Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-28 Thread Roy Bamford
On 2012.12.27 22:13, William Hubbs wrote:
 On Thu, Dec 27, 2012 at 03:14:37PM -0500, Rich Freeman wrote:
  Something I don't like about this whole debate is that it tends to
  come off as I've never run an initramfs and darn it I want to keep
 it
  that way.  Gentoo has always been a cutting-edge/innovative 
 distro.
  We have prefix, hardened, x32, and we were among the first to
 support
  amd64.  Sure, that flexibility also lets you get away without an
  initramfs where other distros simply cannot.  However, the lack of
 an
  initramfs should not be a crutch.
 
 Rich,
 
 you just hit my concern about this debate right on the head. I feel
 like
 the nay-sayers are opposed to it because of the FHS, and the idea of
 critical software going in / and everything else in /usr. The 
 attitude
 seems to be that has always worked, so it must continue to work into
 the
 future, with no regard to the advantages that moving everything to
 /usr
 would give us.
 
 Another concern I've heard says that we shouldn't do this on linux
 because gentoo *bsd doesn't do it. I don't see that as relevant
 because ebuilds can be smart enough to test whether they are being
 emerged on Linux or *BSD.
 
 William
 
 

I don't think the 'luddites' have quite so black and white a view as 
that but if I expand on it much more, I'll reignite a flamewar we have 
already had. 

-- 
Regards,

Roy Bamford
(Neddyseagoon) a member of
elections
gentoo-ops
forum-mods
trustees


pgpM7U4Uf69D8.pgp
Description: PGP signature


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-28 Thread Mike Frysinger
On Thursday 27 December 2012 17:13:56 William Hubbs wrote:
 Another concern I've heard says that we shouldn't do this on linux
 because gentoo *bsd doesn't do it. I don't see that as relevant
 because ebuilds can be smart enough to test whether they are being
 emerged on Linux or *BSD.

+1
-mike


signature.asc
Description: This is a digitally signed message part.


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread Rich Freeman
On Thu, Dec 27, 2012 at 2:55 AM, Tony Chainsaw Vroon
chain...@gentoo.org wrote:
 On Wed, 2012-12-26 at 22:01 -0600, William Hubbs wrote:
 Actually, since ulm pointed out in another thread that the
 council has not mandated that we support separate /usr without an
 initramfs, I am re-considering this.

 So now that the /usr-merge steamroller can not break systems through
 udev, because an alternative now exists... another way must be found?
 That seems rather immature.
 What must be forked next to keep this working? openrc?

Tend to agree, assuming it causes no additional work for package maintainers.

This all started out as udev maintainers wanting to keep things simple
and closer to upstream.  Systems with a separate /usr breaking was a
bit of a side-effect.  The general direction that was chosen was to
provide alternatives for those who don't want to use an initramfs and
allow udev to follow upstream.  Life for the udev team is easier as a
result.

There is no decided strategic direction at Gentoo to move everything
into /usr as there is with Fedora.  It just doesn't make sense to
start pushing packages there.  That potentially CREATES work for
maintainers (bug reports, dealing with change, etc), and there is no
real benefit unless we systematically apply it (moving EVERYTHING into
/usr as with Fedora).  Systematically moving everything isn't going to
happen by just changing an eclass.

If somebody can see a benefit to having things moving in the direction
of /usr then by all means stick a flag in the profiles and use it to
control this behavior, and then we give choice to the end-user.
However, I don't really see the point.  When you change the status quo
it should be because it either lowers cost or produces benefit.

Rich



Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread William Hubbs
On Thu, Dec 27, 2012 at 07:55:38AM +, Tony Chainsaw Vroon wrote:
 On Wed, 2012-12-26 at 22:01 -0600, William Hubbs wrote:
  Actually, since ulm pointed out in another thread that the
  council has not mandated that we support separate /usr without an
  initramfs, I am re-considering this. 
 
 So now that the /usr-merge steamroller can not break systems through
 udev, because an alternative now exists... another way must be found?
 That seems rather immature.
 What must be forked next to keep this working? openrc?

Nothing must be forked. No one has said anything is happening yet. This
is just a discussion.

William



pgpMXyuN3dwx0.pgp
Description: PGP signature


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread William Hubbs
On Thu, Dec 27, 2012 at 08:00:09AM -0500, Rich Freeman wrote:
 On Thu, Dec 27, 2012 at 2:55 AM, Tony Chainsaw Vroon
 chain...@gentoo.org wrote:
  On Wed, 2012-12-26 at 22:01 -0600, William Hubbs wrote:
  Actually, since ulm pointed out in another thread that the
  council has not mandated that we support separate /usr without an
  initramfs, I am re-considering this.
 
