Question about __getattribute__/__getattr__

[Joao Prado Maia]

On Tue, 1 Oct 2002, holger krekel wrote:

> For caching a function's results why do you need __getattr__?
> 

Because with __getattr__ it will work transparently, even if I add new 
methods to the cached class, or change the name of an existing one. 
Instead of having to create one method for each method in the cached 
class, I just use __getattr__ to receive all calls and apropriately cache 
the results.


> It seems you really need another caching policy than the prescribed
> memoize pattern.  For example, you may implement a dictish object that evicts
> old cached values by means of counting accesses, oldest access times
> or whatever.  I still see no use for expecting __getattr_ to know about
> the arguments.  Not?
> 

A dictionary approach is not good enough for my needs, as it will 
eventually hold too much data in memory. As I explained before, saving the 
method return values in md5()'ed named files is the best solution I can 
think of.

But going back to the subject, I need the arguments on __getattr__ so I 
can create unique filenames for the method call signatures. That is, 
calling a method like this:

t.real_method('spam')

would create a filename as:

filename = md5.md5().digest('%%' % ('real_method', 'spam'))

While the following method call:

t.real_method('eggs')

would create a filename as:

filename = md5.md5().digest('%%' % ('real_method', 'eggs'))

This way you can see that both method calls would be cached in different 
files. To see if the method call was already cached, I just need to see if 
the file exists.

I hope this explained a little better the rationale behind the question. 
:)

Cheers,
Joao