ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Mon, May 16, 2011 at 03:40:32PM +0200, Martin Sandve Alnæs wrote:
> On 16 May 2011 15:00, Anders Logg <logg@xxxxxxxxx> wrote:
> > On Mon, May 16, 2011 at 02:53:03PM +0200, Marie E. Rognes wrote:
> >>
> >>
> >> On 16. mai 2011, at 13:07, Anders Logg <logg@xxxxxxxxx> wrote:
> >>
> >> > The suggested plan is as follows:
> >> >
> >> > 1. Release 0.9.11    now
> >> > 2. Release 0.9.12    2011-05-31
> >> > 3. Release 1.0.0-rc1 2011-06-14
> >> >
> >> > No new features should be added after this point and all known
> >> > bugs (unless we decide to push them to post 1.0) should have
> >> > been fixed.
> >> >
> >> > 4. Release rc2, rc3, ... if necessary
> >> > 5. Release 1.0.0 when ready
> >> >
> >> > Between 3 and 4, all depending packages (like cbc.solve etc) have a
> >> > chance to make updates to any interface changes and everyone is
> >> > encouraged to test the release candidate.
> >> >
> >>
> >> Sounds good to me, though "when ready" is not a very definite term.
> >
> > Def: when we have fixed all bugs found during a testing window of
> > length to be determined below...
> >
> >> > How long should this window be? I'd like a very short window (like a
> >> > week). Opinions?
> >>
> >> A week is fine as long as one knows approximately when this week is going to be :-)
> >
> > At the release of 1.0.0-rc1 scheduled for 2011-06-14.
>
> So the rcN run should be like this (starting 2011-06-14):
> 1) Release 1.0.0-rcN
> 2) Wait for a week (fix bugs as they are reported)
> 3) When bug list is empty:
>     3.1) If bugs were found, goto 1)
>     3.2) If no bugs were found, break
> 4) Release 1.0.0

Yes, unless this drags on forever... We might consider inserting some
convergence criterion into that loop that does not require convergence
to zero. Perhaps something like this:

  while len(bugs) > 2:
    ...

:-)

> With the exception that some bugs may be deliberately marked as to be
> delayed after 1.0.0.

Yes.

> A week is short and will give little chance of any significant testing.
> But if the next month can be considered a kind of beta period, I guess
> we'll be fine. No major changes coming up now, right?

Yes, I consider 0.9.x as the beta series, especially starting from 0.9.8.

There are no major changes planned. All plans are listed here:

https://launchpad.net/dolfin/+milestone/0.9.11
https://launchpad.net/dolfin/+milestone/0.9.12
https://launchpad.net/dolfin/+milestone/1.0.0-rc1

> What about the book? I think it would be best to adapt the code examples
> to rcN before 1.0.0 is released, in case they trigger bugs that must be fixed.

Yes, all code examples should be adapted to work with rc-1 once it is
released except for the chapters [hoffman-1/2] where we have made an
exception to stay on the 0.8-something branch.

--
Anders