When using adler32() as a hash function, one should expect rare collisions.

We can do the exact math of collisions probability, but roughly speaking,
since it is a 32-bits hash function, there should not be many collisions
on a sample set of a few thousands items.

This is hardly the case.

Here is an example: let's take strings that include a date in the middle, something like 

"Some prefix text " + date + " some postfix text."

where the dates` format is yyyy-mm-dd, and looping over 2012.

There are 91 collisions in this example!

Even worse: there are 7 cases where 3 dates collided.

How come such a commonly-used hash function perform so poorly?
Or am I missing something?

Here are the detailed results of the above example: 

0x592a0f1f: 2012-01-30, 2012-02-02, 2012-10-21
0x592b0f1f: 2012-02-11, 2012-10-30, 2012-11-02
0x593d0f20: 2012-01-31, 2012-02-03, 2012-10-22
0x593e0f20: 2012-02-12, 2012-10-31, 2012-11-03
0x59410f20: 2012-03-11, 2012-11-30, 2012-12-02
0x59560f21: 2012-03-30, 2012-04-02, 2012-12-21
0x59690f22: 2012-03-31, 2012-04-03, 2012-12-22

0x59020f1d: 2012-01-10, 2012-10-01
0x59150f1e: 2012-01-11, 2012-10-02
0x59160f1e: 2012-01-20, 2012-10-11
0x59170f1e: 2012-02-01, 2012-10-20
0x59180f1e: 2012-02-10, 2012-11-01
0x59280f1f: 2012-01-12, 2012-10-03
0x59290f1f: 2012-01-21, 2012-10-12
0x592c0f1f: 2012-02-20, 2012-11-11
0x592d0f1f: 2012-03-01, 2012-11-20
0x592e0f1f: 2012-03-10, 2012-12-01
0x593b0f20: 2012-01-13, 2012-10-04
0x593c0f20: 2012-01-22, 2012-10-13
0x593f0f20: 2012-02-21, 2012-11-12
0x59400f20: 2012-03-02, 2012-11-21
0x59420f20: 2012-03-20, 2012-12-11
0x59430f20: 2012-04-01, 2012-12-20
0x594e0f21: 2012-01-14, 2012-10-05
0x594f0f21: 2012-01-23, 2012-10-14
0x59500f21: 2012-02-04, 2012-10-23
0x59510f21: 2012-02-13, 2012-11-04
0x59520f21: 2012-02-22, 2012-11-13
0x59530f21: 2012-03-03, 2012-11-22
0x59540f21: 2012-03-12, 2012-12-03
hash hash-collision adler32
Edit
Report