Re: feature request: Pre-seeding the random number generator

[Sean M. Burke]

On 10/22/2006 03:14 AM, Didier LINK wrote:
> Le samedi 21 octobre 2006 à 14:56 -0800, Sean M. Burke a écrit :
> > This is a feature request: I'd really really appreciate if you could
> > make the random number generator in ghostscript pre-seeded.
>
> Why do you need this enhancement exactly ?

I like to generate documents that contain, well, some degree of 
randomness, without having to duck out of PostScript to fetch a new 
srand seed.  For example, I like to reload this:
http://interglacial.com/~sburke/pub/PostScript/dynamically_generated_maze_simpler.ps.txt
but I very often get the same two or three mazes over and over and over.

This problem recurs with many other documents of mine that attempt 
randomness.

If that sounds dull, tell me, and I can make up something totally 
compelling about how I'm actually the person generating the entire 
Albanian Lottery's cards using /just/ sh and GhostScript and a can of 
gold paint, and how the ghost of Hoxha is angry that I keep generating 
the same four or five cards over and over.
How's that for a use-case and/or a rich comedic premise?

> > (Perl has been doing this for about ten years-- in the surprisingly
> > rare circumstances where programmers want /predictable/
> > pseudorandomness, they just need to explicitly feed srand an particular 
> > value.
> > If you want a bit of inspiration from Perl's pre-seeding, the code for
> > it is in here:
> >   http://www.devdaily.com/scw/c/perl-5.8.5/util.c.shtml
> > and look for the function "Perl_seed(pTHX)".)
> >
> > I considered just replacing rand with something that reads from 
> > /dev/random, but besides portability problems, such file operations are 
> > apparently inaccessible in gs's (imminently default) SAFER mode.
>
> I think it's possible to introduce a better initial value in the code
> with a read from /dev/random but it's not portable on all architectures
> that's are supported by ghostscript. I need to investigate in this way.

Excellent!  At this point, anything would be better than the current 
behavior.

(In the old days of my programming in other languages with poor 
pre-seeding: I used to do pretty well with taking the current process ID 
number (or, if under Windows, instead using GetTickCount, since Windows 
often reuses process IDs almost immediately), shift it four bits left, 
and xor that against the lower twelve bits of time or utime or something 
of the sort.)

> > I used to pre-seed by just doing "realtime srand", but now I see that
> > realtime has a very NONrandom value (see below).
>
> Yes I see, and it's plateform dependent, on my amd64 the random numbers
> generated by your perl script are a little bit better. But why do a
> srand at each iteration of the loop ? It's certainly better to
> initialize one time and use rrand function to iterate new random numbers
> inside gs. I'm wrong ?

It was just minimal example code to illustrate the problem of realtime's 
value not being useful as an initial value for srand.

> > I know that PostScript docs typically warn that rand's generator is 
> > typically pretty dumb, but that doesn't mean it /has/ to be dumb in 
> > GhostScript.
> Yes but they need to be compliant with the Adobe's specifications and
> reference implementation of postscript interpreter.

I appreciate you noting that One Must Be Very Careful-- but I swear, if 
Adobe actually does mandate that the random number generator /not/ be 
pre-seeded, I will SHOWER you with gifts.