Update rand() documentation to be more modern and clear

[scottchiefbaker]

Learning more about how Perl implements random numbers lead me to clean up and update the documentation.

[scottchiefbaker]

Any more advice on fixes/updates on wording so this can get merged?

[demerphq]

On Tue, 12 Nov 2024, 14:20 guest20, ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In pod/perlfunc.pod <#22727 (comment)>: > -(Note: If your rand function consistently returns numbers that are too +=item * + +Currently EXPR with the value C<0> is +also special-cased as C<1>. This was undocumented before Perl 5.8.0 +and is subject to change in future versions of Perl. + +=item * + +As of Perl v5.20.0 C<rand> uses the C<drand48()> PRNG to generate random numbers. +This should be acceptable for educational or hobby projects, basic simulations, +or simple random sampling. + +=item * + +If your C<rand> function consistently returns numbers that are too your C<rand> This refers to the rand in the libs/OS you picked. This type of language is used to couch all sorts of porting issues.

Any such language should be removed, we added the hand rolled drand48() specifically to insulate ourselves from OS dependencies. I don't remember if we support a build option to use the system rand/srand (but probably we do), if we do then we can keep these comments but caveat them appropriately; a normal build will use the same (bad) rng everywhere. Yves

…

[demerphq]

On Tue, 12 Nov 2024, 06:44 Scott Baker, ***@***.***> wrote: Any more advice on fixes/updates on wording so this can get merged?

FWIW i don't think we should say that our rand is suitable for any specific purpose. I think we should say that it is not suitable for any mathematically rigorous purpose and leave it to the reader to make their own decisions. Perhaps suggesting some alternatives like MT or WELL variants. WELL-512 is probably one of the better choices these days. Linear congruent random number generators as a whole have a variety of issues, and the one we provide was modern when 16 bit integers were modern and has less state (48 bits of course) than an integer in a standard build. To make things worse iirc we don't even dispose of the high bits in the state, which are substantially less random than the low bits. These days the state of art of non cryptographic generators with limited state is much advanced and people really shouldn't use rand() for anything serious at all. I really wish we bundled a better RNG. Yves

…

[scottchiefbaker]

@demerphq I updated the two sections you referenced.

[demerphq]

On Thu, 14 Nov 2024, 03:32 Scott Baker, ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In pod/perlfunc.pod <#22727 (comment)>: > +As of Perl v5.20.0 C<rand> uses the C<drand48()> PRNG to generate random numbers. +This should be acceptable for educational or hobby projects, basic simulations, +or simple random sampling. + I took out the wording that ChatGPT used

Thanks. Unfortinately it still says: "As a PRNG drand48() should be sufficient for most non-cryptographic needs." I don't think this is correct, an LCG PRNG is not suitable for many tasks, for instance if you use an LCG to choose coordinates you will find that the coordinates chosen all lie on a predictable hyper-lattice. (Wikipedia has a nice animated gif showing this IIRC) IMO Drand48 is basically only suitable for simple sampling tasks which do not require any mathematical rigor. Cheers Yves

[leonerd]

Needs one letter case adjustment but aside from that, LGTM

[scottchiefbaker]

@leonerd good catch... docs updated

[khwilliamson]

@scottchiefbaker These commits should be squashed before we merge