Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 22 Oct 2004 16:14:26 -0400
From:      Paul Chvostek <paul+fbsd@it.ca>
To:        freebsd-ports@freebsd.org
Subject:   Re: ports/www is too full
Message-ID:  <20041022201425.GA36702@it.ca>
In-Reply-To: <200410221824.12294.benlutz@datacomm.ch>
References:  <20041022074529.GN10363@k7.mavetju> <41791AF7.2050009@vonostingroup.com> <200410221824.12294.benlutz@datacomm.ch>

next in thread | previous in thread | raw e-mail | index | archive | help

On Fri, Oct 22, 2004 at 06:24:08PM +0200, Benjamin Lutz wrote:
>
> Why does it matter that some dirs are rather large? Personally I'd rather
> be able to guess where a port resides (which is harder with more
> categories), than to have an "ls /usr/ports/www" fit onto a single
> screen.

Logical divisions are helpful ... which is why we have directories in
the first place.  I agree that splitting for the sake of smaller
directories is the wrong approach, but if a logical division exists in
www, as did in net and (farther back in) x11, should it be resisted
merely to keep the directories large?

At first glance, I don't see a really clear division to the ports in
www.  Edwin's initial suggestion sort of works, but I too think that's
far too many directories.  Too many things belong in multiple places,
and it would feel silly to `cd /usr/ports/*/someport` just because I
can't remember where something lives.  Even splitting into www-server
and www-client doesn't work for many ports.

So until there's a clear reason why a big directory is a problem, I'd
stick with the status quo.

>How about borrowing an idea
> from some of the knowledge databases, and using keywords to mark ports?
> Eg, instead of creating a www-server category, the apache port could be
> marked "server www". linux-opera could be market "binary browser client
> linux www" or something like that.

I really like this idea, and it's at least partly implemented now with a
number of ports (with multiple CATEGORIES; www/gatling, mail/wmbiff,
etc) but implementing this fully is a *much* larger task than merely
applying another logical division to a ports directory.

There are lots of ports that really truly belong in multiple categories;
ftp/wget and ftp/curl are excellent HTTP clients, mail/rlytest verifies
an element of an SMTP security policy, etc.  The problem is that having
the port directory named for its "primary" category is just so damned
convenient.

One solution might be to model the packages/ tree.  Store all ports in
ports/All/, then have *all* the categories broken out as their own
directories, with symlinks pointing back into ../All/portdir/.  Then you
could have multiple symlinks, easily maintained by the meta info in
ports/INDEX.

A single directory with 12000 subdirectories in it may be unruly ... and
we'd have to fix oddities like hydra, jags, replay, etc ... but it would
provide for the most flexible expansion, and the symlink tree would
provide an equivalent interface to folks comfortable with the current
setup.

Thoughts?

-- 
  Paul Chvostek                                             <paul@it.ca>
  Operations / Abuse / Whatever
  it.canada, hosting and development                   http://www.it.ca/



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20041022201425.GA36702>