public string GetValidationError(string itemName)
{
//Validation methods should be called here... pretending nothings wrong for sake of discussion & simplicity
switch (itemName)
{
case "Name":
return string.Empty;
case "Street":
return string.Empty;
case "City":
return string.Empty;
case "State":
return string.Empty;
case "ZipCode":
return string.Empty;
case "Category":
return string.Empty;
default:
return "Invalid item name.";
}
}
#endregion
}
public class Product
{
public IndustryCategory Category { get; private set; }
public string Description { get; private set; }
public Product(IndustryCategory category, string description)
{
this.Category = category;
this.Description = description;
}
}
public class OrderDescription : IValidationInfo
{
public CustomerDescription Customer { get; private set; }
private List<Product> _products = new List<Product>();
public List<Product> Products
{
get { return new List<Product>(_products); }
private set { _products = value; }
}
public OrderDescription(List<Product> products, CustomerDescription customer)
{
this.Products = products;
this.Customer = customer;
}
public void PlaceOrder()
{
//If order valid, place
if (this.IsValid)
{
//Do stuff to place order
}
else
Validate(); //cause the exceptions to be raised with the validate because business rules were broken
}
public void AddProduct(Product product)
{
List<Product> productsToEvaluate = this.Products;
//some special read, validation, quantity check, pre-existing, etc here
// doing other stuff...
productsToEvaluate.Add(product);
this.Products = productsToEvaluate;
}
#region IValidationInfo Members
private readonly string[] _validationItems =
{
"Customer",
"Products"
};
public string[] ValidationItems
{
get { return _validationItems; }
}
public bool IsValid
{
get
{
List<string> validationErrors = GetValidationErrors();
if (validationErrors != null && validationErrors.Count > 0)
return false;
else
return true;
}
}
public void Validate()
{
List<string> errorMessages = GetValidationErrors();
if (errorMessages != null && errorMessages.Count > 0)
{
StringBuilder errorMessageReported = new StringBuilder();
errorMessages.ForEach(em => errorMessageReported.AppendLine(em));
throw new Exception(errorMessageReported.ToString());
}
}
public List<string> GetValidationErrors()
{
List<string> errorMessages = new List<string>();
foreach (string item in this.ValidationItems)
{
string errorMessage = GetValidationError(item);
if (!string.IsNullOrEmpty(errorMessage))
errorMessages.Add(errorMessage);
}
return errorMessages;
}
public string GetValidationError(string itemName)
{
switch (itemName)
{
case "Customer":
return ValidateCustomer();
case "Products":
return ValidateProducts();
default:
return "Invalid item name.";
}
}
#endregion
#region Validation Methods
private string ValidateCustomer()
{
string errorMessage = string.Empty;
if (this.Customer == null)
errorMessage = "CustomerDescription is missing a valid value.";
else
{
if (!this.Customer.IsValid)
{
List<string> customerErrors = this.Customer.GetValidationErrors();
customerErrors.ForEach(ce => errorMessage += "\n" + ce);
}
}
return errorMessage;
}
private string ValidateProducts()
{
string errorMessage = string.Empty;
if (this.Products == null || this.Products.Count <= 0)
errorMessage = "Invalid Order. Missing Products.";
else
{
foreach (Product product in this.Products)
{
if (product.Category != Customer.Category)
{
errorMessage += string.Format("\nThe Product, {0}, category does not match the required Customer category for {1}", product.Description, Customer.Name);
}
}
}
return errorMessage;
}
#endregion
}

Any reason you wouldn't want the constructor to noisily throw an exception if the information is valid? It's best to avoid ever creating an object in an invalid state, in my experience.

It's completely depends on the client. There's a trade-off as you already mentioned. By default approach number 1 is my favorite. Creating smart classes with good encapsulation and hiding details from client. The level of smartness depends who is going to use the object. If client is business aware you can reveal details according to the level of this awareness. This is a dichotomy and should not be treated as black or white.

Well if I correctly understood, there are basically two question - whether you should fail right away or later and whether you should omit/assume certain information.
1) I always prefer failing as soon as possible - good example is failing at compile time vs failing at run time - you always want to fail at compile time. So if something is wrong with the state of some object, as Jon said - throw exception right away as loudly as you can and deal with it - do not introduce additional complexity down the road as you'll be heading for if/elseif/elseif/elseif/else mumbo jumbo.
2) When it comes to user input, if you are in position to simply filter out errors automatically - just do it. For example, I almost never ask users for country - if I really need it, I automatically detect it from IP and display it in the form. It's way easier if user just needs to confirm/change the data - and I don't need to deal with null situation.
Now, in case we are talking about the data generated by code during some processing - for me situation is drastically different - I always want to know an much as possible (for easier debugging down the road) and ideally you never should destroy any piece of information.
To wrap up, in your case I would recommend that you keep IsValid as simple yes/no (not yes/no/maybe/kindaok/etc). If you can fix some problems automatically - do it, but consider that they keep object in IsValid yes. For everything else, you throw exception and go to IsValid=no.