Re: rand is not ISO C compliant in Cygwin

[Bruno Haible <bruno@clisp.org>]

Corinna Vinschen wrote:
> I took a look into POSIX and I'm a bit puzzled now.  From
> https://pubs.opengroup.org/onlinepubs/9699919799/functions/rand.html

Part of the confusion is that POSIX and ISO C have slightly different
wording. This POSIX page says:
   "The functionality described on this reference page is aligned
    with the ISO C standard. Any conflict between the requirements
    described here and the ISO C standard is unintentional. This
    volume of POSIX.1-2017 defers to the ISO C standard."

In ISO C 99 § 7.20.2, the only relevant sentence is:

  "The srand function uses the argument as a seed for a new sequence
   of pseudo-random numbers to be returned by subsequent calls to rand.
   If srand is then called with the same seed value, the sequence of
   pseudo-random numbers shall be repeated."

In ISO C 11 § 7.22.2 and ISO C 17 § 7.22.2, additionally two sentences
were inserted:

  "The rand function is not required to avoid data races with other
   calls to pseudo-random sequence generation functions."

  "The srand function is not required to avoid data races with other
   calls to pseudo-random sequence generation functions."

ISO C 23 (which is still is draft state, but compared to the draft
https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3054.pdf I cannot
see any change regarding rand() in the changes summary
https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3148.doc) has the
same wording.

POSIX does not have these two sentences, but instead has:

  "The rand() function need not be thread-safe."

> RATIONAL
> 
>   The ISO C standard rand() and srand() functions allow per-process
>                                                   ^^^^^ (not requires)
> 
>   pseudo-random streams shared by all threads.

Indeed, "requires" would fit better here, IMO, because the texts of
both ISO C and POSIX have multithreading in mind and still talk about
"subsequent calls to rand" — which makes a reference to time, but not
to threads.

> Ok, so, *iff* rand/srand share per-process state, then they have to
> use locking to prevent MT interference.

... if the implementor wants to prevent MT interference (which both
ISO C and POSIX allows).

> POSIX continues:
> 
>   With regard to rand(), there are two different behaviors that may be
>   wanted in a multi-threaded program:
> 
>   1. A single per-process sequence of pseudo-random numbers that is
>      shared by all threads that call rand()
> 
>   2. A different sequence of pseudo-random numbers for each thread that
>      calls rand()
>     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

This paragraph continues after the two items:
   "This is provided by the modified thread-safe function based on whether
    the seed value is global to the entire process or local to each thread."

My understanding of this paragraph is:
  - If an application wants 1., they can use rand_r with SEED pointing
    to a global variable.
  - If an application wants 2., they can use rand_r with SEED pointing
    to a per-thread variable.

> I read this as the newlib technique being one way of correctly
> implementing rand/srand, no?

I don't think so. The critical sentence is the one with
"subsequent calls to rand".

Bruno