  So now that the /usr-merge steamroller can not break systems through
  udev, because an alternative now exists... another way must be found?
  That seems rather immature.
  What must be forked next to keep this working? openrc?
 
 Tend to agree, assuming it causes no additional work for package maintainers.

As I and others have said on this list a thousdand times, moving
everything to /usr never had anything to do with systemd and udev. This
is a completely separate topic.

The arguments for moving everything into /usr seem to be pretty strong
[1], and as gregkh and others have said, it would benefit us in the longrun
to do it.

Given that, that is not even what I'm discussing. I am just discussing
moving the libraries that we manually install into /lib* back to
/usr/lib* on Linux.

William

[1] http://www.freedesktop.org/wiki/Software/systemd/TheCaseForTheUsrMerge


pgpNxjovyb3us.pgp
Description: PGP signature


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread Mike Gilbert
On Thu, Dec 27, 2012 at 12:03 PM, William Hubbs willi...@gentoo.org wrote:
 On Thu, Dec 27, 2012 at 08:00:09AM -0500, Rich Freeman wrote:
 On Thu, Dec 27, 2012 at 2:55 AM, Tony Chainsaw Vroon
 chain...@gentoo.org wrote:
  On Wed, 2012-12-26 at 22:01 -0600, William Hubbs wrote:
  Actually, since ulm pointed out in another thread that the
  council has not mandated that we support separate /usr without an
  initramfs, I am re-considering this.
 
  So now that the /usr-merge steamroller can not break systems through
  udev, because an alternative now exists... another way must be found?
  That seems rather immature.
  What must be forked next to keep this working? openrc?

 Tend to agree, assuming it causes no additional work for package maintainers.

 As I and others have said on this list a thousdand times, moving
 everything to /usr never had anything to do with systemd and udev. This
 is a completely separate topic.


It has everything to do with udev if you (as the udev maintainer for
Gentoo) decide to put zero effort into keeping udev working with a
traditional split-/usr configuration. Although udev is only one
package of many, it is a pretty damn critical one.



Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread William Hubbs
On Thu, Dec 27, 2012 at 01:35:55PM -0500, Mike Gilbert wrote:
 On Thu, Dec 27, 2012 at 12:03 PM, William Hubbs willi...@gentoo.org wrote:
  On Thu, Dec 27, 2012 at 08:00:09AM -0500, Rich Freeman wrote:
  On Thu, Dec 27, 2012 at 2:55 AM, Tony Chainsaw Vroon
  chain...@gentoo.org wrote:
   On Wed, 2012-12-26 at 22:01 -0600, William Hubbs wrote:
   Actually, since ulm pointed out in another thread that the
   council has not mandated that we support separate /usr without an
   initramfs, I am re-considering this.
  
   So now that the /usr-merge steamroller can not break systems through
   udev, because an alternative now exists... another way must be found?
   That seems rather immature.
   What must be forked next to keep this working? openrc?
 
  Tend to agree, assuming it causes no additional work for package 
  maintainers.
 
  As I and others have said on this list a thousdand times, moving
  everything to /usr never had anything to do with systemd and udev. This
  is a completely separate topic.
 
 
 It has everything to do with udev if you (as the udev maintainer for
 Gentoo) decide to put zero effort into keeping udev working with a
 traditional split-/usr configuration. Although udev is only one
 package of many, it is a pretty damn critical one.

As I said on another thread, there was a misunderstanding on my part
about setting up udev. I am looking into fixing that with the next
release, but I need to coordinate with systemd as well, so I thought it
would be good to wait for 197 to be released, so again, this is not
correct.

William



pgpipNCBpHwXg.pgp
Description: PGP signature


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread Rich Freeman
On Thu, Dec 27, 2012 at 12:03 PM, William Hubbs willi...@gentoo.org wrote:

 As I and others have said on this list a thousdand times, moving
 everything to /usr never had anything to do with systemd and udev. This
 is a completely separate topic.

Understood.  However, the whole request to not have to support a
separate /usr without an initramfs was brought up by the udev team.
If udev doesn't have the need, then they should just go do what they
want to do and stop asking the council to step in, as there apparently
isn't anything for them to decide on.


