csharp-nullable-reference-types

name	csharp-nullable-reference-types
description	Guidelines for introducing and using nullable reference types (NRT) and System.Diagnostics.CodeAnalysis nullable attributes in C# / .NET codebases. Covers the nullability model, flow analysis, the null-forgiving operator, API design rules, the full attribute catalog (AllowNull, DisallowNull, MaybeNull, NotNull, NotNullWhen, MaybeNullWhen, NotNullIfNotNull, MemberNotNull, MemberNotNullWhen, DoesNotReturn, DoesNotReturnIf), the C# 14 field keyword, incremental migration of legacy codebases, and a code-generation checklist.
version	1.0.0
tags	["csharp","nullable","nrt","code-quality","api-design"]

C# Nullable Reference Types

When to Use

• Introducing nullable reference types (NRT) into a codebase that has not yet adopted them
• Writing or refactoring C# code that uses T? / nullable annotations
• Annotating APIs with System.Diagnostics.CodeAnalysis nullable attributes
• Designing public/internal APIs where nullability contracts matter
• Wrapping unannotated or legacy APIs so downstream callers still benefit from NRT
• Reviewing code for correct null-state analysis, guard helpers, and the field keyword

Core Goals

• Prevent NullReferenceException at runtime by making null intent explicit in signatures.
• Express contracts the type system cannot represent directly using the official nullable attributes.
• Adopt NRT incrementally in legacy codebases without a big-bang rewrite.

Core Nullability Model

Non-nullable vs nullable

• string — non-nullable reference. The compiler assumes instances are never null; assigning null or a maybe-null value produces a warning.
• string? — nullable reference. The variable may be null; the compiler requires a null check before dereference.

string name = "Alice";
name = null;          // Warning: assigning null to non-nullable.

string? nickname = null;
Console.WriteLine(nickname.Length); // Warning: possible null dereference.

Null-state analysis (flow)

The compiler tracks whether a reference is definitely non-null or maybe null. Null checks and assignments update this state.

string? message = GetMessageOrNull();

if (message != null)
{
    // message is definitely non-null in this block.
    Console.WriteLine(message.Length);
}

// Outside the if, message is maybe null again.

Introduce explicit null checks (if (x != null), is not null, pattern matching) before dereferencing nullable values. Narrow nullability early and keep the non-null state alive. Null-conditional assignment (C# 14) lets you write customer?.Order = CreateOrder(); — the right side is evaluated only when the receiver is non-null.

Null-forgiving operator (!)

x! tells the compiler "treat x as non-null here." It affects analysis only, not runtime behavior.

• Use ! only when a real invariant guarantees non-null and the compiler cannot see it.
• Do not use ! as a general fix for warnings. Prefer refactoring control flow, adding attributes, or proper member initialization.

_customer = LoadCustomerFromOrm()!; // ORM guarantees this is not null in valid state.

Reorganize code before suppressing warnings

A successful guard clause or pattern match already creates a null-safe region in the current scope. Before adding !, make nullable values cross a checked boundary once and keep the remaining code non-nullable:

• narrow early with a guard clause or pattern match;
• copy nullable fields or properties to a local before checking, so repeated reads cannot change underneath the analysis;
• when a method has complex control flow, optionally move the non-null path into a local function or private method with non-nullable parameters;
• keep nullable handling at the boundary instead of spreading T?, repeated checks, or ! through the implementation.

public void Process(Order? order)
{
    if (order?.Customer is not { } customer)
    {
        return;
    }

    // The pattern match already proved that customer is non-null here.
    Console.WriteLine(customer.Name);
}

Do not extract a function solely to satisfy nullable analysis. Use an explicit non-nullable function boundary when it also simplifies a large or branching implementation. Use ! only when a real external invariant cannot be represented through control flow, signatures, or nullable-analysis attributes.

Project Configuration

Enable NRT for new code:

<PropertyGroup>
  <Nullable>enable</Nullable>
</PropertyGroup>

For legacy codebases, enable incrementally with file-level directives (#nullable enable, #nullable disable, #nullable enable warnings, #nullable enable annotations). Treat CS86xx nullable warnings as important; consider TreatWarningsAsErrors or treating nullable warnings as errors in new projects.

See nrt-migration-playbook-reference.md for the full incremental-adoption strategy, #nullable directive reference, legacy interop, and known static-analysis limitations (arrays, default(struct)).

API Design Rules (Signatures)

These rules apply to public and internal APIs and to models.

