[Vm-dev] New Cog VMs available...

[David T. Lewis]

On Wed, Jan 09, 2013 at 10:20:45PM +0100, Nicolas Cellier wrote:
>  
> 2013/1/9 Frank Shearar <frank.shearar at gmail.com>:
> >
> > On 9 January 2013 02:38, David T. Lewis <lewis at mail.msen.com> wrote:
> >>
> >> On Tue, Jan 08, 2013 at 09:17:52PM -0500, David T. Lewis wrote:
> >>> On Wed, Jan 09, 2013 at 02:28:58AM +0100, Nicolas Cellier wrote:
> >>> >
> >>> > Great, but now everybody could see my bug if she tries to connect
> >>> > primitive 20...
> >>> >
> >>> > http://code.google.com/p/cog/issues/detail?id=117
> >>>
> >>>
> >>> Your fix from http://code.google.com/p/cog/issues/detail?id=117 is in
> >>> VMMaker-dtl.296 for the interpreter VM.
> >>>
> >>
> >> And I note for the record that Eliot and I applied your patch to the two
> >> VMM branches simultaneously within a two minute time window. How's that
> >> for some good teamwork?
> >>
> >> :)
> >>
> >> Dave
> >
> > It's not all bad news:
> > http://squeakci.org/job/SqueakTrunk/116/testReport/ shows Nicolas' new
> > primitives working on Cog r.2664!
> >
> > frank
> 
> Ah, yes, but the short-lived bug can only show up when using a certain
> conjunction of VM and image
> - old VM are OK, primitive 20 does primitiveFail
> - new VM are OK, bug corrected
> - only the interim VM (COG 2662 and classical 4.10.9) can cause a bug,
> but only if image would be using <primitive: 20>
> Until now, trunk image doesn't, this change is still in the inbox, so
> I'm quite sure CI is safe.
> 
> Anyway, if we push inbox change in trunk, and some unlucky guy use the
> wrong VM, #rem: would be false only if
> - dividand is a large integer with magnitude fitting in 64 bits,
> - and result is different from zero
> - and divider is a negative, small or large integer, with magnitude
> fitting in 64 bits too.
> 
> Since rem: is used principally in Fraction arithmetic (but more than
> often the divider is positive), and some dateAndTime handling (case of
> negative divider does not seem the usual case neither), seeing a
> consequence of the bug would be a great case of bad luck !
> 
> So I think we can soon proceed with the change :)

+1

And thanks for all of these numerical method and VM improvements :)

Dave

> 
> Nicolas