[Rd] package 'stats' needs import directive for 1.9.0?

[Prof Brian Ripley]

On Fri, 27 Feb 2004, Paul Gilbert wrote:

> Prof Brian Ripley wrote:
> 
> >On Fri, 27 Feb 2004, Swinton, Jonathan wrote:
> >
> >From: Prof Brian Ripley <ripley at stats.ox.ac.uk>
> >Subject: Re: [Rd] package 'stats' needs import directive for 1.9.0?
> >To: "Heywood, Giles" <Giles.Heywood at CommerzbankIB.com>
> >Cc: R-devel at stat.math.ethz.ch
> >  
> >
> >>... 
> >>    
> >>
> >>>I suggest you don't try: there are already quite a few 1.9.0-specific 
> >>>versions of packages on CRAN, and 1.8.x will be history in a 
> >>>couple of 
> >>>months.
> >>>
> >>>      
> >>>
> >>I'm not sure I understand the advice here. I have also hit the same problem,
> >>and it seems reasonable to me to aspire to maintain a single package which
> >>works on both 1.8.x and 1.9.x. 
> >>    
> >>
> >
> >It is not easily possible if namespaces are involved.  We failed to 
> >anticpate that we would need version-dependent namespaces until rather 
> >late and then decided not to set up an elaborate solution when most people 
> >affected had already made separate versions for 1.9.0.
> >  
> >
> I'm confused (not for the first time).  I can see that namespaces are an 
> additional problem, but I thought the original question was about 
> defining an S4 generic "start." 

That's not what the subject line says, and the problem I was addressing is
that the NAMESPACE file needs to be different in 1.8.1 and 1.9.0-to-be.

> This  would conflict (and override?) the 
> S3 generic in base, which I don't think is allowed in 1.9.0. Until 
> recently I have been overriding the start function in base (and some 
> others) with my own S3 generic, and have just gone through a rather 
> lengthy exercise to  rename these functions and some objects and other 
> methods so that things do work in 1.9.0. (They do also work in 1.8.1.)
> 
> Will S3 and S4 generics for start co-exist? What happens to all the S3 
> methods for start if the S4 generic is loaded? Won't this break a lot of 
> code other than mine?

There should be no problem with an S4 generic, as it calls the actual
existing S3 generic as its default S4 method.  It's done for plot(), for
example.  What will not work is having a different S3 generic of the same
name in a different namespace (including in the residual non-namespace).

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595