Re: comparing version numbers

[Paul Kienzle]

On Mar 24, 2006, at 3:46 AM, David Bateman wrote:

> Paul Kienzle wrote:
>
> > On Mar 23, 2006, at 5:22 PM, John W. Eaton wrote:
> >
> > > I changed the version number in CVS earlier today to be 2.9.5+ and 
> > > the
> > > following test from cell2mat.m is now failing:
> > >
> > >   %!test
> > >   %! if ([1e6,1e4,1] * str2num (split (version, '.')) > 2010064)
> > >   %!   assert (cell2mat (D), F);  % crashes octave 2.1.64
> > >   %! endif
> > >
> > > Do we really need the version number test here?
> > >
> > > If there is a need to compare version numbers, then perhaps we should
> > > have a function to do it that can handle things like 2.9.5+ or
> > > 2.9.5+SVN-serial-number (or similar).
> > >
> > > Comments?
> >
> >
> > When deploying software which depends on Octave I do not
> > want to force people to upgrade.  In order to support
> > older versions there will sometimes be a need to compare
> > version numbers.  Alois' NaN and TSA packages in
> > octave-forge/extra are designed to support multiple versions
> > of octave and matlab if you want a real world example of the
> > sort of tests that are needed.*
> >
> > - Paul
> Paul,
>
> The difference here is that cell2mat.m has migrated to octave and so 
> the
> version in octave is always particular to the version of octave with
> which it was delivered. In that case it doesn't make sense to have a
> version number test like this. I agree completely with the case that
> octave-forge, and any eventual external packages might need such tests
> however, and that we need a means of testing version numbers, including
> any version numbers for CVS versions..

Agreed.  Furthermore, with the 2.1.x branch and the 2.9.x
trunk we are free to purge a lot of tests from octave-forge
as well.

- Paul