GetHashCode Guidelines in C#

I read in the Essential C# 3.0 and .NET 3.5 book that:

>GetHashCode()’s returns over the life of a particular object should be constant (the same value), even if the object’s data changes. In many cases, you should cache the method return to enforce this.

Is this a valid guideline?

I have tried a couple built-in types in .NET and they didn't behave like this.

It's been a long time, but nevertheless I think it is still necessary to give a correct answer to this question, including explanations about the whys and hows. The best answer so far is the one citing the MSDN exhaustivly - don't try to make your own rules, the MS guys knew what they were doing.

But first things first: The Guideline as cited in the question is wrong.

Now the whys - there are two of them

First why: If the hashcode is computed in a way, that it does not change during the lifetime of an object, even if the object itself changes, than it would break the equals-contract.

Remember: "If two objects compare as equal, the GetHashCode method for each object must return the same value. However, if two objects do not compare as equal, the GetHashCode methods for the two object do not have to return different values."

The second sentence often is misinterpreted as "The only rule is, that at object creation time, the hashcode of equal objects must be equal". Don't really know why, but that's about the essence of most answers here as well.

Think of two objects containing a name, where the name is used in the equals method: Same name -> same thing. Create Instance A: Name = Joe Create Instance B: Name = Peter

Hashcode A and Hashcode B will most likely not be the same. What would now happen, when the Name of instance B is changed to Joe?

According to the guideline from the question, the hashcode of B would not change. The result of this would be: A.Equals(B) ==> true But at the same time: A.GetHashCode() == B.GetHashCode() ==> false.

But exactly this behaviour is forbidden explicitly by the equals&hashcode-contract.

Second why: While it is - of course - true, that changes in the hashcode could break hashed lists and other objects using the hashcode, the reverse is true as well. Not changing the hashcode will in the worst case get hashed lists, where all of a lot of different objects will have the same hashcode and therefor be in the same hash bin - happens when objects are initialized with a standard value, for example.

---

Now coming to the hows Well, on first glance, there seems to be a contradiction - either way, code will break. But neither problem does come from changed or unchanged hashcode.

The source of the problems is well described in the MSDN:

From MSDN's hashtable entry:
> Key objects must be immutable as long > as they are used as keys in the > Hashtable.

This does mean:

Any object that creates a hashvalue should change the hashvalue, when the object changes, but it must not - absolutely must not - allow any changes to itself, when it is used inside a Hashtable (or any other Hash-using object, of course).

First how Easiest way would of course be to design immutable objects only for the use in hashtables, that will be created as copys of the normal, the mutable objects when needed. Inside the immutable objects, it's obviusly ok to cache the hashcode, since it's immutable.

Second how Or give the object a "you are hashed now"-flag, make sure all object data is private, check the flag in all functions that can change objects data and throw an exception data if change is not allowed (i.e. flag is set). Now, when you put the object in any hashed area, make sure to set the flag, and - as well - unset the flag, when it is no longer needed. For ease of use, I'd advise to set the flag automatically inside the "GetHashCode" method - this way it can't be forgotten. And the explicit call of a "ResetHashFlag" method will make sure, that the programmer will have to think, wether it is or is not allowed to change the objects data by now.

Ok, what should be said as well: There are cases, where it is possible to have objects with mutable data, where the hashcode is nevertheless unchanged, when the objects data is changed, without violating the equals&hashcode-contract.

This does however require, that the equals-method is not based on the mutable data as well. So, if I write an object, and create a GetHashCode method that does calculate a value only once and stores it inside the object to return it on later calls, then I must, again: absolutely must, create a Equals method, that will use stored values for the comparison, so that A.Equals(B) will never change from false to true as well. Otherwise, the contract would be broken. The result of this will usually be that the Equals method doesn't make any sense - it's not the original reference equals, but it is neither a value equals as well. Sometimes, this may be intended behaviour (i.e. customer records), but usually it is not.

So, just make GetHashCode result change, when the object data changes, and if the use of the object inside of hash using lists or objects is intended (or just possible) then make the object either immutable or create a readonly flag to use for the lifetime of a hashed list containing the object.

(By the way: All of this is not C# oder .NET specific - it is in the nature of all hashtable implementations, or more generally of any indexed list, that identifying data of objects should never change, while the object is in the list. Unexpected and unpredictable behaviour will occur, if this rule is broken. Somewhere, there may be list implementations, that do monitor all elements inside the list and do automatic reindexing the list - but the performance of those will surely be gruesome at best.)

The answer is mostly, it is a valid guideline, but perhaps not a valid rule. It also doesn't tell the whole story.

The point being made is that for mutable types, you cannot base the hash code on the mutable data because two equal objects must return the same hash code and the hash code has to be valid for the lifetime of the object. If the hash code changes, you end up with an object that gets lost in a hashed collection because it no longer lives in the correct hash bin.

For example, object A returns hash of 1. So, it goes in bin 1 of the hash table. Then you change object A such that it returns a hash of 2. When a hash table goes looking for it, it looks in bin 2 and can't find it - the object is orphaned in bin 1. This is why the hash code must not change for the lifetime of the object, and just one reason why writing GetHashCode implementations is a pain in the butt.

Update
Eric Lippert has posted a blog that gives excellent information on GetHashCode.

Additional Update
I've made a couple of changes above:

1. I made a distinction between guideline and rule.
2. I struck through "for the lifetime of the object".

A guideline is just a guide, not a rule. In reality, GetHashCode only has to follow these guidelines when things expect the object to follow the guidelines, such as when it is being stored in a hash table. If you never intend to use your objects in hash tables (or anything else that relies on the rules of GetHashCode), your implementation doesn't need to follow the guidelines.

