Random Generate & Thread Safety

[Derek Clegg]

On Nov 17, 2011, at 12:11 PM, Torbjorn Granlund wrote:

> "james at jamesbb.co.uk" <james at jamesbb.co.uk> writes:
> 
> To achieve thread-safety, is it sufficient to keep the gmp_randstate_t structure in thread-local storage?
> 
> No.  Just like for *any* variable in GMP or elsewhere, gmp_randstate_t cannot safely be written by two threads, or read by one thread and written by another, without exclusion mechanisms.  There is nothing magic about gmp_randstate_t that would make it different.

If it's in thread-local storage, how can it be written by two threads, or read by one thread and written by another?

Derek