I can verify that running updatedb through
/etc/cron.daily/slocate.cron does seem to ignore PRUNEPATHS
environment variable set in /etc/updatedb.conf.

To work around this problem, I have had to modify the PRUNEPATHS
variable set in /etc/updatedb.conf to use comma-delimited paths, then
add -e ${PRUNEPATHS} as arguments to the slocate.cron script.

Forgot one step to fix, I also had to tell the slocate.cron script to
execute /etc/updatedb.conf, like FC3 does.

This bug also affects FC3 (and at least FC2 as well).

Created attachment 110848 [details]
workaround

A workaround that covers at least the PRUNEPATHS case.	I don't know 
if PRUNEFS is also a problem.

The problem appears to be that when invoked as updatedb (or with the
option -c), it reads the /etc/updatedb.conf, but it cannot parse it
well.  The /etc/cron.daily/slocate.cron implies that it executes
/etc/updatedb.conf as a script, when in fact updatedb ignores
everything executed in it.  I'd like to illustrate that with examples.

I will attach two examples of /etc/updatedb.conf, and you can try
them by doing the following:

- backup your current /etc/updatedb.conf where you can restore it later.
- note the size of the /var/lib/slocate/slocate.db file.  This is the
  size when the filesystems and paths that should be pruned are indeed
  pruned.
- For each of the following two examples of updatedb.conf, copy
  them to /etc/updatedb.conf, run /etc/cron.daily/slocate.cron, and
  note the size of the resulting /var/lib/slocate/slocate.db
- when you are done, restore your old updatedb.conf, and rerun the
  slocate.cron script to restore everything back the way it should.

Created attachment 111344 [details]
test case 1 - not intended as permanent updatedb.conf

Created attachment 111345 [details]
test case 2: not intended as a permanent updatedb.conf

Bjorn, thanks a lot for the test cases!

I knew that sourcing updatedb.conf in slocate.cron had no effect,
but updatedb.conf was working perfectly for me.
Both the problem and its cause is obvious with your examples.

I see two solutions:

- Either document in /etc/updatedb.conf exactly how the updatedb program
  parses it, and delete the ". /etc/updatedb.conf" line in the 
  /etc/cron.daily/slocate.cron file, so people are not confused

  OR (preferably, in my opinion)

- Leave the /etc/updatedb.conf as a script that exports the PRUNEFS and
  PRUNEPATHS variables, and then in /etc/cron.daily/slocate.cron execute
  the following:
       /usr/bin/slocate -u -e $PRUNEPATHS -f $PRUNEFS
  instead of the current call to /usr/bin/updatedb.  Note that the -e
  and -f parameters are to comma-delimited, so either it needs to be
  that way in /etc/updatedb.conf, or you can execute the 'sed' command
  that Devin Reade proposed:
      sed -e 's/  */,/g;'`

I prefer actually fixing the updatedb parser :)

slocate-2.7-17, to appear in rawhide soon, ignores comments in /etc/updatedb.conf.

If there are multiple definitions of the same PRUNE* variable, the first one
is used; I think it is more important to preserve compatibility with previous
slocate versions than compatibility with GNU locate (which we stopped using
sometime around RHL 6.0).

The new slocate does indeed ignore comments.

You mention two clauses:
  1. ignores comments in /etc/updatedb.conf
  2. If there are multiple definitions of the same PRUNE* variable, the
     first one is used.

But even if these two clauses are met, you can still have problems.  (These
two clauses take care of the two earlier examples, but they were just
examples, not an exhaustive listing.)  The following three examples meet
the above two conditions, and a user might expect the following to work,
as they do for shell variables:

1.  PRUNEPATHS = \
    "/tmp /usr/tmp /var/tmp /afs /net /sfs /selinux /udev /mnt/floppy"

2.  PRUNEPATHS = '/tmp /usr/tmp /var/tmp /afs /net /sfs'

3.  PRUNEPATHS = "/tmp /udev /disk*"

    (for 3., assume that the system has /disk1, /disk2, etc.)


So I think that it is still worth documenting updatedb.conf someplace.
Currently, neither the updatedb.conf, nor the man pages of slocate and
updatedb mention updatedb.conf, so a user has no obvious way to
even know that the updatedb program uses that file.

My concern about documenting the format is that the format is very close to
that of setting variables in a shell, but not the same.  For example,
it is possible to use multi-line statements. But not like in example 1
above.  Instead, the following works:

PRUNEPATHS="/tmp /usr/tmp /var/tmp /afs /net /sfs
           /selinux /udev /mnt/floppy"

OK, that is a good reason to add updatedb.conf(5), added in rawhide slocate-2.7-19.

(Note that multi-line statements working this way are probably working
only because the parser is very permissive, not by design - and it is
possible they will stop working in a new upstream slocate version.)

Excellent!  That's what we needed.

This will be fixed in RHEL 5.