Re: [PATCH] Implementations of random, srandom, initstate, setstate, rand, srand

[Bruno Haible]

Hi,

Richard W.M. Jones wrote:
> My take from this is we need separate modules, rand and random, is
> that right?

Yes, that's my understanding too.

> The existing random_r module is folded into random.

Or maybe 'random' needs only to depend on 'random_r'? (I haven't looked into
the details.)

> 'rand' is a new module which is advisory for those platforms that
> believe they may have a losing rand() implementation, eg. ancient
> versions of Solaris.

The question is whether that module is needed at all. People have been
warned for years that rand() does not yield more than 16 random bits.
You are providing 'random' and 'rand48' as alternatives. And packages that
need high-quality random numbers, such as for crypto, are using true
random numbers, e.g. Simon's gc-random module.

Anyway, I'm fine with a module 'rand', if it documents in which ways the
system's implementation is considered to be "losing".

Bruno