Parameters — if null is not allowed, use a non-nullable type and add a runtime guard for public APIs:

public void SendEmail(string recipient)
{
    ArgumentNullException.ThrowIfNull(recipient);
    // Implementation
}

If null is allowed and meaningful, use T?, document how null is interpreted, and implement correct null behavior.

Return types — Customer when the method never returns null; Customer? when it can legitimately return null (callers must check, and the compiler enforces it).

public Customer GetRequiredCustomer(Guid id);  // Throws on failure.
public Customer? TryGetCustomer(Guid id);       // Returns null on failure.

Properties and fields — follow the same rules as parameters and return types. Non-nullable members must be initialized in constructors, via required properties with object initializers, via field-backed lazy properties, or via helpers annotated with [MemberNotNull].

public class Order
{
    public required string Id { get; init; }
    public required Customer Customer { get; init; }
    public string? Comment { get; init; } // Optional.
}

When contracts depend on input/output behavior, conditional behavior, or member initialization, apply the nullable attributes described in nullable-attributes-reference.md.

Public API compatibility for libraries

Treat nullable annotations and nullable-analysis attributes as part of a shipped API contract. T and T? have the same CLR type, so annotation-only changes are generally binary compatible, but they can be source breaking by introducing warnings for nullable-enabled consumers. Those warnings often become build failures when consumers treat warnings as errors.

Review public nullability changes before release, especially:

• weakening an output from T to T? or adding [MaybeNull];
• tightening an input from T? to T or adding [DisallowNull];
• changing generic constraints such as class?, class, or notnull;
• changing annotations or attributes on virtual members, interfaces, delegates, and implementations, where mismatches produce compiler warnings.

Adding annotations to a previously nullable-oblivious API can create the same source-compatibility problems. Compare the annotated surface with the last released version, test a nullable-enabled consumer, and document or version intentional source-breaking changes according to the library's compatibility policy.

The field Keyword (C# 14 / .NET 10)

The field contextual keyword lets you write a property accessor body without declaring an explicit backing field. This is a primary NRT scenario (lazily-initialized properties) and the compiler performs a special null-resilience analysis so you do not get nuisance CS8618 in constructors:

public class C
{
    public C() { } // No warning: the getter is null-resilient.
    string Prop => field ??= GetPropValue();
}

See nullable-attributes-reference.md for the full field nullability rules (null-resilient vs non-resilient getters, the [field: AllowNull, MaybeNull] escape hatch, and setter/constructor analysis).

Reference Files

• nullable-attributes-reference.md: The complete System.Diagnostics.CodeAnalysis attribute catalog — preconditions (AllowNull, DisallowNull), postconditions (MaybeNull, NotNull), conditional postconditions (NotNullWhen, MaybeNullWhen, NotNullIfNotNull), helper methods (MemberNotNull, MemberNotNullWhen), unreachable-code helpers (DoesNotReturn, DoesNotReturnIf), and the field keyword nullability rules. Each with intent, pattern, and agent rules.
• nrt-migration-playbook-reference.md: Incremental adoption strategy, #nullable directive reference, legacy/unannotated API interop, polyfilling nullable attributes for older target frameworks (with tradeoffs and a confirm-before-adding decision process), known static-analysis limitations (arrays of non-nullable references, default(struct) with reference fields), warning handling, and the full generation checklist.

Generation Checklist (Summary)

1. Project — <Nullable>enable</Nullable> present; disable only around unavoidable legacy code.
2. Types — non-nullable for required params/returns/properties; T? only when null is valid and expected.
3. Initialization — constructors, required + object initializers, field-backed lazy getters, or [MemberNotNull] helpers. Avoid null! except as a documented escape hatch.
4. Null checks — explicit guards at public boundaries; narrow with control flow, and extract a non-nullable helper only when it improves complex code; ! only with a clear invariant.
5. Attributes — apply to express contracts the type system cannot express (see reference file).
6. Interop — trust BCL/annotated libraries; add your own guards and attributes when wrapping unannotated APIs.
7. Warnings — never ignore; fix design or add attributes rather than suppressing with ! or #pragma.
8. Compatibility — for released libraries, review public nullability changes as potential source breaks and test nullable-enabled consumers.

References

• Nullable reference types (overview)
• Attributes for null-state static analysis
• Nullable migration strategies
• Breaking change: Nullable reference type annotation changes
• Tutorial: Express your design intent with nullable and non-nullable reference types
• What's new in C# 14 (extension members, field keyword GA, null-conditional assignment)
• The field contextual keyword (feature spec)