bug-gnu-emacs
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

bug#32605: [w64] (random) never returns negative


From: Andy Moreton
Subject: bug#32605: [w64] (random) never returns negative
Date: Sat, 14 Aug 2021 12:06:00 +0100
User-agent: Gnus/5.13 (Gnus v5.13) Emacs/28.0.50 (windows-nt)

On Sat 14 Aug 2021, Eli Zaretskii wrote:

>> From: Andy Moreton <andrewjmoreton@gmail.com>
>> Date: Sat, 14 Aug 2021 09:31:06 +0100
>> 
>> I'm not an expert on random numbers either, and your efforts are not an
>> annoyance, but I am puzzled why you so strongly prize performance over
>> correctness in this instance.
>
> Because I have no idea how important the "correctness" part is, or
> why.  OTOH, this stuff, when used, tends to be in the inner loops, so
> performance matters.

I doubt anyone expects cryptographic quality randomness or any given
statistical distribution from such a general purpose routine, but they
have a reasonable expectation that the results from 'get_random' do not
have stuck bits that are always non-random.

In which case perhaps the solution is to change the RAND_BITS logic
in sysdep.c on Windows to override the RAND_BITS definition:

+   #ifdef WINDOWSNT
+   /* Use w32.c replacement for random().  */
+   # define RAND_BITS 15
+   #endif

    #ifndef RAND_BITS
    # ifdef HAVE_RANDOM
    #  define RAND_BITS 31
    # else /* !HAVE_RANDOM */
    ...
    #endif

..and then in w32.c make 'random' return the 15bit value from
'rand_as183':

    int
    random (void)
    {
      /* rand_as183 () gives us 15 random bits.  */
      return rand_as183 ();
    }

That should result in 'get_random' receiving 15 bits of randomness in
each loop iteration and thus computing a valid result.

[This could obviously be optimised to open code 'rand_as183' in 'random',
 or allow the compiler to inline it by moving the w32.c implementations
 of 'random' and 'srandom' into sysdep.c]

As 'get_random_bignum' (in fns.c) calls 'get_random' in a loop, that
should also remove bugs from that function on this platform.


Perhaps this would be helped by having a test for 'get_random', to check
that every bit of a fixnum is toggled after a reasonable number of
calls. While that does not test the statistical distribution of the
random number sequence, it would ensure that the values returned by
'get_random' are not always positive, or always even, etc.

    AndyM







reply via email to

[Prev in Thread] Current Thread [Next in Thread]