"Todd A. Cook" <tcook@blackducksoftware.com> writes:
> If hashint8() is ultimately invoked by TupleHashTableHash() in execGroups.c,
> it might be magnifying the difficulties here. The least significant bits,
> which are used as the bucket number in simplehash.h, are not very well
> distributed:
> select val, to_hex(val), to_hex(hashint8(val)) from temp_f_03 limit 15 ;
> val | to_hex | to_hex
> ----------------------+------------------+----------
> 4444319256653758784 | 3dad64d121468140 | 805ffffe
> 554179993563924608 | 7b0d7c49a018880 | 84dffffb
> -3383965646518123872 | d109bd2c6b2982a0 | 9c3ffff7
> -4706811054739454944 | beae0c48915f8420 | 191ffff6
> 618200668902031424 | 8944a3ba5d08040 | 2a3ffff0
> 5074043922812601024 | 466aa01079f982c0 | 7effffee
> -8783188184262212928 | 861bd8e1b9a482c0 | a6bfffea
> -4597800992953433792 | c031545b6b128140 | b1dfffea
> 8563040839807173408 | 76d608465dde8320 | 7d9fffe6
> 6092569112843158816 | 548d27540c888520 | 6f9fffe6
> -7313351060369079936 | 9a81c1f558f98180 | 68ffffe5
> -1786712428165627488 | e7345283536981a0 | 73ffffe5
> -6153596242570280896 | aa9a08d20e6b8040 | ac3fffd8
> 88426174078092128 | 13a272306c58360 | b57fffd8
> -5305589938458295680 | b65ec20faa4e8280 | ba9fffd3
Checking out hashint8() on random data shows no such obvious fault.
You've apparently got a data set that exposes a weakness in hashint8,
but it's not very clear what that is.
In any case, the hashtable code needs to not fall over in the
presence of a lot of collisions, regardless of the exact reason
for there being a lot.
regards, tom lane
