Why are these hashcodes equals?

Question

This test is failing :

var hashCode = new 
{
    CustomerId = 3354,
    ServiceId = 3,
    CmsThematicId = (int?)605,
    StartDate = (DateTime?)new DateTime(2013, 1, 5),
    EndDate = (DateTime?)new DateTime(2013, 1, 6)
}.GetHashCode();
var hashCode2 = new
{
    CustomerId = 1210,
    ServiceId = 3,
    CmsThematicId = (int?)591,
    StartDate = (DateTime?)new DateTime(2013, 3, 31),
    EndDate = (DateTime?)new DateTime(2013, 4, 1)
}.GetHashCode();
Assert.AreNotEqual(hashCode, hashCode2);

Can you tell me why ?

He's right. This code does produce identical hash codes. Don't down-vote so quickly people. — Samuel Neff, Jan 29 '13 at 14:26
@dasblinkenlight, run the code. The hash codes are the same. — Samuel Neff, Jan 29 '13 at 14:32
@SamuelNeff I just did - my comment above is a link to ideone. I got 429571660 and -626441557. — Sergey Kalinichenko, Jan 29 '13 at 14:33
@dasblinkenlight, I get 1660129507 and 1660129507. ideone is using mono. Looks like mono's implementation of anonymous type hash codes is different. — Samuel Neff, Jan 29 '13 at 14:35
@SamuelNeff That was pretty much my point: calculation of hash code in anonymous classes is heavily dependent on the system. Ideone runs Mono. — Sergey Kalinichenko, Jan 29 '13 at 14:37
@dasblinkenlight, but that doesn't invalidate the question. Perhaps it should have been more specific, but it's still a valid question. — Samuel Neff, Jan 29 '13 at 14:38

score 3 · Answer 1 · edited May 23 '17 at 10:24

It's kinda amazing you found this coincidence.

Anonymous classes have a generated GetHashCode() method that generates a hash code by combining the hash codes of all properties.

The calculation is basically this:

  public override int GetHashCode()
  {
    return        -1521134295 * 
                ( -1521134295 * 
                ( -1521134295 * 
                ( -1521134295 * 
                ( -1521134295 * 
                   1170354300 + 
                  CustomerId.GetHashCode()) +
                  ServiceId.GetHashCode()) + 
                  CmsThematicId.GetHashCode()) + 
                  StartDate.GetHashCode()) + 
                  EndDate.GetHashCode();
  }

If you change any of the values of any of the fields, the hash code does change. The fact that you found two different sets of values that happen to get the same hash codes is a coincidence.

Note that hash codes are not necessarily unique. It's impossible to say hash codes would always be unique since there can be more objects than hash codes (although that is a lot of objects). Good hash codes provide a random distribution of values.

NOTE: The above is from .NET 4. Different versions of .NET may different and Mono differs.

If you want to actually compare two objects for equality then use .Equals(). For anonymous objects it compares each field. An even better option is to use an NUnit constraint that compares each field and reports back which field differs. I posted a constraint here:

https://stackoverflow.com/a/2046566/118703

Jim Mischel · Answer 2 · 2023-02-13T16:26:38.887

Did you run into this when processing a fairly large amount of data?

Welcome to the wonderful world of hash codes. A hash code is not a "unique identifier." It can't be. There is an essentially infinite number of possible different instances of that anonymous type, but only 2^32 possible hash codes. So it's guaranteed that if you create enough of those objects, you're going to see some duplicates. In fact, if you generate 70,000 of those objects randomly, the odds are better than 50% that two of them will have the same hash code.

See Birthdays, Random Numbers, and Hash Codes, and the linked Wikipedia article for more info.

As for why some people didn't see a duplicate and others did, it's likely that they ran the program on different versions of .NET. The algorithm for generating hash codes is not guaranteed to remain the same across versions or platforms:

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.

I need these to be unique because I want to make sure there is no record in the DB with the same combination of values, how do you think I should do it ? I needs to be efficient because I'm dealing with a lot of data. — remi bourgarel, Jan 30 '13 at 08:25
@remibourgarel: It's hard to give a recommendation without knowing more about the data and how you're using it, or what you want this unique key to do. In general, no hashing scheme will guarantee uniqueness; there's always a risk of collision. Unless you're talking about perfect hashing, which is a very special case. Post another question that details what you're trying to do, and it's likely we can supply a solution. — Jim Mischel, Jan 30 '13 at 12:33
http://stackoverflow.com/questions/14599976/how-to-prevent-parameters-to-be-used-again#comment20384762_14599976 — remi bourgarel, Jan 30 '13 at 12:52

score 1 · Answer 3 · edited Jan 31 '13 at 13:02

1

Your test is not valid.

Because hash codes are not guaranteed to be unique (see other answers for a good explanation), you should not test for uniqueness of hash codes.

When writing your own GetHashCode() method, it is a good idea to test for even distribution of random input, just not for uniqueness. Just make sure that you use enough random input to get a good test.

The MSDN spec on GetHashCode specifically states:

For the best performance, a hash function must generate a random distribution for all input.

This is all relative, of course. A GetHashCode() method that is being used to put 100 objects in a dictionary doesn't need to be nearly as random as a GetHashCode() that puts 10,000,000 objects in a dictionary.

edited Jan 31 '13 at 13:02

Theodoros Chatzigiannakis

28,773
8
68
104

answered Jan 29 '13 at 15:22

Joel Rondeau

7,486
2
42
54

what do you mean by random distribution ? – remi bourgarel Jan 30 '13 at 08:17
Random distribution is not an accurate statement. What it should state is even distribution of random input. I'll update. – Joel Rondeau Jan 30 '13 at 15:16

remi bourgarel · Accepted Answer · 2013-01-31T13:07:44.930

Jim suggested me (in the chat room) to store my parameters so when i display my parameters, select the not used ones, then when someone registers I flag it as used. But it's a big PITA to generate all the parameters.

So my solution is to build a int64 hashcode like this

const long i = -1521134295;    
return -i * (-i * (-i * (-i * -117147284 + customerId.GetHashCode()) + serviceId.GetHashCode()) + cmsThematicId.GetHashCode()) + startDate.GetHashCode();

I removed the end date because Its value was depending on serviceId and startDate so I shouldn't have add this to the hashcode in the firstplace. I copy/pasted it from a decompilation of the generated class. I got not colision if I do a test with 300 000 differents combinations.

Why are these hashcodes equals?

4 Answers4

Linked