With udev, are dev and MAKEDEV still required?

[David Hollis]

With udev now handling /dev and repopulating upon reboot, it seems that
the dev and MAKEDEV packages are no longer relevant.  However, they can
not be uninstalled due to manual requirements for some packages:

[root@dhollis-lnx i2c]# rpm -e dev MAKEDEV
error: Failed dependencies:
        dev is needed by (installed) which-2.16-4
        dev is needed by (installed) mod_ssl-2.0.50-4
        dev is needed by (installed) kdelibs-3.3.0-1
        dev is needed by (installed) mkinitrd-4.1.1-1
        dev is needed by (installed) initscripts-7.67-1
        MAKEDEV >= 3.0 is needed by (installed) raidtools-1.00.3-8

I'm not terribly concerned about these packages from a space perspective
(especially since all of dev's files get blown away anyway!) but now rpm
--verify dev  blows up in all kinds of fun ways since all of its files
are missing.  With the active discussions about udev, I'm quite certain
that it is not the only way to do things at this point and the dev and
MAKEDEV packages will remain for those that prefer that method, but
should these dependencies be investigated to see just how necessary they
really are?


-- 
David T Hollis <dhollis(a)davehollis.com&gt;

[Michel Salim]

On Tue, 24 Aug 2004 21:15:34 -0400, David T Hollis
<dhollis(a)davehollis.com&gt; wrote:
>  With udev now handling /dev and repopulating upon reboot, it seems
> that
>  the dev and MAKEDEV packages are no longer relevant.  However, they can
>  not be uninstalled due to manual requirements for some packages:
>  
>  [root@dhollis-lnx i2c]# rpm -e dev MAKEDEV
>  error: Failed dependencies:
>          dev is needed by (installed) which-2.16-4
>          dev is needed by (installed) mod_ssl-2.0.50-4
>          dev is needed by (installed) kdelibs-3.3.0-1
>          dev is needed by (installed) mkinitrd-4.1.1-1
>          dev is needed by (installed) initscripts-7.67-1
>          MAKEDEV >= 3.0 is needed by (installed) raidtools-1.00.3-8
>  
>  I'm not terribly concerned about these packages from a space perspective
>  (especially since all of dev's files get blown away anyway!) but now rpm
>  --verify dev  blows up in all kinds of fun ways since all of its files
>  are missing.  With the active discussions about udev, I'm quite certain
>  that it is not the only way to do things at this point and the dev and
>  MAKEDEV packages will remain for those that prefer that method, but
>  should these dependencies be investigated to see just how necessary they
>  really are?
>   
Having udev Provides: dev would be a way out, in that case, since that
is precisely what packages that depend on dev need, device nodes?

-- 
Michel Salim 林智勇
http://salimma.livejournal.com

[Enrico Scholz]

dhollis(a)davehollis.com (David T Hollis) writes:

>  With udev now handling /dev and repopulating upon reboot, it seems
> that
>  the dev and MAKEDEV packages are no longer relevant. 
ACK.

A problem with MAKEDEV is, that it places the MAKEDEV *binary* into
/dev. This is a really bad place for it; devfs under 2.4 removed it and
buildsystems which need a special /dev will remove it also.

Especially in the latter case, this is very problematic as some packages
need it to create %files list. I would suggest to move MAKEDEV into
/sbin and to create a symlink into /dev on demand.


>  [root@dhollis-lnx i2c]# rpm -e dev MAKEDEV
>  error: Failed dependencies:
>          dev is needed by (installed) which-2.16-4
>          dev is needed by (installed) mod_ssl-2.0.50-4
>          dev is needed by (installed) kdelibs-3.3.0-1
>          dev is needed by (installed) mkinitrd-4.1.1-1
>          dev is needed by (installed) initscripts-7.67-1
>          MAKEDEV >= 3.0 is needed by (installed) raidtools-1.00.3-8 
There are probably some additional BuildRequires: on MAKEDEV missing in
this list.


>  I'm not terribly concerned about these packages from a space
> perspective
>  (especially since all of dev's files get blown away anyway!) but now rpm
>  --verify dev 
Adding '%_netsharedpath /dev' into your /etc/rpm/macros file and to
reinstall the dev package would be a workaround.




Enrico

[Alan Cox]

On Wed, Aug 25, 2004 at 04:57:55AM +0200, Enrico Scholz wrote:
>  A problem with MAKEDEV is, that it places the MAKEDEV *binary* into
>  /dev. This is a really bad place for it; devfs under 2.4 removed it and
>  buildsystems which need a special /dev will remove it also. 
Unix tradition is the essential reason for this. Nothing more.

[Nalin Dahyabhai]

On Sat, Aug 28, 2004 at 06:40:22AM -0400, Alan Cox wrote:
>  On Wed, Aug 25, 2004 at 04:57:55AM +0200, Enrico Scholz wrote:
>  > A problem with MAKEDEV is, that it places the MAKEDEV *binary* into
>  > /dev. This is a really bad place for it; devfs under 2.4 removed it and
>  > buildsystems which need a special /dev will remove it also.
>  
>  Unix tradition is the essential reason for this. Nothing more. 
The wording in FHS [1] seems a bit vague about this case.  Opinions?

Nalin

[1] http://www.pathname.com/fhs/pub/fhs-2.3.html#SPECIFICOPTIONS4

[Alan Cox]

On Mon, Aug 30, 2004 at 12:19:47PM -0400, Nalin Dahyabhai wrote:
>  On Sat, Aug 28, 2004 at 06:40:22AM -0400, Alan Cox wrote:
>  > On Wed, Aug 25, 2004 at 04:57:55AM +0200, Enrico Scholz wrote:
>  > > A problem with MAKEDEV is, that it places the MAKEDEV *binary* into
>  > > /dev. This is a really bad place for it; devfs under 2.4 removed it and
>  > > buildsystems which need a special /dev will remove it also.
>  > 
>  > Unix tradition is the essential reason for this. Nothing more.
>  
>  The wording in FHS [1] seems a bit vague about this case.  Opinions? 
Looks ok to me. It belongs there unless it isnt needed. Since udev doesn't
create every possible device in every configuration I see no harm putting
it there.

