Debian Bug report logs - #923091
base-installer: Allow installing w/o the broken merged-usr-via-symlinks

Reply or subscribe to this bug.

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Guillem Jover <guillem@debian.org>

To: submit@bugs.debian.org

Subject: base-installer: Allow installing w/o the broken merged-usr-via-symlinks

Date: Sun, 24 Feb 2019 03:10:17 +0100

Package: base-installer
Version: 1.187
Severity: wishlist

Hi!

The current base-installer uses the default debootstrap settings
which end up unconditionally installing systems with the
merged-usr-via-symlinks deployment method which is broken by design,
please see:

  <https://lists.debian.org/debian-devel/2019/02/msg00251.html>

I'm aware the original request to change the debootstrap default got
unfortunately moved to the tech-ctte. :(

But regardless of that outcome I'd like to request to have a way to
install using debootstrap's --no-merged-usr option, to not have to
install from stretch and then upgrade to buster, or having to drop into
a shell and do manual stuff from within the installer.

In addition it would be also nice if that option was passed whenever
/usr is not on its own partition, because then the properties from
the merged-/usr concept are not relevant anymore, but we get all the
downsides of the broken deployment method.

If this was to be applied for buster, I'd be happy to provide patches.

Otherwise I guess I might need to end up looking to generate
alternative netinst somewhere else or something. :(

Thanks,
Guillem

Message #10 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Cyril Brulebois <kibi@debian.org>

To: Guillem Jover <guillem@debian.org>, 923091@bugs.debian.org

Subject: Re: Bug#923091: base-installer: Allow installing w/o the broken merged-usr-via-symlinks

Date: Sun, 24 Feb 2019 19:49:23 +0100

[Message part 1 (text/plain, inline)]

Hi,

Guillem Jover <guillem@debian.org> (2019-02-24):
> The current base-installer uses the default debootstrap settings
> which end up unconditionally installing systems with the
> merged-usr-via-symlinks deployment method which is broken by design,
> please see:
> 
>   <https://lists.debian.org/debian-devel/2019/02/msg00251.html>

I won't be commenting on the brokenness you claim in this bug report,
and in the mail you linked to, and in the subthread; that doesn't mean
I agree with your assessment.

Instead, I'll concentrate on the actual request for a change in
bootstrap-base below:

> I'm aware the original request to change the debootstrap default got
> unfortunately moved to the tech-ctte. :(
> 
> But regardless of that outcome I'd like to request to have a way to
> install using debootstrap's --no-merged-usr option, to not have to
> install from stretch and then upgrade to buster, or having to drop into
> a shell and do manual stuff from within the installer.

I'm not sure what costs (initial addition then maintenance) an
expert-only option would have, but I'm naively expect them to be rather
low? It seems to me that this would suit your needs and also make that
configurable through preseeding?

If that were to be implemented: the question must not be asked in a
normal installation, I think it shouldn't be translatable (at least at
first) so that it doesn't put more burden on l10n people (coordinator
and translators) for an expert-only option. Cc-ing Holger for input.

> In addition it would be also nice if that option was passed whenever
> /usr is not on its own partition, because then the properties from
> the merged-/usr concept are not relevant anymore, but we get all the
> downsides of the broken deployment method.

That shouldn't be tracked in this bug report as that seems quite
orthogonal (base-installer runs in d-i, not on installed systems)?

> If this was to be applied for buster, I'd be happy to provide patches.

Not promising anything, but I think I should be able to test/review such
patches once I'm done with updating the test suite.

> Otherwise I guess I might need to end up looking to generate
> alternative netinst somewhere else or something. :(

That won't be a factor regarding the inclusion of possible patches into
d-i, for buster or afterward.


Cheers,
-- 
Cyril Brulebois (kibi@debian.org)            <https://debamax.com/>
D-I release manager -- Release team member -- Freelance Consultant

[signature.asc (application/pgp-signature, inline)]

Message #15 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Holger Wansing <hwansing@mailbox.org>

To: Cyril Brulebois <kibi@debian.org>

Cc: Guillem Jover <guillem@debian.org>, 923091@bugs.debian.org

Subject: Re: Bug#923091: base-installer: Allow installing w/o the broken merged-usr-via-symlinks

Date: Sun, 24 Feb 2019 21:10:18 +0100

Hi,

Cyril Brulebois <kibi@debian.org> wrote:
> 
> If that were to be implemented: the question must not be asked in a
> normal installation, I think it shouldn't be translatable (at least at
> first) so that it doesn't put more burden on l10n people (coordinator
> and translators) for an expert-only option. Cc-ing Holger for input.