 The arguments for moving everything into /usr seem to be pretty strong
 [1], and as gregkh and others have said, it would benefit us in the longrun
 to do it.

 Given that, that is not even what I'm discussing. I am just discussing
 moving the libraries that we manually install into /lib* back to
 /usr/lib* on Linux.

I think moving everything into /usr is a good idea.  However:

1.  It isn't my decision to make.  This is the role of the Council.
2.  It doesn't make sense for every dev to just stick stuff wherever
they personally feel is best.
3.  Moving just a bunch of libraries to /usr and nothing else is dumb.
 It brings none of the benefits of the /usr move, and gets rid of all
of the benefits of complying with FHS (like systems booting fine with
a separate /usr - and yes I know this is already broken despite the
fact that it works just fine for 99% of the people running in this
configuration).  This is one of those situations where you need to
have a plan and do it right, or don't do it at all.

If people want to argue for a /usr move by all means do so.  If people
want to beg the Council to back this, by all means do so.  If people
want to run for Council by all means do so.  If you want to build a
mechanism that gives the choice to the end user based on a profile
setting or some other sensible mechanism, by all means do so.

But, until the Council decides that we're really doing a coordinated
/usr move, then let's leave things alone.  Sticking stuff in random
locations per the whim of individual maintainers will cause nothing
but trouble.

Rich



Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread William Hubbs
On Thu, Dec 27, 2012 at 01:49:50PM -0500, Rich Freeman wrote:
 On Thu, Dec 27, 2012 at 12:03 PM, William Hubbs willi...@gentoo.org wrote:
 
  As I and others have said on this list a thousdand times, moving
  everything to /usr never had anything to do with systemd and udev. This
  is a completely separate topic.
 
 Understood.  However, the whole request to not have to support a
 separate /usr without an initramfs was brought up by the udev team.
 If udev doesn't have the need, then they should just go do what they
 want to do and stop asking the council to step in, as there apparently
 isn't anything for them to decide on.
 
 I wasn't actually asking the council to step in. I was just trying to
 have a discussion here.

 
  The arguments for moving everything into /usr seem to be pretty strong
  [1], and as gregkh and others have said, it would benefit us in the longrun
  to do it.
 
  Given that, that is not even what I'm discussing. I am just discussing
  moving the libraries that we manually install into /lib* back to
  /usr/lib* on Linux.
 
 I think moving everything into /usr is a good idea.  However:
 
 1.  It isn't my decision to make.  This is the role of the Council.

Tell me if I am wrong here. My understanding is that this is only true
if the community itself doesn't make the decision first.

 2.  It doesn't make sense for every dev to just stick stuff wherever
 they personally feel is best.
 3.  Moving just a bunch of libraries to /usr and nothing else is dumb.
  It brings none of the benefits of the /usr move, and gets rid of all
 of the benefits of complying with FHS (like systems booting fine with
 a separate /usr - and yes I know this is already broken despite the
 fact that it works just fine for 99% of the people running in this
 configuration).  This is one of those situations where you need to
 have a plan and do it right, or don't do it at all.
 
 Ok, I can agree with this.

 If people want to argue for a /usr move by all means do so.  If people
 want to beg the Council to back this, by all means do so.  If people
 want to run for Council by all means do so.  If you want to build a
 mechanism that gives the choice to the end user based on a profile
 setting or some other sensible mechanism, by all means do so.
 
 But, until the Council decides that we're really doing a coordinated
 /usr move, then let's leave things alone.  Sticking stuff in random
 locations per the whim of individual maintainers will cause nothing
 but trouble.
 
 There was a long thread a while back where the /usr merge was discussed
 in depth and there was no escalation to the council to make the
 decision [1]. Unless I don't remember something significant out of that
 thread, we agreed that even though some of us don't like the /usr
 merge, it is probably a good thing for us to do it in the longrun.

If I were to start that thread now, I would change my introduction to
not specifically mention udev, systemd and kmod, but my view still is
that it will be better for us in the longrun if we do it. Maybe that is
a topic for another thread though.

William

[1]
http://archives.gentoo.org/gentoo-dev/msg_c3c5bdabbe058b08627ff04cee896af3.xml


pgps9xpLHSSH9.pgp
Description: PGP signature


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread Rich Freeman
On Thu, Dec 27, 2012 at 2:48 PM, William Hubbs willi...@gentoo.org wrote:
 On Thu, Dec 27, 2012 at 01:49:50PM -0500, Rich Freeman wrote:
 Understood.  However, the whole request to not have to support a
 separate /usr without an initramfs was brought up by the udev team.
 If udev doesn't have the need, then they should just go do what they
 want to do and stop asking the council to step in, as there apparently
 isn't anything for them to decide on.

