Roland Roberts <roberts@panix.com> writes:
> Call random() several times and test the maximum value against your
> thresholds of 2^15 and 2^31. If random() is generating values in the
> range 1:2^31-1, you would expect half of your values to be greater
> than 2^15-1; more importantly, if you generate, say, 10 values, you
> expect only a 1:1024 chance that they are all below 2^15.
Actually the odds are far better than that. If the range is 2^31-1
then only about 2^-16th of the outputs should be less than 2^15.
So ten probes gives you a failure probability of about 2^-160 not
2^-10.
Generalizing, you could tell the difference between widths of 31,
47, or 63 bits with the same level of reliability.
regards, tom lane