https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124256
Jonathan Wakely <redi at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
Resolution|--- |INVALID
--- Comment #4 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Carlos Galvez from comment #1)
> Possibly this commit is the culprit:
> https://github.com/gcc-mirror/gcc/commit/
> 866bc8a9214b1d00402fdbdcb976688a95c98f65
>
> Adding -D_GLIBCXX_USE_OLD_GENERATE_CANONICAL reverts the previous behavior.
Yes, and this is clearly documented in the draft release notes:
https://gcc.gnu.org/gcc-16/changes.html#libstdcxx
(In reply to Carlos Galvez from comment #3)
> I understand that only the *engine* has reproducibility guarantees, but they
> can't be used without a distribution, or?
It depends whether you just want random bits, or random numbers that are
distributed according to some function.
> In other words is there other way
> I can write code to generate random numbers and achieve reproducibility,
> without relying on distributions?
You can write your own distribution.