[Bioc-devel] best practices for exporting/importing methods and generics?

[Martin Morgan]

On 12/15/2011 05:23 AM, Michael Lawrence wrote:
> Hi guys,
>
> An issue has come up since the move of many generic functions to
> BiocGenerics.
>
> Often, a package will implicitly import a generic, along with any methods,
> from a package using importMethodsFrom(). That of course only works if the
> package in question exports the generic. This can happen implicitly via
> exportMethods() (if the package defines the generic) or explicitly via
> export(). If the package is using exportMethods(), then it will stop
> exporting the generic when the generic is pushed up the package stack. To
> be defensive, the importing package could importFrom() the generic, which
> would then give an error if the package has been moved. Otherwise, R will
> not throw a warning if methods are imported, but not the generic (it
> probably should, right?).
>
> So the question is whether the burden should be on the exporter (to export
> the generic explicitly) or the importer (to import the generic explicitly).
> In my opinion, it should be on the exporter, because then the importer
> never needs to change the NAMESPACE when things move around. Also, each
> exporter probably has multiple importers (dependencies). Really, the
> exporter only needs to export() the generic to also export the methods, but
> as the R docs say, it may be cleaner to also have exportMethods() when
> methods are defined on the generic.
>
> The case could be made that this is a problem with R. The exportMethods()
> directive should always export the generic, even if the generic is defined
> in another package, i.e., the chaining happens automatically.
>
> Here is an example:
>
> IRanges defined the unique() generic and exported it via exportMethods(),
> but it was moved to BiocGenerics. Now IRanges no longer exports the
> unique() generic, so a package like rtracklayer that has
> importMethodsFrom(IRanges, unique) will not get the generic and code
> breaks. To fix this, either rtracklayer needs to importFrom(BiocGenerics,
> unique) or IRanges needs to export(unique). Which one should happen?

Taking the first fix, does rtracklayer need both

   importFrom(BiocGenerics, unique)
   importMethodsFrom(IRanges, unique)

or is importFrom(BiocGenerics, unique) sufficient? I would guess that, 
in general, both importFrom() and importMethodsFrom() are necessary.

Martin

>
> The workspace has not been affected by this, since IRanges has BiocGenerics
> in Depends.
>
> Michael
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> Bioc-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/bioc-devel


-- 
Computational Biology
Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N. PO Box 19024 Seattle, WA 98109

Location: M1-B861
Telephone: 206 667-2793