dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, Mar 22, 2007 at 02:22:55PM +0100, Johan Jansson wrote:
> On Thu, Mar 22, 2007 at 01:17:48PM +0100, Anders Logg wrote:
> > On Thu, Mar 22, 2007 at 11:49:03AM +0100, Johan Jansson wrote:
> > > On Thu, Mar 22, 2007 at 11:11:26AM +0100, Anders Logg wrote:
> > > 
> > > ...
> > > 
> > > > I don't understand this. If you want to develop against the development
> > > > version of DOLFIN, then get dolfin-dev.
> > > > 
> > > > If you get dolfin-fsi, you'll need to cope with both a changing kernel
> > > > *and* potentially unstable module development in unrelated modules.
> > > 
> > > The kernel will improve, but not break. If you develop against
> > > dolfin-dev, then after a pull you might discover that assembly no
> > > longer works (which is ok in the kernel development tip), or something
> > > more subtle. This is what we want to filter out with dolfin-fsi.
> > > 
> > >   Johan
> > 
> > How will the kernel improve? Will the kernel in dolfin-fsi have new
> > functionality not present in dolfin-dev that will be merged back into
> > dolfin-dev, or will dolfin-fsi pull certain things from dolfin-dev?
> > Or both?
> > 
> > /Anders
> 
> With improve, I meant that new features or fixes from dolfin-dev will
> propagate into dolfin-fsi, not that special improvements will only be
> developed in dolfin-fsi. dolfin-fsi is not a carrier of kernel
> development, it's simply there for people who want to develop
> modules/applications for DOLFIN but are unable to keep up with the
> development speed at the tip.
> 
> In practice, dolfin-fsi will just be a snapshot of dolfin-dev where
> someone has taken care that the snapshot doesn't happen just when a
> core function happens to be broken.
> 
>   Johan

It sounds like current DOLFIN + quality assurance, which sounds very
reasonable but also very ambitious.

But what makes it confusing is that it seems to be a mix of
"dolfin-qa" and "arbitrary work on a subset of the modules".

How about instead creating "mini releases" as often as needed
(whenever you make a snapshot of the development tree and verify that
it passes all the tests you require for it to be pulled onto the
dolfin-fsi tree)?

We could name these releases

    dolfin-x.y.z-n.tar.gz

or

    dolfin-x.y.{z+1}-rcn.tar.gz

Then these snapshots could be really useful to everyone.

But if the goal is less ambitious (to have a repository that is
tailored to your specific development model at KTH + eventually anyone
else who might be interested) then I have no opinions about the
repository.

/Anders