Re: findutils-4.2.15 -- "make check" fails + request + 2 patches

[Peter Breitenlohner]

On Mon, 31 Jan 2005, James Youngman wrote:

> On Mon, Jan 31, 2005 at 02:26:36PM +0100, Peter Breitenlohner wrote:
>
> > 2. I'd like to ask for an enhancement in updatedb as per the attached patch
> > "path-01-pathfilter".
>
> This is a good idea.  However the change I actually plan to make
> (eventually) is one that will properly fix the problem outlined by
> Bruno Haible at
> http://savannah.gnu.org/bugs/?func=detailitem&item_id=8599.  That
> really needs a more general solution than this one, since any given
> NFS server may have several clients, each of which mount the
> filesystem in a different location.
>
> Therefore, although your idea is a good one, I'm not going to apply
> your patch.  However, if you have some ideas as to how to solve the
> problem in the more general case, I'd encourage you to work with Bruno
> to find a solution.  I've copied Bruno on this email.

Maybe I should first clarify our setup. Our patch is used in a situation
where the NFS filesystems are automounted due to shell startup scripts
(/etc/profile resp. /etc/csh.login) which append certain directories on them
to PATH and the corresponding locatedb's to LOCATEPATH (and info dir's to
INFOPATH). The filesystems are always mounted at fixed locations and
updateddb is run at the server (thus avoiding some of the problems addressed
by Bruno). In this situation our patch is fine but it certainly doesn't
cover more general situations.

For the more general solution sought by Bruno I have these ideas (which in
part repeat Bruno's arguments):

1. A separate locatedb for each NFS filesystem is maintained by (updatedb is
run on) the server for reasons of efficiency as well as permissions. There
might even be several such databases, one for the complete exported
filesystem and others for certain subtrees.

A separate updatedb-exports script might be useful but an ordinary cron job
could as well do that. I'm not sure that one should automatically create
locatedb's based on the output of "showmount -e".

2. On clients these locatedb's are added to LOCATEPATH by a suitable
mechanism (after all that's what LOCATEPATH is good for). This could also be
a virtual addition based on the contents of a some configuration file (the
man_db package does a very good job in this direction for man hierarchies).

3.2. When running updatedb on the server (with a special option) only the
relative paths are stored in the database. These paths are relative to the
directory containing the database, without leading slash. Of course,
updatedb should then detect an attempt to include files outside this subtree
and complain.

For such a database locate automatically prepends the directory containing
the database.

This would even cover the case of mounting a subtree of an exported
filesystem, provided there is a database at the root of this subtree.

regards
Peter Breitenlohner <address@hidden>