On Thu, Nov 15, 2018 at 4:38 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Thomas Munro <tmunro@postgresql.org> writes:
> > Increase the number of possible random seeds per time period.
>
> Um, this bit is *not* right:
>
> + ((unsigned int) MyStartTimestamp >> 20));
>
> You're cutting it down to 32 bits and then right shifting, which
> means you are shifting in a lot of zeroes when you could be shifting
> in something that's not quite as predetermined. I had in mind to do
> the shifts in uint64 arithmetic and then narrow the final XOR result
> to int (maybe let the compiler do that implicitly).
Will this close the case?
- srandom(((unsigned int) MyProcPid) ^
- ((unsigned int) MyStartTimestamp << 12) ^
- ((unsigned int) MyStartTimestamp >> 20));
+ srandom(((uint64) MyProcPid) ^
+ ((uint64) MyStartTimestamp << 12) ^
+ ((uint64) MyStartTimestamp >> 20));
--
Thomas Munro
http://www.enterprisedb.com