Proposed solution to the "predetermined states" problem Wolfgang Corcoran-Mathe (22 May 2026 23:18 UTC)
Re: Proposed solution to the "predetermined states" problem Wolfgang Corcoran-Mathe (22 May 2026 23:26 UTC)
Re: Proposed solution to the "predetermined states" problem Peter McGoron (23 May 2026 00:02 UTC)
Re: Proposed solution to the "predetermined states" problem Peter McGoron (23 May 2026 01:18 UTC)
Re: Proposed solution to the "predetermined states" problem Peter McGoron (23 May 2026 13:44 UTC)
Re: Proposed solution to the "predetermined states" problem Wolfgang Corcoran-Mathe (23 May 2026 15:15 UTC)
Re: Proposed solution to the "predetermined states" problem Wolfgang Corcoran-Mathe (23 May 2026 16:28 UTC)
Re: Proposed solution to the "predetermined states" problem Peter McGoron (23 May 2026 17:00 UTC)
Re: Proposed solution to the "predetermined states" problem Peter McGoron (23 May 2026 17:02 UTC)
Re: Proposed solution to the "predetermined states" problem Wolfgang Corcoran-Mathe (24 May 2026 17:15 UTC)

Re: Proposed solution to the "predetermined states" problem Peter McGoron 23 May 2026 01:17 UTC

The best solution to this problem is still to require (at least one) RNG
to take non-negative exact integers as seed arguments. The
transformation of integer to seed is implementation dependent and the
implementer can add whatever checks are necessary to make sure that the
RNG is in a reasonable state.

For an example of why seeding with repetitive patterns is bad, try
seeding xoshiro256++ with the seed {1, 1, 1, 1} and printing out the
first few numbers in hexadecimal. There are many repeated bits:

0x1000001
0x800001
0x10
0x2210000220010
0x4002200044c00025
0x202a84444800899
0x4113008091a88885
0x1198002260100eb5
0x4546a2b755accc17
0x9b82e94f147868a9

It stabilizes after that, but there are still patterns I can see by eye.

-- Peter McGoron