https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88935
Andrea Griffini <agriff at tin dot it> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |agriff at tin dot it
--- Comment #12 from Andrea Griffini <agriff at tin dot it> ---
Even assuming rand() were generating hardware random numbers (not allowed by
the standard because of srand, obviously), gcc would still be broken and
performing a terrible random_shuffle with ranges larger than 64k elements
(indeed non-uniformity becomes evident even at much smaller ranges).
Mingw's rand() numbers are decent enough (and the period is not just 2**16, for
example) but they're only 16 bit. This is allowed.
The gcc bug is that it uses rand() (that can be just 16) bit to pick a random
number between 0 and the size of the range even when the range is much bigger
than 65536. Using rand() to feed data into a xorshift64 or something similar
only for large ranges would fix, and would also keep shuffle of small ranges
backward compatible (shuffling elements in small ranges exactly as current
implementation does if starting from the same random seed).
I've somewhere an half backed patch for doing that but quite frankly I don't
know if I should polish it and submit here as, in all honesty, seems that in
this case the maintainers simply don't want to have this bug fixed, for reasons
that I really cannot understand.
