lookbusy1344.RecordValueAnalyser 0.9.6

Value-Semantics Analyser for C# Records

This project is a C# Roslyn code analyser to check records for correct value semantics.

https://github.com/lookbusy1344/RecordValueAnalyser

Why?

Records are a feature in modern C#. They are intended to be used for immutable data with value semantics. This means that two instances of the same record type should be considered equal if all their members are equal. This is the same as the behaviour of struct and tuple types.

Internally records are regular classes (or structs), but they have a synthesized Equals method that compares all their members. Without this Equals method, different instances would never be equal:

class TestClass
{
    public int A { get; set; }
    public string B { get; set; }
}

var a = new TestClass { A = 1, B = "Hello" };
var b = new TestClass { A = 1, B = "Hello" };

In this case a never equals b, because they are different instances. But with records:

record TestRecord(int A, string B);

var a = new TestRecord(1, "Hello");
var b = new TestRecord(1, "Hello");

Now a and b do equal, because the compiler has synthesized an Equals method that compares the members. This is more natural behaviour.

But...

There is a gotcha. If one of your record members lacks value-semantics itself, the synthesized Equals method will not work correctly. For example:

record TestRecord(int A, string B, IReadOnlyList<int> C);

var a = new TestRecord(1, "Hello", new[] { 1, 2, 3 });
var b = new TestRecord(1, "Hello", new[] { 1, 2, 3 });

The C member is an array, and these lack value semantics. Therefore a and b do not equal any more!

What the analyser does

It scans your records, and reports any members that don't have value semantics. It also scans into any nested stucts and tuples. In the above example, it would cause a warning on the C field:

record TestRecord(int A, string B, IReadOnlyList<int> C);
                                   ~~~~~~~~~~~~~~~~~~~~  JSV01: member lacks value semantics

It was built for C# 11 and .NET 7. It checks record class and record struct types for the following:

• if the record has a Equals(T) method, it is ok and no more checks are performed
• Otherwise all members are checked for:
  • the member is a primitive type, enum or string (these are ok)
  • it is a object or dynamic (these are never ok)
  • it has Equals(T) or Equals(object) method overriden directly in the type (these are ok)
  • it is a record (these will be checked elsewhere, so are assumed ok here)
  • it is a class (without Equals method, these are not ok)
  • it is a tuple or struct (without Equals method, their members are checked recursively)

It works in Visual Studio 2022 and Visual Studio Code, and also on the command line.

Warnings

It can produce 4 warnings:

• JSV01 - a record member lacks value semantics eg record Test(IList<int> Fail)
• JSV02 - a record propery fails
• JSV03 - a record field fails
• JSV04 - a nested tuple or struct fails

Examples

These are taken from the test project. Members expected to pass are named Pass, and those expected to fail are named Fail.

// these check members, called parameters in Roslyn

public record class A(F FooFail, G BarFail, string SPass, StructA SaFail);

public record class B(int IPass, IReadOnlyList<int> JFail, int? NullableIntPass, StructB BaPass);

public record struct AS(F FooFail, G GShouldFail, H HShouldPass, string SPass, Inner InnPass, object OFail);

public record struct Tup1(int IPass, (int a, int b) TupPass, DateTime? DtPass);

public record struct Tup2(int IPass, (int a, int[] b, object o) TupFail);

public record struct Tup3(int IPass, (bool, int) TupPass);

// this checks fields and properties
public record class RecFields(int IPass, string SPass, object OFail)
{
	public IList<string>? FieldFail;
	public int[]? PropertyFail { get; set; }

	public int FieldPass;
	public string? PropertyPass { get; set; }

    //uncomment, and the failures will disappear
	//public virtual bool Equals(RecFields? other) => true;

	public override int GetHashCode() => 1;
}

// the record class has an Equals method, so its assumed to be ok
public record class HasEqualsRecordClass(IReadOnlyList<int> NumsPass)
{
	public virtual bool Equals(HasEqualsRecordClass? other) => other != null && NumsPass.SequenceEqual(other.NumsPass);

	public override int GetHashCode() => NumsPass.GetHashCode();
}

// the record struct has an Equals method, so its assumed to be ok
public record struct HasEqualsRecordStruct(IReadOnlyList<int> NumsPass)
{
	public readonly bool Equals(HasEqualsRecordStruct other) => NumsPass.SequenceEqual(other.NumsPass);

	public override readonly int GetHashCode() => NumsPass.GetHashCode();
}

// ============= Supporting types =============

// this is nested inside another record
public record Inner(int IPass, string JPass, DateTime DtPass);

// when used in record, this fails because it is a class with no Equals method
public class F { public int[]? n; }

// when used in record, this fails because it is a class with no Equals method
public class G { public int i; }

// when used in record, this passes because it has Equals(T)
public class H
{
	public int i;

	public bool Equals(H? other) => other != null && other.i == this.i;

	//public override bool Equals(object? obj) => Equals(obj as H);

	public override int GetHashCode() => i.GetHashCode();
}

// when used in a record, this fails because no Equals method
public struct StructA { public int[] Numbers; }

// when used in a record, this passes because its fields have value semantics
public struct StructB { public int A; public string S; }