NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/lang/c-part1/7496

Internet Message Format | 1996-08-05 | 3.6 KB

Path: mail2news.demon.co.uk!genesis.demon.co.uk
From: Lawrence Kirby <[e-mail]>
Newsgroups: comp.lang.c
Subject: Re: How to get a random strin
Date: Tue, 27 Feb 96 00:04:01 GMT
Organization: none
Message-ID: <[e-mail]>
References: <4g19id$[e-mail]> <[e-mail]> <4gqir9$[e-mail]>
Reply-To: [e-mail]
X-NNTP-Posting-Host: genesis.demon.co.uk
X-Newsreader: Demon Internet Simple News v1.27
X-Mail2News-Path: genesis.demon.co.uk

In article <4gqir9$[e-mail]>
           [e-mail] "Simon Hosie" writes:

>In article <4g19id$[e-mail]> [e-mail] (Martin Ambuhl)
> writes:
>> [e-mail] (Clapton Chan) in <4fh5od$[e-mail]> asks:
>>
>> [A poor practice follows]
>> if you do not #include <time.h>, you need
>>     srand((unsigned)time(NULL));
>
>Alan Bowler:
>> Worse than poor.  time() returns a time_t which might not be an
>> integer type.  The above code implicitly declares "time()" as returning
>> int.  Adding the cast to unsigned will not change that fact that
>> this implicit declaration means your code could be looking in the
>> wrong place.  For example:  Suppose "time_t" is actually "double"
>> and not "int", and that the machine uses has separate floating
>> point and integer registers.  (x86, pdp-11, ibm/370).  Then the
>> above code could well result in the same vale being passed to srand()
>> every time the program is called, because time() set the floating
>> point result register and the above code is looking at an integer
>> result register which untouched by time().
>
>  I don't follow that.. if time returns a float then won't the result be
>cast and truncated to an int?

Only if the compiler knows that time() returns float. The only way it
can know is if there is a declaration of time() in scope (e.g. from
including time.h). Without that the compiler must assume that time()
returns int hence will generate the code required (whixh may be none) to
convert int to unsigned, *not* float to unsigned.

With an appropriate declaration this will work if the truncated value is
representable as an unsigned int. If not the result is undefined behaviour.
On most systems floats can represent much larger values than ints.

While integer operations where the result is unsigned cannot overflow
(and produce undefined behaviour) floating point -> unsigned conversions
*can*.

>Is the following legal, by the way?
>
>        char Temp[sizeof(time_t) + sizeof(unsigned)];
>
>        time((time_t *)Temp);
>        srand(*(unsigned *)Temp);

There is no guarantee that Temp is aligned suitably for either a time_t or
an unsigned. You could overcome that using malloc.

There is no guarantee that time_t is as long as unsigned. This means that
not all bytes of *(unsigned *)Temp may be set.

C doesn't guarantee that accessing an object via an unsigned lvalue will
be aliased to accessing it via a time_t lvalue.

What you can do is something like:

     {
         time_t timeval = time(NULL);
         unsigned char *ptr = (unsigned char *)&timeval;
         unsigned seed = 0;
         int    i;

         for (i = 0; i < sizeof timeval; i++)
              seed = (seed + ptr[i]) * N;

         srand(seed);
     }

where N can be chosen to 'diffuse' small changes in timeval throughout
the seed. It should be at least odd so that lower order bits are affected.
Making it prime may or may not produce any benefit but it shouldn't do
any harm.

--
-----------------------------------------
Lawrence Kirby | [e-mail]
Wilts, England | [e-mail]
-----------------------------------------