  I wasn't actually asking the council to step in. I was just trying to
  have a discussion here.

The Council WAS asked to step in:
http://www.gentoo.org/proj/en/council/meeting-logs/20120403-summary.txt
http://thread.gmane.org/gmane.linux.gentoo.project/1864/focus=1867

However, you are right, the udev team did not actually request this.
So, if udev 180+ doesn't break anything that wasn't already broken in
udev 179- then just go about your business...  :)

 1.  It isn't my decision to make.  This is the role of the Council.

 Tell me if I am wrong here. My understanding is that this is only true
 if the community itself doesn't make the decision first.

True, but I don't see any consensus on this topic.  The /usr move is
VERY controversial, at least within Gentoo.  This really doesn't fall
into the domain of any one project either - this affects the whole
distro.  Even if it did fall into the domain of a single project,
anybody with half a brain would realize that you don't just do
something like this on the initiative of a few individuals unless you
want a really big mess on your hands.

 If I were to start that thread now, I would change my introduction to
 not specifically mention udev, systemd and kmod, but my view still is
 that it will be better for us in the longrun if we do it. Maybe that is
 a topic for another thread though.

Agreed.  There is no harm in discussing it.  I'd love to see this as a
supported Gentoo configuration, and perhaps even as the default.
However, this should come down to a discussion of pros/cons,
especially in terms of what kinds of opportunities it creates.

Something I don't like about this whole debate is that it tends to
come off as I've never run an initramfs and darn it I want to keep it
that way.  Gentoo has always been a cutting-edge/innovative distro.
We have prefix, hardened, x32, and we were among the first to support
amd64.  Sure, that flexibility also lets you get away without an
initramfs where other distros simply cannot.  However, the lack of an
initramfs should not be a crutch.

I could see the exact same argument unfolding 15 years ago about
forcing users to have a bootloader like grub.  Go bring up the
suggestion that the kernel should support direct booting on lkml and
I'm sure Linus will tell you to bugger_off...

Rich



Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread Tony Chainsaw Vroon
On Thu, 2012-12-27 at 15:14 -0500, Rich Freeman wrote:
 Go bring up the suggestion that the kernel should support direct
 booting on lkml 

And be pointed at EFI_STUB functionality. Next?

Regards,
Tony V.


signature.asc
Description: This is a digitally signed message part


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread Rich Freeman
On Thu, Dec 27, 2012 at 3:27 PM, Tony Chainsaw Vroon
chain...@gentoo.org wrote:
 On Thu, 2012-12-27 at 15:14 -0500, Rich Freeman wrote:
 Go bring up the suggestion that the kernel should support direct
 booting on lkml

 And be pointed at EFI_STUB functionality. Next?

I was referring to booting from a legacy BIOS - hence my comment about
15 years ago - back when this was a completely supported linux
configuration.



Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread Mike Frysinger
On Wednesday 26 December 2012 23:01:46 William Hubbs wrote:
 On Mon, Dec 24, 2012 at 10:48:23PM +0100, Diego Elio Pettenò wrote:
  On 24/12/2012 20:08, Mike Frysinger wrote:
   i.e. saying we should get rid of gen_usr_ldscript and use
   --libdir=/lib makes absolutely no sense.  it's just begging for
   people to screw things up constantly and waste developer time for 0
   gain.
  
  Amen.
 
 Actually, since ulm pointed out in another thread that the
 council has not mandated that we support separate /usr without an
 initramfs, I am re-considering this.
 
 In linux-only ebuilds, if we install everything in /usr as gregkh and
 others have suggested, we can remove this call from them. Also, for the
 other ebuilds that have this call, we can eventually disable the
 function on Linux systems.

as mentioned in bug 417451, the ebuilds won't drop the `gen_usr_ldscript` 
call.  we'll update the gen_usr_ldscript itself to be a no-op.  that way non-
linux systems continue to work, as well as linux users who want to live in the 
past.

on the upside, i will no longer have compassion for keeping / small, so we can 
close all the existing bugs about pkg foo in / is linked against lib bar in 
/usr by dumping these calls.  or maybe we symlink /usr/lib to /lib ? :)
-mike


signature.asc
Description: This is a digitally signed message part.