I'm unable to completely follow this whole 'merged-usr' thing, but to me
personally it seems like a somewhat controversial issue, which may see some
more discussion and changings.

So I tend to follow Kibi's proposal to keep the strings untranslateable for 
now, to allow everything to settle down, and make them translatable within 
the Buster+1 development cycle.

Filing a separate bugreport for this untranslatable -> translatable change
will be the best way, to make sure we don't forget about this.


So long
Holger


-- 
Holger Wansing <hwansing@mailbox.org>
PGP-Fingerprint: 496A C6E8 1442 4B34 8508  3529 59F1 87CA 156E B076

Message #20 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Ian Jackson <ijackson@chiark.greenend.org.uk>

To: 923091@bugs.debian.org

Cc: debian-ctte@lists.debian.org, debian-release@bugs.debian.org

Subject: That merged-usr is mandatory is RC

Date: Mon, 13 May 2019 11:22:35 +0100

Control: severity -1 serious

In #923091, Guillem (with dpkg maintainer hat on) asks for a
base-installer option to allow installing buster without merged-usr.

Guillem filed the bug as `wishlist' but given the controversy it seems
to me that it should be RC if for no other reasons than social
cohesion.

CCing the TC FYI (they have already been involved in merged-usr
debates via #914897) and the release team, in case they have an
opinion.  FAOD I am not a maintainer of base-files but AFAICT the
base-files maintainer has not expressed an opinion about severity.

Ian.

-- 
Ian Jackson <ijackson@chiark.greenend.org.uk>   These opinions are my own.

If I emailed you from an address @fyvzl.net or @evade.org.uk, that is
a private address which bypasses my fierce spamfilter.

Message #27 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Ian Jackson <ijackson@chiark.greenend.org.uk>

To: debian-release@lists.debian.org

Cc: 923091@bugs.debian.org, debian-ctte@lists.debian.org

Subject: Re: That merged-usr is mandatory is RC

Date: Mon, 13 May 2019 11:30:06 +0100

(sending this because I got the release team address wrong)

Ian Jackson writes ("That merged-usr is mandatory is RC"):
> Control: severity -1 serious
> 
> In #923091, Guillem (with dpkg maintainer hat on) asks for a
> base-installer option to allow installing buster without merged-usr.
> 
> Guillem filed the bug as `wishlist' but given the controversy it seems
> to me that it should be RC if for no other reasons than social
> cohesion.
> 
> CCing the TC FYI (they have already been involved in merged-usr
> debates via #914897) and the release team, in case they have an
> opinion.  FAOD I am not a maintainer of base-files but AFAICT the
> base-files maintainer has not expressed an opinion about severity.

-- 
Ian Jackson <ijackson@chiark.greenend.org.uk>   These opinions are my own.

If I emailed you from an address @fyvzl.net or @evade.org.uk, that is
a private address which bypasses my fierce spamfilter.

Message #32 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Bastian Blank <waldi@debian.org>

To: Ian Jackson <ijackson@chiark.greenend.org.uk>, 923091@bugs.debian.org

Cc: Guillem Jover <guillem@debian.org>

Subject: Re: Bug#923091: That merged-usr is mandatory is RC

Date: Mon, 13 May 2019 17:24:55 +0200

Control: severity -1 wishlist

Hi Ian

On Mon, May 13, 2019 at 11:22:35AM +0100, Ian Jackson wrote:
> In #923091, Guillem (with dpkg maintainer hat on) asks for a
> base-installer option to allow installing buster without merged-usr.

No, he did not mention dpkg.  But as always, please provide a patch and
answer kibi's questions.

Regards,
Bastian

-- 
First study the enemy.  Seek weakness.
		-- Romulan Commander, "Balance of Terror", stardate 1709.2

Message #39 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Guillem Jover <guillem@debian.org>

To: Bastian Blank <waldi@debian.org>

Cc: Ian Jackson <ijackson@chiark.greenend.org.uk>, 923091@bugs.debian.org

Subject: Re: Bug#923091: That merged-usr is mandatory is RC

Date: Wed, 15 May 2019 13:52:16 +0200

On Mon, 2019-05-13 at 17:24:55 +0200, Bastian Blank wrote:
> On Mon, May 13, 2019 at 11:22:35AM +0100, Ian Jackson wrote:
> > In #923091, Guillem (with dpkg maintainer hat on) asks for a
> > base-installer option to allow installing buster without merged-usr.
> 
> No, he did not mention dpkg.

