On Tue, Dec 04, 2001 at 04:02:13PM -0500, Paul Lussier wrote:
> 
> In a message dated: Tue, 04 Dec 2001 15:26:11 EST
> Paul Iadonisi said:
> 
> >  I normally don't speak up either, but I'm of the same mind as Red Hat.
> >To be blunt, as a sysadmin, I despise the concept of throwing *all* of a
> >packages files under a single directory. 
> 
> Well, to be fair I never advocated putting all files in one directory, I 
> advocated placing packages files in logical locations and used apache 
> as an example of a location which made sense, and that RH moved to 
> locations which didn't make sense or, confused long-time admins of 
> apache.  I do, however, don't think it's necessarilly a bad idea to 
> keep a package like apache under a hierarchy like /usr/apache.
> (at the very least, get rid of /home/apache, which is my biggest 
> gripe :)

I thought /home/httpd was bizarre as well.  The good news is, as of I think
Red Hat 7.0, they did change it to /var/www.  I still don't like /var/www,
but it does make more sense than /home/httpd (especially in automounted
/home situations -- ICK).  I usually change my document root to /data or
some other such location that is a separate partition or disk.  Red Hat's
file layout is starting to make more sense for each release.  They even
moved the samba configs from /etc to /etc/samba.


> > It typically makes PATH variables a mess.  I'm sure I could think of other
> > annoyances about that approach, but I'm pleased that the FHS exists, now,
> > and I'm glad to see Red Hat (others?) start to move in that direction.
> 
> Well, in general, /usr/local/[s]bin is almost always in the PATH 
> variable, and if not, it's one more addition, so I don't really see 
> that as a problem.

True, but I was referring more to the packages that like to put everything
including user invoked binaries in a subdirectory, meaning that it must
be added to the path.  What does make sense, however, is when there are
a large number of binaries/scripts that could pontentially conflict with other
more common commands, they can be put into a subdirectory and invoked by
a wrapper placed in /usr/[s]bin.  FreeS/WAN does this with the ipsec command
placed in /usr/sbin and all of the subcommands in /usr/lib/ipsec invoked with
'ipsec <subcommand>'.  (Well, they actually default to /usr/local/sbin and
/usr/local/lib, but I just patch the make file as part of my rpm spec.)

> >  My basic view is that each package should fit well into the structure of
> >the operating system.  I'ld rather see config files in /etc/<tool>, transient
> >data in /var/<whatever>, read-only stuff in /usr, etc. (that's et cetera,
> >not /etc :-)).
> 
> I have no argument with that, just so well as it's logical. For 
> example, /home/apache, to me, isn't all that logical.  I'd rather see 
> this stuff in /var/apache or /var/www like Debian has it.  To me that 
> makes sense.

As mentioned above, RH did at least fix that.

> > A package manager such as rpm makes it easy to find stuff.
> > I half jokingly say that one of the first things I aim to do to any
> >production system that I am given responsibility to maintain is
> >'rm -rf /usr/local'.  I consider /usr/local for testing, getting something
> >up and running quickly, just to try something out, for temporary use only.
> 
> That's actually not a bad idea. (of course, it's completely contrary 
> to the way I've done things in the past, but I like it :)
> 
> >I put everything I possibly can into an rpm package.  To be fair, I've spent
> >at least the last four years digging into the depths of rpm and know it
> >well enough to do that.
> 
> This here is the problem, though.  Now you need to know about a whole 
> bunch of package managers if you're in charge of a heterogeneous 
> environment; rpm, dpkg, pkgadd, whatever HP and Compaq use, etc.)

Agreed.  I was working on a project a while ago to take source rpms and
poop out pkgadd packages.  I had gotten pretty far, but I only have a
dinky little Sparc5 with Solaris 7, so development was kind of slow.
Really, I was just writting a bash script which was glue between rpm and
epm (http://www.easysw.com).  I'ld use rpm to build a binary rpm and then
run my bash script to create the epm config file and directory structure
and then invoke epm.  I worked pretty well, but there were still a number
of kinks to work out.

> Sure, you could compile rpm for each of them and use rpm for stuff 
> you add and the native one for base OS stuff, but if you're going to 
> go that far, why not just put everything into /usr/local from source 
> and be done with it.
> 
> Though, in the end, the only thing that really matters is that
> you should be consistent in whatever you do.

Which is why I was working on the project above.  Someday, I'll have the
motivation to revive it.

> >  Hell, I even refuse to use a kernel unless I can put it into an rpm format.
> >But again, to be fair, I've spent quite a bit of time learning how to
> >unravel kernel*.src.rpm files, modify them, and build them with my own
> >changes even while I sleep (or in lieu of sleep).
> 
> That's a little extreme for me. especially since I use Debian :)
> But seriously, why go through all that effort for the kernel?
> (unless you're just trying to be consistent :)

Just trying to stay consistent.  Makes upgrades more straightforward.  Though,
admittedly, I have rarely trusted operating system upgrades on ANY OS.  I
almost always do clean installs and migrate my changes/configs.

> >  Sorry for the rant.  I'm just a bit grumpy because of problems I found with
> >evolution 1.0 last night.  I'm opened to being convince of the error of my
> >ways, but it'll take something I haven't heard before.
> 
> What!  Bugs in a 1.0 version of code meant to mimick OutLook?  I'm 
> shocked! ;)

:-)

Turns out that evolution won't work for users in LDAP and no local /etc/passwd
entry.  It might work for NIS users -- I didn't try it.  The cause is that
wombat segfaults while doing the uid lookup.  A few minutes ago I found that
running nscd is reasonable workaround.  Before that, I had to add a local
username entry in /etc/passwd, which, to me, is an *un*acceptable workaround.


> 
> Seeya,
> Paul
> ----
> 
>                         God Bless America!
> 
>       ...we don't need to be perfect to be the best around,
>               and we never stop trying to be better. 
>                      Tom Clancy, The Bear and The Dragon
> 
> 
> 

-- 
-Paul Iadonisi
 Senior Systems Administrator
 Red Hat Certified Engineer / Local Linux Lobbyist
 Ever see a penguin fly?  --  Try Linux.
 GPL all the way: Sell services, don't lease secrets

*****************************************************************
To unsubscribe from this list, send mail to [EMAIL PROTECTED]
with the text 'unsubscribe gnhlug' in the message body.
*****************************************************************

Reply via email to