[Zope-dev] Re: [Zope2] ZODB load optimization ineffective due to ExtensionClass flaw

[Jim Fulton]

Dieter Maurer wrote:
> The ZODB allows to include the class (a reference to it)
> into persistent references in order to avoid loading the object
> state when a ghost object needs
> to be created for the persistent reference.
> 
> For this to work, the class must not require arguments to "__new__".
> The ZODB checks whether the class defines a "__getnewargs__" method.
> It it does, the ZODB concludes that its "__new__" might need arguments
> and does not try to optimize.
> 
> Unfortunately, "ExtensionClass.Base" defines "__getnewargs__".
> It is the base class for all Zope2 classes. Zope2 persistent
> references are not optimized...
> 
> 
> An analysis of our application revealed that the missed
> optimization opportunity increases the amount of storage loads
> by a factor of 4 to 5. Storage loads are often the dominating
> factor -- especially for searches.
> 
> 
> @Jim:
> Is the "__getnewargs__" definition in "ExtensionClass.Base" vital?
> It returns the empty tuple. Thus, there seems to be a chance to
> get rid of it.
> 
> I know that some code (especially in "ExtensionClass/picke/pickle.c")
> needs to be changed as it depends on the existence of "__getnewargs__".
> 
> If you think, there is a chance to get rid of "__getnewargs__",
> I can do the work and put the changes into the collector.
> 
> Zope2 might become a bit faster...

I'd have to review this, but I'm pretty sure that we dealt with this
issue, but I could be mistaken.

Jim

-- 
Jim Fulton           mailto:jim at zope.com       Python Powered!
CTO                  (540) 361-1714            http://www.python.org
Zope Corporation     http://www.zope.com       http://www.zope.org