[Gretl-devel] RFC: $sigma & co.

[Allin Cottrell]

On Wed, 30 Apr 2008, Sven Schreiber wrote:

> Allin Cottrell schrieb:
> > On Mon, 28 Apr 2008, Sven Schreiber wrote:
> > 
> 
> > Suppose we change something that breaks an existing package: I 
> > guess it's OK if we're explicit about that, and put up a 
> > modified version of the package that works with the new 
> > release?  (With judgment required, of course, on how often to 
> > do that sort of thing.)
> 
> In my view, yes, the bug fix may (and probably often will) 
> consist of an updated package. All I'm saying is that gretl 
> development should not adopt the attitude "if the package 
> suffers it's not my problem". So IMHO it's not (only) the 
> original author of the package who is responsible for fixing the 
> bug (=getting it working again).

Agreed.

> > > 2. For each gretl release some notes (readme) are prepared 
> > > which complement the changelog. They explain the 
> > > backwards-incompatible changes and how to solve the 
> > > associated problems.
> > 
> > Yes, we should do that.
> 
> If you want me to do that, I will need a couple of days warning 
> before an imminent release (maybe the same warning as for 
> translators)...

Thanks, but I think it's probably easiest for me or Jack to do 
that bit.

> > > 3. Another sourceforge tracker is introduced as a database 
> > > for incompatible changes, including information on when such 
> > > a change was introduced (date and version numbers) and how 
> > > to deal with it.
> > 
> > If you (or someone else) is willing to take charge of that, 
> > that's fine by me.
> 
> Ok I will see if I can set it up soon. It should probably be 
> read-only for the general public?

Yes, I'd say so.

> > > Number 1 is two-fold: the testing would need volunteers 
> > > ("package maintainers"?), but simply agreeing to the bug 
> > > policy is just a decision, not work.
> > 
> > I think it would be easier to roll such tests into the 
> > existing gretl regression suite.  I can make this available 
> > via the web.
> 
> What do you mean by that? Posting a big meta-script? But it 
> sounds nice.

It's a large collection of scripts with a Makefile system to run 
tests.  The basic test is to run the scripts and diff the output 
against known good output.  There's also a facility to test all 
the scripts for memory usage with valgrind.  There are several 
subdirectories to test specific sorts of functionality (mle, gmm, 
panel models and so on).

Allin.