Package "datasets" not exporting anything on a recent R-2.15.1-patched

On 26.08.2012 20:01, Laurent Gautier wrote:

On 2012-08-26 19:27, Prof Brian Ripley wrote:

On 26/08/2012 18:20, Laurent Gautier wrote:

On 2012-08-26 19:03, Prof Brian Ripley wrote:

On 26/08/2012 17:25, Laurent Gautier wrote:

Hi,

I just stumbled on the following apparent oddity: the package
"datasets"
does not appear to export anything out of its namespace:

 > ns_datasets <- getNamespace('datasets')
 > getNamespaceExports(ns_datasets)

character(0)

Not the case with other packages (example here with "utils"):

 > ns_utils <- getNamespace('utils')
 > head(getNamespaceExports(ns_utils))

[1] "?"            ".DollarNames" "adist"        "alarm" "apropos"
[6] "aregexec"

Is this a temporary glitch, or is there something new and 
specific to
"datasets" ?

Not new, been so since R 2.14.0.

The package contains no R code: the only reason it has a namespace is
that since 2.14.0 all packages must have one.

There are other data-only packages, but not in base R.

Lazy-loaded data has not been part of the namespace for a long time:
they are directly in the package's environment.  The reason is the
namespace scoping rules: if the package's datasets were in its
namespace, they would be found before any other dataset of that name
by the package's R functions.

Thanks for the quick answer and explanations. The last paragraph above
is leaving me with the wish to understand more before I comment on it.
Would you have an example of finding a dataset in a package's 
namespace
(before a dataset with the same name in an other package further 
down in
the search path) being a problem ?

It does not happen now, but for example MASS contains both datasets
and analysis functions and when the datasets were in the namespace you
got them before other datasets (e.g. modified versions) of the same 
name.

I am seeing two ways for datasets to be used:
A- as internal data for necessary to a function in a package (e.g,
conversion tables). Not very frequent, from my limited experience.
B- as example data, used in the documentation. The most frequent usage.

For A/, the user is not expected to modify the data.

In that case, you can provide the data as an object defined within the 
/R directory, for example.

For B/, the data are typically passed to functions (to use an example
from MASS, "huber(chem)").

In that case, you do not want to have it in your NAMESPACE, since you 
just pass those data in a function call.

Uwe Ligges

I thought a bit about it, but still do no see where exactly this would
be a problem.

An other example available in R is the vector "letters" (in base). It
can be thought of as a dataset, yet it is
- in a namespace

 > "letters" %in% getNamespaceExports(getNamespace("base"))

[1] TRUE

- probably used in the two situations A/ and B/ above.

Best,

Laurent