[Rd] Minimal DESCRIPTION file

[Prof Brian Ripley]

On Wed, 25 Aug 2004, David Brahm wrote:

> My "g.data" contributed package fails under R-devel.  g.data creates minimal
> packages which consist only of a "DESCRIPTION" file and the "R" and "data"
> subdirectories, but .find.package() now rejects them because the DESCRIPTION
> file lacks a "Version" field.  I will fix g.data to add a "Version" field.
> 
> My question is, what else is necessary for a minimal package?  "Writing R
> Extensions" states:
>    "The `Package', `Version', `License', `Description', `Title', `Author',
>     and `Maintainer' fields are mandatory..."
> So should I really add all these (superfluous) fields, to avoid future
> rejection?

At least Package, Version and Title are not superfluous: they 
are used by library().

> (A comment inside .find.package() refers to tools:::check_package_description,
> but I could not find such a function.  

It is tools:::.check_package_description
              ^
> Is there a plan for a universal package
> validator, and if so, what are its requirements?)

It's called R CMD check, and also we have long required that packages be 
INSTALLed.  So we require valid metadata in pkg/Meta and a Built: field in 
the DESCRIPTION, since 2.0.0 will require a package to have been INSTALLed 
under 2.0.0.  (That has implications for the indices too.)

> Bigger picture, will g.data be rendered obsolete by lazy loading?

Is any one using it?  Lazy data loading in a package seems much simpler to 
me, as did direct use of delay() earlier.

-- 
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