In SciSmalltalk we added some more random generators but people should have a look....

Stef


Le 22/10/15 18:21, Nicolas Cellier a écrit :
For these reasons, I often prefer the Marsaglia's KISS

2015-10-22 17:57 GMT+02:00 Henrik Johansen <[email protected] <mailto:[email protected]>>:


    > On 22 Oct 2015, at 2:49 , Tobias Pape <[email protected]
    <mailto:[email protected]>> wrote:
    >
    >
    >
    > On 22.10.2015, at 14:37, David T. Lewis <[email protected]
    <mailto:[email protected]>> wrote:
    >
    >> On Thu, Oct 22, 2015 at 06:28:09AM +0000,
    [email protected] <mailto:[email protected]> wrote:
    >>> Tobias Pape uploaded a new version of Network to project The
    Trunk:
    >>> http://source.squeak.org/trunk/Network-topa.166.mcz
    >>>
    >>> ==================== Summary ====================
    >>>
    >>> Name: Network-topa.166
    >>> Author: topa
    >>> Time: 21 October 2015, 8:26:14.656 pm
    >>> UUID: d6d9910f-fa67-4c69-9a89-030c81233e90
    >>> Ancestors: Network-topa.165
    >>>
    >>> UUIDGenerator
    >>> - Use new Random>>#nextBytes:into:startingAt: (needs
    Kernel-ul.960) for even more speed
    >>> - Fix UUIDGenerator class>>#initialize to correctly register
    at startup
    >>> - Do not reset default on startup but rather reseed TheRandom
    >>>
    >>> Timings improved:  '1,190,000 per second. 842 nanoseconds per
    run.' (0.25 times slower than primitive version)
    >>>
    >>
    >> This is outstanding. The UUID plugin has long been a source of
    problems in
    >> the VM, partly due to issues in various Linux distributions.
    >>
    >> http://bugs.squeak.org/view.php?id=7358
    >>
    >> Making the plugin unnecessary is a big improvement.
    >
    > Thats what I thought.
    > Probably we want to present this (Mersenne PRNG + new UUID gen) to
    > the Pharo core developers (cc) so that we can just remove the
    UUID plugin
    > from the VMs (cc vm-dev)?
    >
    > Best regards
    >       -Tobias

    In the release version of Squeak 5.0, the Mersenne Twister is
    initialized using a single seed restricted to 32bit.
    Has this been changed since to use a wider ranged value?
    Otherwise, you run into problems with potential collisions from
    separate image starts rather too fast for UUID uses... (IMHO)

    Cheers,
    Henry



Reply via email to