www/64120: /mnt/www/en/ports/needs to be re-run

[Mark Linimon]

Alexey Zelkin <phantom at FreeBSD.org.ua> wrote:

> IMO, The proper solution is to leave it as is.  Even using INDEX
> www/en/ports is most time and cpu consuming part (except *maybe*
> doc/) of www tree. We don't need to get www builder machine into
> yet another ports index tester.

Believe me, I am *very* aware of how long it takes to build INDEX --
I wrote a lot of code to extract that information on an incremental
basis, because otherwise portsmon would never have correct
information.

So, while I don't think anyone wants that, the fact of the matter is
that the ports webpage has not reflected reality for just over 3 weeks.
This makes these pages something worse than useless -- they are
flat-out misleading.  If they can't be updated to reflect a category
split, then what event is sufficient to trigger an update?

Perhaps we're just trying to solve the wrong problem.  Bear with
me here.

Perhaps it's entirely fair to step back and ask what it is that we
expect the ports web-pages to be.  If it's just a general introduction
to the ports system, fine.  If it's supposed to be a canonical reference
to the ports tree, it's not serving that purpose in its present form.

The first observation is that if the pages are only updated once every
once in a while, then the most useless data is the dependency
information, which can change on a daily basis.  We should determine
whether anyone really _relies_ on that information being on that
page.  If so, then we should investigate something like Kris is talking
about, bringing over an INDEX file as a cron job or something.  But
if not, let's just delete those lines altogether.  This would not only
save browser-display time, but also enable us to consider getting
INDEX out of the equation altogether.

By doing so, we could create a cron job that would extract the more
minimal information directly from the ports tree (with make -V) and use
that to rebuild the pages.  While this would not be instantaneous, it
would IMHO be at least twice as fast.  A cron job could build that,
reduced, file, once a day, and then regenerate the web pages.

If that's still too much of a load, my preference would be to delete
the detail pages altogether rather than having them always be out
of date, and just refer people to e.g. FreshPorts, which updates more
on the order of once an hour.

Or perhaps there's some middle-ground here which is not occuring
to me?

Executive summary: I'm pretty allergic to INDEX, but the web pages
ought to be more "right" than they are right now.

mcl