[Jef Spaleta]

On Tue, 24 Aug 2004 21:15:34 -0400, David T Hollis
<dhollis(a)davehollis.com&gt; wrote:
>  With udev now handling /dev and repopulating upon reboot, it seems
> that
>  the dev and MAKEDEV packages are no longer relevant.   
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=130746

I'm not sure its safe to say the static dev packages are completely
irrelevant now.


-jef

[David Hollis]

On Tue, 2004-08-24 at 23:07 -0400, Jeff Spaleta wrote:
>  On Tue, 24 Aug 2004 21:15:34 -0400, David T Hollis
>  <dhollis(a)davehollis.com&gt; wrote:
>  > With udev now handling /dev and repopulating upon reboot, it seems that
>  > the dev and MAKEDEV packages are no longer relevant.  
>  
>  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=130746
>  
>  I'm not sure its safe to say the static dev packages are completely
>  irrelevant now.
>   
In looking into the dev requirement for which (which struck me as quite
odd), I found this bug report:
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=99275 .  The which
package needs dev so that the postinstall scripts calling install-info
can use "> /dev/null".  There were ordering problems with 'which'
being
installed before 'dev' since 'which' doesn't have much in the way of
requirements.  I suppose it may be better to have a requirement
on /dev/null, though currently only 'dev' provides it.  If there is a
future without 'dev' (optional or mandatory), that sort of scenario will
need to be addressed.


-- 
David T Hollis <dhollis(a)davehollis.com&gt;

[dragoran]

David T Hollis schrieb:

> On Tue, 2004-08-24 at 23:07 -0400, Jeff Spaleta wrote:
>   
>
> >On Tue, 24 Aug 2004 21:15:34 -0400, David T Hollis
> ><dhollis(a)davehollis.com> wrote:
> >    
> >
> >>With udev now handling /dev and repopulating upon reboot, it seems that
> >>the dev and MAKEDEV packages are no longer relevant.  
> >>      
> >>
> >https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=130746
> >
> >I'm not sure its safe to say the static dev packages are completely
> >irrelevant now.
> >
> >    
> >
>
> In looking into the dev requirement for which (which struck me as quite
> odd), I found this bug report:
> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=99275 .  The which
> package needs dev so that the postinstall scripts calling install-info
> can use "> /dev/null".  There were ordering problems with 'which'
> being
> installed before 'dev' since 'which' doesn't have much in the way of
> requirements.  I suppose it may be better to have a requirement
> on /dev/null, though currently only 'dev' provides it.  If there is a
> future without 'dev' (optional or mandatory), that sort of scenario will
> need to be addressed.
>
>
>   
>  
does /dev/null exists when using udev?

[David Hollis]

On Wed, 2004-08-25 at 15:36 +0200, dragoran wrote:
>  >
>  >In looking into the dev requirement for which (which struck me as quite
>  >odd), I found this bug report:
>  >https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=99275 .  The which
>  >package needs dev so that the postinstall scripts calling install-info
>  >can use "> /dev/null".  There were ordering problems with
> 'which' being
>  >installed before 'dev' since 'which' doesn't have much in the way
> of
>  >requirements.  I suppose it may be better to have a requirement
>  >on /dev/null, though currently only 'dev' provides it.  If there is a
>  >future without 'dev' (optional or mandatory), that sort of scenario will
>  >need to be addressed.
>  >
>  >
>  >  
>  >
>  does /dev/null exists when using udev?
>  
>   
On a normally running system - yes.  udev does not create /dev/null upon
RPM install/upgrade and there are no postinstall scripts, only a preun
to remove the service.

-- 
David T Hollis <dhollis(a)davehollis.com&gt;

[Manuel Moreno]

On Tue, 2004-08-24 at 23:07 -0400, Jeff Spaleta wrote:
>  On Tue, 24 Aug 2004 21:15:34 -0400, David T Hollis
>  <dhollis(a)davehollis.com&gt; wrote:
>  > With udev now handling /dev and repopulating upon reboot, it seems that
>  > the dev and MAKEDEV packages are no longer relevant.  
>  
>  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=130746
>  
>  I'm not sure its safe to say the static dev packages are completely
>  irrelevant now. 
Yes, please I want static dev, too,

(me too? blush) 
-- 
Manuel Moreno <manolo(a)miconexion.com&gt;

[Jeremy Katz]

On Tue, 2004-08-24 at 21:15 -0400, David T Hollis wrote:
>  With udev now handling /dev and repopulating upon reboot, it seems
> that
>  the dev and MAKEDEV packages are no longer relevant.  However, they can
>  not be uninstalled due to manual requirements for some packages: 
[snip]
>  I'm not terribly concerned about these packages from a space
> perspective
>  (especially since all of dev's files get blown away anyway!) but now rpm
>  --verify dev  blows up in all kinds of fun ways since all of its files
>  are missing.  With the active discussions about udev, I'm quite certain
>  that it is not the only way to do things at this point and the dev and
>  MAKEDEV packages will remain for those that prefer that method, but
>  should these dependencies be investigated to see just how necessary they
>  really are? 
dev is of potentially dubious relevance now, yes.  MAKEDEV is still
quite relevant (although the comment about moving it to /sbin makes good
sense to me).  

The full fallout of udev has just begun, though... :)

Jeremy