[R] Destructor for S4 objects?

[Prof Brian Ripley]

On Sat, 8 Jan 2005, McGehee, Robert wrote:

> Adam, I'm about to encounter a very similar problem, so would be curious
> if you find a good solution, although your thoughts with reg.finalizer
> look promising. The thoughts I had sketched out for tackling this
> problem:
>
> 1) Have all ancillary resources freed when the package is detached. The
> new() function would create the object, allocate additional resources,
> and store some kind of handle for these resources in the package
> environment as a cleanup object. The cleanup object is then referenced
> when .Last.lib() is run, and all ancillary resources are deallocated.
>
> 2) Write methods for rm() / remove() that cleanup the resources at the
> same time as removing the object from the environment. A handle to these
> resources would have to be stored either with the object like you
> suggested, or in a cleanup object (like #1) that stores the handles of
> every object needing cleanup.
>
> Conceptually, #2 seems very similar to your suggestion of using
> reg.finalizer. However, it had seemed to make more sense (to me) to
> deallocate resources at rm() time rather than gc() time, as I had
> thought that garbage collection happened at R's convenience, rather than
> when the user explicitly removes the object from the environment (which
> is presumably garbage collected sometime later).

rm() does not remove the object, just mark it as no longer in use by 
removing the symbol.  If you want to release the storage (to R, at least), 
you need to call gc().

> However, any information about the mechanics of removal and garbage 
> collection, and the correct way to cleanup objects is greatly 
> appreciated, as I had never quite understood when reg.finalizer should 
> be used.
>
> As a corollary, I have some concern that if R quits, then the garbage
> collector never runs (and rm() and .Last.lib() as well). The resources

There are ways around that, as finalizers can be told to run then.
See the examples in the RODBC package.

OTOH, .Last.lib is run only when a package is explicitly detached, and not 
at the end of the session.

> in my case are database objects that may need to be altered or deleted
> after the corresponding R object is removed (or R quits), and I have yet
> to find any solution that runs at quit time (although maybe rewriting
> the q() function might not be a bad idea, although I'm hesitant to
> rewrite base code). Let me know if you come up with anything more
> elegant.

Quitting does not necssarily happen through q() (EOF is another way): but 
you can use .Last.

-- 
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