Dieter Maurer wrote:
Jim Fulton wrote at 2005-5-27 08:29 -0400:


Then, we probably do something wrong...

That's always a possibility.  I think what we are doing is
pretty reasonable.  Perhaps you have other suggestions.

I think we need more control over what modifications trigger
what reindexing events.

I am not yet sure about the best (or even a good) approach.


Even computing the value for a text index (without any change
to the index itself) can be very expensive: it may
include expensive fetching of a large object,
an expensive conversion (text extraction), expensive splitting
and comparison to what is currently indexes.

Perhaps. It depends a lot on the application.

I suggest that, if this optimization is important, it might
be much easier and cleaner to make text extracttion and comparison
cheap, rather than, trying to solve the problem with a more complex
event model.

You cannot make text extraction cheap (as it handles potentially large

You can't make it cheap in all applications.  For most applications,
text extraction and comparison is very cheap.

I'm guessing that you are refering to indexing large (book size)
documents.  I would argue that this is pretty specialized.

You could make comparison cheap -- e.g. by storing last modification
dates and comparing them.
But, I fear, you would just move the problem to when changing the
modification date.

I think this is a nice solution for those special cases where text
extraction  is expensive.  The nice thing about this solution is that
it involves a contract between the content and the index without
complicating the event framework.

I think it would be very difficult to come up with rules
for deciding which events might effect a text value and which would not.
For example, I can easily imagine objects who's searchable text
depends on their workflow state.

Indeed, such objects are easily imaginable.
But usually, it is not the case.

And it is usually not the case that text extraction is expensive.

The problem is obviously difficult -- not solvable with
a trivial event model and trivial reindexing dispatching.



Jim Fulton           mailto:[EMAIL PROTECTED]       Python Powered!
CTO                  (540) 361-1714  
Zope Corporation
Zope3-dev mailing list

Reply via email to