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread Mike Frysinger
On Thursday 27 December 2012 13:49:50 Rich Freeman wrote:
 I think moving everything into /usr is a good idea.  However:

i don't think it's hard to support both.  the majority of packages just want 
to relocate shared libs into / from /usr and that's easy with one line:
gen_usr_ldscript -a foo
put a knob into the func itself (perhaps a var set in the profile's 
make.defaults) and you've addressed more than 50% of the problem.

very few packages actually install into /bin and /sbin, and i don't mind a 
USE=sep-usr flag for them (relevant since i also see that i'm maintaining most 
of those packages).

 3.  Moving just a bunch of libraries to /usr and nothing else is dumb.
  It brings none of the benefits of the /usr move

sure

 and gets rid of all
 of the benefits of complying with FHS (like systems booting fine with
 a separate /usr - and yes I know this is already broken despite the
 fact that it works just fine for 99% of the people running in this
 configuration).

strictly speaking, i don't think FHS mandates sep /usr be supported.  it's 
fairly easy to read a merged /usr setup as being FHS compliant.

 But, until the Council decides that we're really doing a coordinated
 /usr move, then let's leave things alone.  Sticking stuff in random
 locations per the whim of individual maintainers will cause nothing
 but trouble.

aka today's status quo
-mike


signature.asc
Description: This is a digitally signed message part.


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-27 Thread William Hubbs
On Thu, Dec 27, 2012 at 03:14:37PM -0500, Rich Freeman wrote:
 Something I don't like about this whole debate is that it tends to
 come off as I've never run an initramfs and darn it I want to keep it
 that way.  Gentoo has always been a cutting-edge/innovative distro.
 We have prefix, hardened, x32, and we were among the first to support
 amd64.  Sure, that flexibility also lets you get away without an
 initramfs where other distros simply cannot.  However, the lack of an
 initramfs should not be a crutch.

Rich,

you just hit my concern about this debate right on the head. I feel like
the nay-sayers are opposed to it because of the FHS, and the idea of
critical software going in / and everything else in /usr. The attitude
seems to be that has always worked, so it must continue to work into the
future, with no regard to the advantages that moving everything to /usr
would give us.

Another concern I've heard says that we shouldn't do this on linux
because gentoo *bsd doesn't do it. I don't see that as relevant
because ebuilds can be smart enough to test whether they are being
emerged on Linux or *BSD.

William



pgptEYdZoDwIs.pgp
Description: PGP signature


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-26 Thread William Hubbs
On Mon, Dec 24, 2012 at 10:48:23PM +0100, Diego Elio Pettenò wrote:
 On 24/12/2012 20:08, Mike Frysinger wrote:
  i.e. saying we should get rid of gen_usr_ldscript and use --libdir=/lib 
  makes absolutely no sense.  it's just begging for people to screw things up 
  constantly and waste developer time for 0 gain.
 
 Amen.

Actually, since ulm pointed out in another thread that the
council has not mandated that we support separate /usr without an
initramfs, I am re-considering this.

In linux-only ebuilds, if we install everything in /usr as gregkh and
others have suggested, we can remove this call from them. Also, for the
other ebuilds that have this call, we can eventually disable the
function on Linux systems.

William



pgp7xHIN3FrCs.pgp
Description: PGP signature


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-26 Thread Tony Chainsaw Vroon
On Wed, 2012-12-26 at 22:01 -0600, William Hubbs wrote:
 Actually, since ulm pointed out in another thread that the
 council has not mandated that we support separate /usr without an
 initramfs, I am re-considering this. 

So now that the /usr-merge steamroller can not break systems through
udev, because an alternative now exists... another way must be found?
That seems rather immature.
What must be forked next to keep this working? openrc?

Regards,
Tony V.


signature.asc
Description: This is a digitally signed message part


Re: [gentoo-dev] gen_usr_ldscript --libdir=/lib

2012-12-24 Thread Diego Elio Pettenò
On 24/12/2012 20:08, Mike Frysinger wrote:
 i.e. saying we should get rid of gen_usr_ldscript and use --libdir=/lib 
 makes absolutely no sense.  it's just begging for people to screw things up 
 constantly and waste developer time for 0 gain.

Amen.

-- 
Diego Elio Pettenò — Flameeyes
flamee...@flameeyes.eu — http://blog.flameeyes.eu/



signature.asc
Description: OpenPGP digital signature