Using GetHashCode in Equals

I have a scenario when I want to have a HashSet<ITrackableObjectData>, where ITrackableObjectData has string TypeName and object Id readonly immutable properties. Id is a primitive value, ie. string, int, long, Guid etc.

I want to use these equality rules:

• If the TypeName property is different, should anyway be considered non-equal.
• If the Id is zero, null etc, should use the base object hashcode (reference equal).
• If Id is not zero, then it should compare the TypeName and Id.

I created the following comparer to provided to the HashSet's constructor, and I'd like to hear from you of any improvements/vandalism in my code. As you can see I'm using the GetHashCode in the Equals method, is this a bad-idea?
Note that these are my equality requirements in my scenario.

class ITrackableObjectDataComparer : EqualityComparer<ITrackableObjectData>
{
public override bool Equals(ITrackableObjectData x, ITrackableObjectData y) =>
GetHashCode(x) == GetHashCode(y);

public override int GetHashCode(ITrackableObjectData obj)
{
if (obj == null) return 0;
var typeHash = obj.TypeName == null ? 0 : obj.TypeName.GetHashCode();
var idHash = obj.Id == null ? 0 : obj.Id.GetHashCode();
idHash = idHash == 0 ? obj.GetHashCode() : idHash;
return typeHash ^ idHash;
}
}


Note that this EqualityComparer is a private class and is only meant to serve the purpose of this specific HashSet which is not exposed at all.
In fact I don't also really care about the Equals, but rather about the GetHashCode, so I know whether an object with my specs have already been added to a HashSet.

Update:

This is basically the equality checker that fulfills my requirements (but again, I don't need an equality comparer, I need a hash equality-comparer)

public override bool Equals(ITrackableObjectData x, ITrackableObjectData y)
{
if (x == y) return true; //ref eq.
else if (x == null)
return y == null;
else if (y == null)
return false;
else if (x.TypeName != y.TypeName)
return false;
else
{
if (x.Id == null)
return y.Id == null;
else if (y.Id == null)
return false;
else
{
int xId = x.Id.GetHashCode(),
yId = y.Id.GetHashCode();
if (xId == 0 || yId == 0)
return x.Equals(y); //use default eq.
else
return xId == yId;
}
}
}

• It would have been interesting to see this ITrackableObjectData and one of its implementations... Especially given the nullability of both values involved. Jun 5 '17 at 13:46
• Two reference-equal objects are guaranteed to have the same hash code, but two equal-hash-code objects are not guaranteed to be reference equal. Why? Because there can be more than 2<sup>32</sup> reference objects, but there are only that many hash codes, so in that case there must be at least one collision. Remember, references are allowed to be 64 bits. Jun 5 '17 at 17:50
• @Shimmy: You're not following the logic. The logic is not there will only be a collision when there are four billion objects. The logic is there MUST be a collision when there are four billion objects and therefore collisions are POSSIBLE. Your theory that collisions are impossible amongst reference objects when there are a small number of them is simply false. You are required to write logic that assumes that hashes may collide. They collide with surprising frequency thanks to the birthday "paradox". Your code is completely and utterly wrong and you should not do this. Jun 5 '17 at 21:10
• You are allowed to conclude that unequal hash codes imply unequal items. That is, you may implement Equals as if (x.GetHashcode() != y.GetHashcode()) return false; else ... but there has to be something in that else that deals with the situation where there are equal hash codes but unequal objects. Jun 5 '17 at 21:13
• @Shimmy you are missing Eric's point. Your Equals implementation does not follow the requirements you listed. It's just wrong and it won't work. It might work most of the time due to pure luck, but sooner or later you will lose objects due to hashcode collisions, which your implementation does not account for. You might not care about it, but users of your software most likely will when they lose their data. Whether or not this class is private is completely irrelevant. Private class should work correctly too. Jun 5 '17 at 23:12

As Nikita mentioned, never use GetHashCode for equality comparison!

This SO answer shows a more sophisticated GetHashCode implementation. Translated to you case it could look like:

public override int GetHashCode(ITrackableObjectData obj)
{
if (obj == null)
return 0;
if (obj.TypeName == null && obj.Id == null)
return obj.GetHashCode();

unchecked // disable overflow, for the unlikely possibility that you
{         // are compiling with overflow-checking enabled
int hash = 27;
hash = (13 * hash) + obj.TypeName?.GetHashCode() ?? 0;
hash = (13 * hash) + obj.Id?.GetHashCode() ?? 0;
return hash;
}
}

• In my scenario, since TypeName is never expected to be equal to Id, sum looks redundant to me. Jun 5 '17 at 12:42
• I think you misunderstand why we combine with something more sophisticated than ^. It's not to avoid a hash of 0 (which is a perfectly acceptable hash value); it's to reduce the likelihood of collision (where two non-equal objects have the same hash value). Collisions aren't wrong, but if you have too many, they can affect performance of algorithms that depend on hash codes. For example, we'd like (0,1) to produce a different result to (1,0), which a simple ^ or + won't achieve. Jun 5 '17 at 16:57
• If a class can be expected to cache the result from GetHashCode in cases where it's expensive to compute, it can sometimes be advantageous to call GetHashCode as the first step in an equality comparison. If, for example, one has two deeply-immutable trees which are identical except for the last branch, having Equals check GetHashCode() values before doing anything else may allow it to quickly return false without having to scan through all the matching branches first. Such an approach would be counterproductive if GetHashCode isn't cached, however. Jun 5 '17 at 17:56
• @Shimmy: Do you know for which Id values Id.GetHashCode() returns 0? Jun 6 '17 at 6:27
• Than try: new Guid("d7a24d8c-75a2-4d64-9061-a21aba171be8").GetHashCode() ;) Jun 6 '17 at 17:43

I think you are missing the point. Different objects can produce same hashcode. This can happen:

obj.Equals(otherObj)  // false
obj.GetHashCode() == otherObj.GetHashCode() //true


If for some reason you want your Equals methods to return true when hashcodes are equal, then by all means. Make sure to document this behavior though.

However if you want to compare references , then use ReferenceEquals method instead. Hashcode comparison will not work for that scenario, because hashcode equality does not guarantee that object are (reference-)equal.

P.S. Also keep in mind, that hashcodes should not change, otherwise hash-tables might break. Make sure that Id and TypeName are read-only properties.

• Thank you. See my updates. The cautions you mentioned are valid in my scenario. Jun 5 '17 at 12:34
• @Shimmy I'm not sure I follow your explanation. You may not care about Equals method, but hash-tables will use it internally to compare items, it is pivotal. If equality comparison does not work correctly, neither will your HashSet. Jun 5 '17 at 12:45
• @Shimmy,GetHashCode method is used to find correct bucked inside a HashSet. Equals is then used to find correct item inside that bucket. Those are two different methods that have different signature and serve different purpose. I don't know what you mean, when you say that they are "equal". Does your Equals return true when hashcodes are equal? Yes. Does this behavior match your requirements? Most likely not, since instead of comparing IDs you compare IDs' hashcodes, like it's the same thing. It is not the same thing. Jun 5 '17 at 13:36
• One of the other cases to consider is that two sets of different values can return the same hash code. Jun 5 '17 at 13:58
• @Shimmy That’s fair enough, but then don’t implement Equals for your class: by doing so you’re lying to the user of your class (even if that’s just you — it’s simply bad documentation). Unfortunately C# doesn’t allow you to delete the method outright but you should then leave it with its default semantic: reference equality. Jun 5 '17 at 21:52