Austin Ziegler
9/19/2003 2:58:00 AM
On Fri, 19 Sep 2003 10:37:05 +0900, Gavin Sinclair wrote:
> On Friday, September 19, 2003, 7:44:07 AM, Austin wrote (in part):
>> [Transaction::Simple] does not reuse the same object identifiers
>> (unless the object responds to #replace), so anything that depends on
>> #__id__ will break on a rewind or abort.
> Does it concern you at all that an object might implement #replace but
> with entirely different semantics? I know it's not that likely, but it's
> worth considering.
A bit. I believe that this would be fixed if, like, #__id__, we had
#__replace__. If I don't test for #replace, though, I would have to do
specific tests for String, Hash, and Array (and other items that support
#replace). Maybe I'll test two things: respond_to?(:replace) and the #arity
of the method (e.g., 1).
Perhaps this would make a good RCR?
Apparently, there's another method in 1.8 that I can use to reconstruct the
object without messing up the #__id__ -- I just can't remember it offhand,
and I haven't had time to look it up in the last two months of methods. If
anyone can remember it, I'd appreciate it.
I am going to be making a change to Transaction::Simple based on what
Madeleine does, by offering alternatives to Marshal in the same way. It's a
good enhancement, even if Marshal probably provides the most efficient way
of handling the data.
-austin
--
austin ziegler * austin@halostatue.ca * Toronto, ON, Canada
software designer * pragmatic programmer * 2003.09.18
* 22.41.30