C# Best Practices

Owned by Stephen Fuqua
Last updated: Oct 09, 2024
6 min read

Introduction

This page is a companion to the C# Coding Standards, sharing some of the best practices that are generally employed in Ed-Fi Alliance supported software. While these are not fixed and absolute rules, code reviewers can and should push back if they feel that a deviation from best practice hinders any of the -ilities, such as readability, maintainability, and so forth.

Naming Conventions

What's in a name? Everything. Use clear and descriptive names for classes, methods, fields and variables to reduce maintenance costs.

  1. Make sure that after refactoring code that the names are still appropriate.

    A Method Name That Does Not Reflect Its Behavior
    // The method name includes the text "Not" and returns a boolean value.
// This could create scenarios where callers are performing double
// negatives, which are hard to read (e.g., if (!IsNotDuplicate(…) { )
private static bool IsNotDuplicate(Dictionary<Tuple<string, string>, int> idsByValue,
      Tuple<string, string> key,
      // Since the main behavioral aspect of this method is to modify
      // the StringBuilder, it would make more sense to pass it first
      // in the list of arguments.
      StringBuilder duplicatesMessageBuilder)
{
    if (idsByValue.ContainsKey(key))
    {
        // The behavior of the method actually appends to the provided
        // StringBuilder, but this is not reflected in the method name
        // at all and is only discoverable by a developer inspecting
        // the code.
        duplicatesMessageBuilder.AppendFormat("\tDescriptor Type: {0}, Code Value: {1}\r\n", key.Item1, key.Item2);
        return false;
    }
      
    return true;
}
    Improved Name and Signature
    private static bool TryAppendDuplicateValueMessage(StringBuilder duplicatesMessageBuilder,
      Tuple<string, string> key,
      Dictionary<Tuple<string, string>, int> idsByValue)
{
    …
}


  2. Name dictionaries using a name format of {ValueName}By{KeyName}.

    • Precisely describe the keys and values (e.g., SchoolNameById would indicate that you can obtain a school's name by its identifier).

    • If each entry’s value is a single item, the name should be singularized (e.g., StudentById).

    • If each entry’s value is a collection, the name should be pluralized (e.g., StudentsBySectionId).

  3. Use the following guidelines when defining generic types:

    • For types with a single generic type, prefer the use of T.

      public interface IList<T>

    • For types with multiple generic types, use a capital "T" followed by an optional secondary name for additional clarity. In all cases, start the type with a capital letter.

      public interface IService<TRequest, TResponse>

  4. Name custom attribute and exception classes using suffixes of Attribute and Exception, respectively.

  5. Name enumerations in the singular (e.g., FileMode) unless the enumeration is representing a bit flag value (e.g., FileAttributes) in which case the [Flags] attribute should also be applied.

Exception Handling

  1. Express exception messages as grammatically correct sentences.

    1. When appropriate, format the message using string interpolation.

      throw new Exception($"There is no data for resource '{resourceName}'.");
 
throw new Exception($"Type '{typeName}' not found in assembly '{assemblyName}'.");

    2. For exceptions that are targeted for a developer audience (such as configuration errors that shouldn't happen in production), you may include a helpful suggestion or question to help resolve the error message.

  2. Before throwing Exception, consider whether a custom exception or one of the following .NET exceptions would provide semantic value:
    1. ArgumentException.
    2. ArgumentOutOfRangeException.
    3. ArgumentNullException.
    4. IndexOutOfRangeException.
    5. InvalidOperationException. Per the Microsoft .NET documentation, "Typically, it is thrown when the state of an object cannot support the method call" such as when an instance variable is null and therefore you can't reference a property on that variable.
    6. NotImplementedException.
  3. Do not reuse .NET exceptions outside of their semantic context (e.g., ObjectNotFoundException is defined in multiple places, and AuthorizationFailedException and SecurityAccessDeniedException have specific contexts in which they're used).

Using Static Structures

  1. Call static members by using the class name: ClassName.StaticMember. This practice makes code more readable by making static access clear.
  2. Do not qualify a static member defined in a base class with the name of a derived class. While that code compiles, the code readability is misleading, and the code may break in the future if you add a static member with the same name to the derived class.
  3. Avoid taking dependencies on static .NET Framework classes whose results are not idempotent or require access to infrastructure (e.g., System.DateTime, System.IO.File and System.Configuration.ConfigurationManager, but not System.IO.Path). Instead, define an interface using a naming convention of I{StaticClassName} and build a minimal (but expanding, as necessary) facade with an implementation named as {StaticClassName}Wrapper. This will enable the behavior to be controlled for testing purposes.

    1. For System.IO.File, this would look like this: 

      public interface IFile
{
    bool Exists(string path);
}
 
public class FileWrapper : IFile
{
    public bool Exists(string path)
    {
        return File.Exists(path);
    }
}
    2. The functionality of the ConfigurationManager has been abstracted using a different approach due to legacy code reuse. Review the IConfigValueProvider, IConfigSectionProvider and IConfigConnectionStringsProvider interfaces and associated implementations.

Design Guidance

SOLID Principles

Apply SOLID principles everywhere. Chapters 8–12 of Agile Principles, Patterns, and Practices by Robert C. Martin are highly recommended reading.

