repoview in our repositories

[Michael Schwendt]

I believe the repoview pages in the Extras repository still cause some
mirrors to choke on mirroring many hundreds to thousands of regenerated
html files all few days:

  extras-tree$ find . -name \*.html|wc -l
  61450

That's how many html pages are in Extras 5+6+devel presently!

The reason is that although repoview doesn't need to create all its web
pages from scratch, it changes many of them too often due to the many
package versions in them, which point to adjacent packages.

For example:

http://wftp.tu-chemnitz.de/pub/linux/fedora-core-extras/development/i386/repodata/repoview/yap-devel-0-5.1.1-3.fc7.html

The navigation bar at the left contains links to 20 packages, which come
before and after "yap-devel" in the alphabetically sorted list of package
names. It links to web page file names, which contain not only the package
name, but also version and release. Changing any of the 20 packages
changes the web pages for all its neighbours. Plus, the navigation bar
lists not only the latest package release, but all releases (which is by
design) found in the repo metadata, although it hides the version-release
and causes confusion.

The more often packages in the neighbourhood of one package are changed,
added or removed, the more web pages are updated due to the heavy use of
package versions in them. Even deleting an old package release triggers
such a rebuild of a chain of web pages.

IMO, we would do better in the short-term if we modified repoview to
simplify the web page file names to just "%{name}.html". Then a package
page would only be updated if the package changed actually or if
the list of its 19 neighbours changed.

What is the estimated popularity of the repoview pages on our master
download site and on the mirrors? Perhaps there are even plans on
making a public interface to the package database obsolete repoview?

Comments?