Re: Class related regressions in GNU Octave 4.0

[Oliver Heimlich]

On 12.04.2015 15:06, Mike Miller wrote:
> On Sun, Apr 12, 2015 at 10:55:42 +0200, Oliver Heimlich wrote:
> > On 26.03.2015 23:09, Jordi Gutiérrez Hermoso wrote:
> > > On Thu, 2015-03-26 at 22:31 +0100, Oliver Heimlich wrote:
> > > > I guess that now it is too late and the regressions are not going to
> > > > be fixed until the final 4.0 release and therefore break some
> > > > functionality for a certain time?!
> > >
> > > I think we ought to consider regressions release-critical. I'll take a
> > > look at these three bugs during the weekend.
> > >
> > > - Jordi G. H.
> >
> > Jordi, I did you check the regressions already? Should I put them on the Bug
> > Fix List for the release [1]?
>
> I looked your list over. I haven't verified whether they are regressions
> or not, or whether they are critical for the release or not. But it's
> worth noting that they all only apply to old-style classes that override
> the "size" function.

I have verified that #44334, #44498, and #43925 all are regressions, 
which have been working in 3.8.2 and are broken in 4.0.


In order to not break some functionality in 4.0 in the interval package, 
I would have to provide three workaround methods. Here is the big 
question (taken from #44498), which affects all these bugs:

What is the desired behaviour?
1) A user-defined class must override all methods together to get a 
consistent behaviour: size, isvector, isscalar, and isequal. Any builtin 
functions will assume that the object's size is always [1 1].
2) A user-defined class must override size only. The builtin functions 
will respect the overriden size method and produce a consistent 
behaviour for the user (like they did in version 3.8.2).