  1. Single Responsibility Principle. There should only be one reason for a class to change. If you're having to change an existing class, ask yourself, "Is there an abstraction (interface) missing here?"
  2. Open/Closed Principle. The system should be open for extension, but closed to modification. You should be able to change the behavior of the system by adding a new implementation of an existing interface.
  3. Liskov Substitution Principle. A derived class should be substitutable for its base class, and should not change the fundamental nature of the abstraction.
  4. Interface Segregation Principle. An interface should be highly focused. Many interfaces in the Ed-Fi ODS / API have just one or two methods. If you implement some methods of an interface using new NotImplementedException(), the interface probably needs to be decomposed.
  5. Dependency Inversion Principle. External dependencies are injected into classes, preferably via their constructors. The Ed-Fi ODS / API uses Castle Windsor as its Inversion of Control container.

Class Design

  1. Avoid deep class hierarchies (more than 2 total levels). Prefer composition over inheritance.
  2. Keep logic in constructors simple, primarily focused on capturing the injected dependencies to the field values.
  3. Prefer constructor injection to property injection.
    1. A property injector may be appropriate for injecting a fake during unit testing, where changing the constructor and setting up the IoC container might not be worthwhile. Use sparingly.
  4. Do not use properties to modify class state outside of the property being set. If other state must be changed, prefer the use of a read-only property in combination with a well-named method.
  5. Prefer the use of return types that are abstractions rather than concrete types from public members. In other words, prefer IList<T> to List<T> and IDictionary<TKey, TValue> over Dictionary<TKey, TValue>.

  6. Make sure the return types from public members match the intended semantics. In other words, prefer IEnumerable<T> over IList<T> and IReadOnlyDictionary<TKey, TValue> over IDictionary<TKey, TValue>, unless modifications by the caller are intended.

    // Good
public interface IStudentDataProvider
{
    IEnumerable<Student> GetAll();
    IReadOnlyDictionary<string, Student> GetStudentByNameDictionary();
}
 
// Potentially problematic for maintenance, due to semantics of returned values
public interface IStudentDataProvider
{
    IList<Student> GetAll();
    IDictionary<string, Student> GetStudentByNameDictionary();
}
  7. Do not use the new inheritance qualifier. Instead, reevaluate the design of the class.

  8. Do not make members of a class public for the sake of unit testing. Find a better way to test the functionality.

  9. When providing an implementation of an interface that does nothing, use the Null Object Pattern.

    public interface IETagProvider
{
    string GetETag(object value);
 
    DateTime GetDateTime(string etag);
}

public class NullETagProvider : IETagProvider
{
    public string GetETag(object value)
    {
        return string.Empty;
    }
 
    public DateTime GetDateTime(string etag)
    {
        return default(DateTime);
    }
}

  10. The Provider Pattern (also known as the Strategy Pattern) is used heavily in the code base.

    1. If the contract being defined basically allows the caller to get something, the Provider suffix is preferred.

    2. If the contract allows the caller to get and set something, consider splitting the methods into Reader and Writer interfaces (which could still be implemented by the same class).  The intent here is to provide more clearly stated intent when a caller intends to modify the underlying data exposed by the provider.

      // Instead of this...
public interface ICacheProvider
{
    void RemoveCachedObjects(string keyContains);
    void RemoveCachedObject(string keyName);
    bool TryGetCachedObject(string key, out object value);
    void SetCachedObject(string keyName, object obj);
    void Insert(string key, object value, System.Web.Caching.CacheDependency dependencies, DateTime absoluteExpiration, TimeSpan slidingExpiration);
    void Insert(string key, object value, DateTime absoluteExpiration, TimeSpan slidingExpiration);
}

// Consider decomposing the interfaces ...
public interface ICacheReader
{
    bool TryGetCachedObject(string key, out object value);
}

public interface ICacheWriter
{
    void RemoveCachedObjects(string keyContains);
    void RemoveCachedObject(string keyName);
    void SetCachedObject(string keyName, object obj);
    void Insert(string key, object value, System.Web.Caching.CacheDependency dependencies, DateTime absoluteExpiration, TimeSpan slidingExpiration);
    void Insert(string key, object value, DateTime absoluteExpiration, TimeSpan slidingExpiration);
}

    3. If the contract being defined performs some sort of action, the suffix may be removed entirely.

      public interface IObjectValidator
{
    ICollection<ValidationResult> ValidateObject(object @object);
}
  11. Use creational patterns for creating objects that involve significant logic.
    1. When the logic required to create something doesn't fit well into a class constructor, use a creational pattern like the Factory Pattern.
    2. When an item can built in one step, use "Create" semantics (see the Factory Method Pattern).
    3. When an item is built up over multiple steps, using "Build" semantics (see the Builder Pattern).

  12. When a class has the concept of an empty instance, implement a static read-only Empty property.

    public class UserLookupResult
{
    public static readonly UserLookupResult Empty = new UserLookupResult();
    …
}

Portions of this document are based on the Microsoft C# Coding Conventions, which have been reproduced and modified under the terms of the Creative Commons Attribution 4.0 International license.

Contents