When you see "for the lifetime of the object", you should read "for the time the object needs to co-operate with hash tables" or similar. Like most things, GetHashCode is about knowing when to break the rules.

From MSDN

> If two objects compare as equal, the > GetHashCode method for each object > must return the same value. However, > if two objects do not compare as > equal, the GetHashCode methods for the > two object do not have to return > different values. > > The GetHashCode method for an object > must consistently return the same hash > code as long as there is no > modification to the object state that > determines the return value of the > object's Equals method. Note that this > is true only for the current execution > of an application, and that a > different hash code can be returned if > the application is run again. > > For the best performance, a hash > function must generate a random > distribution for all input.

This means that if the value(s) of the object change, the hash code should change. For example, a "Person" class with the "Name" property set to "Tom" should have one hash code, and a different code if you change the name to "Jerry". Otherwise, Tom == Jerry, which is probably not what you would have intended.

---

Edit:

Also from MSDN: > Derived classes that override GetHashCode must also override Equals to guarantee that two objects considered equal have the same hash code; otherwise, the Hashtable type might not work correctly.

From MSDN's hashtable entry: > Key objects must be immutable as long as they are used as keys in the Hashtable.

The way I read this is that mutable objects should return different hashcodes as their values change, unless they are designed for use in a hashtable.

In the example of System.Drawing.Point, the object is mutable, and does return a different hashcode when the X or Y value changes. This would make it a poor candidate to be used as-is in a hashtable.

I think that the documentation regarding GetHashcode is a bit confusing.

On one hand, MSDN states that the hashcode of an object should never change , and be constant On the other hand, MSDN also states that the return value of GetHashcode should be equal for 2 objects, if those 2 objects are considered to be equal.

MSDN: > A hash function must have the following properties: > > - If two objects compare as equal, the GetHashCode method for each object > must return the same value. However, > if two objects do not compare as > equal, the GetHashCode methods for the > two object do not have to return > different values. > - The GetHashCode method for an object must consistently return the > same hash code as long as there is no > modification to the object state that > determines the return value of the > object's Equals method. Note that this > is true only for the current execution > of an application, and that a > different hash code can be returned if > the application is run again. > - For the best performance, a hash function must generate a random > distribution for all input.

Then, this means that all your objects should be immutable, or the GetHashcode method should be based on properties of your object that are immutable. Suppose for instance that you have this class (naive implementation):

public class SomeThing
{
      public string Name {get; set;}

      public override GetHashCode()
      {
          return Name.GetHashcode();
      }

      public override Equals(object other)
      {
           SomeThing = other as Something;
           if( other == null ) return false;
           return this.Name == other.Name;
      }
}

This implementation already violates the rules that can be found in MSDN. Suppose you have 2 instances of this class; the Name property of instance1 is set to 'Pol', and the Name property of instance2 is set to 'Piet'. Both instances return a different hashcode, and they're also not equal. Now, suppose that I change the Name of instance2 to 'Pol', then, according to my Equals method, both instances should be equal, and according to one of the rules of MSDN, they should return the same hashcode.
However, this cannot be done, since the hashcode of instance2 will change, and MSDN states that this is not allowed.

Then, if you have an entity, you could maybe implement the hashcode so that it uses the 'primary identifier' of that entity, which is maybe ideally a surrogate key, or an immutable property. If you have a value object, you can implement the Hashcode so that it uses the 'properties' of that value object. Those properties make up the 'definition' of the value object. This is of course the nature of a value object; you're not interested in it's identity, but rather in it's value.
And, therefore, value objects should be immutable. (Just like they are in the .NET framework, string, Date, etc... are all immutable objects).

Another thing that comes in mind:
During which 'session' (I don't know really how I should call this) should 'GetHashCode' return a constant value. Suppose you open up your application, load an instance of an object out of the DB (an entity), and get its hashcode. It will return a certain number. Close the application, and load the same entity. Is it required that the hashcode this time has the same value as when you loaded the entity the first time ? IMHO, not.

This is good advice. Here's what Brian Pepin has to say on the matter:

> This has tripped me up more than > once: Make sure GetHashCode always > returns the same value across the > lifetime of an instance. Remember that > hash codes are used to identify > "buckets" in most hashtable > implementations. If an object's > "bucket" changes, a hashtable may not > be able to find your object. These can > be very hard bugs to find, so get it > right the first time.

Not directly answering your question, but - if you use Resharper, do not forget it has a feature that generates a reasonable GetHashCode implementation (as well as the Equals method) for you. You can of course specify which members of the class will be taken into account when computing the hashcode.

Check out this blog post from Marc Brooks:

VTOs, RTOs and GetHashCode() -- oh, my!

And then check out the follow up post (can't link as I'm new, but there's a link in the initlal article) which discusses further and covers some minor weaknesses in the initial implementation.

This was everything I needed to know about creating a GetHashCode() implementation, he even provides a download of his method along with some other utilities, in short gold.

The hashcode never changes, but it's also important to understand where the Hashcode is coming from.

If your object is using value semantics, i.e. the object's identity is defined by its values (like String, Color, all structs). If your object's identity is independent of all of its values, then the Hashcode is identified by a subset of its values. For example, your StackOverflow entry is stored in a database somewhere. If you change your name or email, your customer entry stays the same, although some values have changed (ultimately you're usually identified by some long customer id #).

So in short:

Value type semantics - Hashcode is defined by values Reference type semantics - Hashcode is defined by some id

I suggest you read Domain Driven Design by Eric Evans, where he goes into entities vs value types (which is more or less what I attempted to do above) if this still doesn't make sense.

Check out Guidelines and rules for GetHashCode by Eric Lippert