On Thu, Sep 17, 2009 at 8:39 AM, Andrew Garrett <agarrett(a)wikimedia.org> wrote:
Why do you want to scrap this, and embark on a
several-month-long
development effort to reimplement it with more complexity? There
certainly isn't any significant technical benefit in it, besides the
fact that it seems a little cleaner from the outside.
If, for instance, all category links were stored only in
categorylinks, we wouldn't have the problem of denormalization.
categorylinks isn't fully reliably right now because it duplicates
information from the page text, and might fall out of sync. Storing
things in a single blob instead of broken into tables -- or
duplicating data in multiple places -- is a violation of the
relational model and tends to hurt correctness, performance, or both.