Re: Attn Packagers

["Andreas J. Guelzow" <aguelzow math concordia ab ca>]

The Mozilla project, for example, has its main builds ignoring some of 
the sub project, but also builds that include them (eg. Mozilla-mathml). 
Wouldn't it similarly make sense to package say a standard 
gnumeric(without bonobo) as well as a gnumeric-with-bonobo (requiring 
bonobo >=1.0.9) and let the user choose which package to install. Norman 
(and many others) would obviously choose the former, while others are 
likely to use the latter. (IMHO gnumeric with bonobo is much more stable 
then we may give it credit. We are using gnumeric in a production 
environment as the spreadsheet of use in our intro stats course and our 
students in this course are probably as unsophisticated a user as you 
are likely to encounter.  We used to use StarCalc but our students 
managed to lock it up/crash it too frequently.)

Andreas

Norman Reitzel wrote:

> Gregory:
>
> Jody's concerns about bug load - particularly from people who want
> to install from RPM as opposed to build from source - is a valid
> point, but I have a view from the opposite end.  I build from the
> CVS tree, and from my point of view as a user of Gnumeric, I find
> that bonobo itself is not ready for prime time.  As a developer, I
> know that this circumstance would like to have maximum exposure to
> shorten the debug time, but from a user standpoint, this is akin to
> Microsoft' "Let the User be the Beta" attitude.  Right now, and for
> the near future, Gnumeric is far more reliable if one can just
> leave bonobo out of the build, period.  I know that for me - at
> least for now - the first nut that requires bonobo will be the one
> after the last one that I download and build.
>
> I have been working with *nixes as a developer since 1979 (v6), and
> I think that the KISS philosophy is the magic bullet.  Keep
> features as independant as possible, and debug the pieces
> thoroughly.  I'm philosophically - and vehemently - against the
> mixmaster method, where there is one package that contains
> everything.  Among other things, building features as separate
> packages requires that the interface be well defined, thoroughly
> documented, and static, which is an extremely desirable thing from
> the standpoint of reliability.
>
> -- Norm Reitzel
>         Blue Water Ventures
>
>
> On Wed, 10 Oct 2001 23:58:14 -0400, Jody Goldberg wrote:
>
> > On Tue, Oct 09, 2001 at 09:49:06PM -0700, Gregory Leblanc wrote:
> >
> > > Is the bonobo support in the main gnumeric executable, or in separate
> > > files?  If it's in separate files, it's easy to create a gnumeric and
> > > gnumeric-bonobo RPMs, allowing end users to not install the bonobo
> > > support if they don't want it.  
> > > If it's not possible to separate them this way, then I'm all for putting
> > > a BuildRequires:  bonobo >= 1.0.9 into the spec file, and getting people
> > > to do some more thorough bonobo testing.
> > Bonobo is mostly isolated in the build, but it does change the
> > compilation.  However, I am against having a Requires for bonobo for
> > the same reason that I'd like to avoid Requiring guile-1.5 or python
> > or guppi.  There are already too many complaints about Gnumeric's
> > requirements.  The main people having problems are the people who
> > are no longer newbies, but have yet to reach grand master status.
> > They do not rely completely on packages, but still use the canned
> > rpm spec to build.  All sorts of problems arise trying to get the
> > various dependencies installed.  I would much prefer to keep the
> > list (and my bug load) down by requiring the absolute minimum number
> > of packages in the default spec.

--
Prof. Dr. Andreas J. Guelzow                    
Assoc. Prof of Mathematics
Concordia University College of Alberta
http://www.math.concordia.ab.ca/aguelzow