cloning and Storable Re: [Bioperl-l] bugs on branch; tests on main trunk

[Ewan Birney]

On Wed, 3 Sep 2003, Heikki Lehvaslaiho wrote:

> I've removed the dependency for Storable. Storable is still used if it
> is installed. Local code can clone everything except circular
> references. If someone knows how to do it, I'd be happy to receive help.
> Not having it here does not really matter because the the main use of
> the clone method is to allow in-memory creation of a new enzyme based on
> an existing one.
>
> The clone code is written in very general way and should be able to deep
> copy any in-memory objects. If you need to add a clone method your own
> classes, copy from there. Ewan feels strongly that deep cloning is too
> prone to errors to be a general property of bioperl objects, so better
> not add this into Bio::Root::Root, although it would be handy.

I am willing to be overruled if there are alot of people who agree with
Heikki, but clone() methods are, in my view, just promise something (the
ability to correctly make a independent copy of all connected objects)
without being able to deliver.


The problem is with objects that either have eccentric memory layouts
(such as bound XS code; not that we have many of these) or have implicit
singleton style characteristics (eg, adaptors to databases which have
session information). a clone() which naively attempts to just in-memory
copy everything with truely fall over on teh first case and probably cause
a complex problem on the second case. Remember that these objects may not
be the top level ones, but rather be held onto in the object graph.


Furthermore, I rarely see the need for clone; in most systems just
reference passing is fine, and clone() is at best used as a shorthand for
a specific constructor, (which is what it is doing in restriction enzyme)
where I would argue the "full memory copy" is really a shorthand for
"build me a new RE with precisely the same attributes" which can then be
modified.


So, I would argue that clone() on RE's is better written as a type of
new option

  $new_re = new RestrictionEnzyme ( -template => $old_re);


and we don't have clone on the Root::Object. Current Heikki is swayed
enough by this argument to keep the clone() method specific to RE's.


If Jason/Lincoln/Hilmar all (or mostly...) liked clone() on the Root
object then I'd have to conceed