Indeed I didn't, even though (as I've mentioned on d-d) I consider any
system installed with such layout completely broken from dpkg's PoV. I
could not be bothered filing this with an RC severity, given that AFAIR
one of the Release Team members uploaded the breaking change, and didn't
(and still do not) have the energy or will to get into such an argument
over severities and similar.

I do still consider this pretty much unsupported from dpkg's PoV though…

Thanks,
Guillem

Message #44 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Sam Hartman <hartmans@debian.org>

To: Ian Jackson <ijackson@chiark.greenend.org.uk>

Cc: debian-release@lists.debian.org, 923091@bugs.debian.org, debian-ctte@lists.debian.org

Subject: Re: That merged-usr is mandatory is RC

Date: Sun, 19 May 2019 07:22:08 -0400

[Message part 1 (text/plain, inline)]

>>>>> "Ian" == Ian Jackson <ijackson@chiark.greenend.org.uk> writes:

    Ian> (sending this because I got the release team address wrong) Ian
    Ian> Jackson writes ("That merged-usr is mandatory is RC"):
    >> Control: severity -1 serious
    >> 
    >> In #923091, Guillem (with dpkg maintainer hat on) asks for a
    >> base-installer option to allow installing buster without
    >> merged-usr.
    >> 
    >> Guillem filed the bug as `wishlist' but given the controversy it
    >> seems to me that it should be RC if for no other reasons than
    >> social cohesion.
    >> 
    >> CCing the TC FYI (they have already been involved in merged-usr
    >> debates via #914897) and the release team, in case they have an
    >> opinion.  FAOD I am not a maintainer of base-files but AFAICT the
    >> base-files maintainer has not expressed an opinion about
    >> severity.

I've been debating doing this, but continue to believe that it's
important after several days of pondering.  So, per constitution 5.1
(2), I'd like to explicitly lend support to the idea that it would be
really good if we provide our users a way to install buster without
merged /usr.  I think that if we do not do so now, we need to be open to
the possibility that if users are stymied in doing their work, we will
need to do so in a buster point release even if we would not normally
add something some might consider a feature in a point release.

I'm not speaking to whether I think it should be RC or even whether an
expert only option is good enough.
I am simply saying that with my DPL hat on, I think this issue is
important enough it deserves real consideration.


I think that the TC's ruling and ongoing experience suggests we have
carefully considered how we want to approach merged /usr for our own
internal work developing Debian and come to a position that at least for
the moment seems to be working.

What I'm most concerned about is people who use Debian to develop
software they plan to use on Debian but who are not part of Debian.
Examples of this include people within organizations who build programs
to distribute within their organization.  People who build upstream
programs using configure from source.  That sort of thing.

These people may not use packages.  These people may not use chroots.

They are our users; they are our priority.  Even if we believe using
chroots or containers would be better for them, I don't think we should
force people into changing their build processes.


I don't think we have a good idea how big the impact will be for these
users, and so, I think we should be conservative.

If we don't choose to be conservative, I think we should be extra
willing to revisit our decision if we find we are wrong.

Again, all I'm saying is that I think this issue is important enough to
consider seriously.  I am not in a position to balance this issue
against other things before us.
I'm speaking as the DPL because I'm trying to consider something that is
a project level concern.  However, this statement has no actual force as
clearly spelled out in the constitution.
I'm speaking in the hopes of getting people to take a moment, think
about this issue and come to their own conclusions.


--Sam

[signature.asc (application/pgp-signature, inline)]

Message #49 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Colin Watson <cjwatson@debian.org>

To: Sam Hartman <hartmans@debian.org>, 923091@bugs.debian.org

Cc: Ian Jackson <ijackson@chiark.greenend.org.uk>, debian-release@lists.debian.org, debian-ctte@lists.debian.org, Guillem Jover <guillem@debian.org>, Cyril Brulebois <kibi@debian.org>

Subject: Re: Bug#923091: That merged-usr is mandatory is RC

Date: Mon, 20 May 2019 13:55:20 +0100

On Sun, May 19, 2019 at 07:22:08AM -0400, Sam Hartman wrote:
> >>>>> "Ian" == Ian Jackson <ijackson@chiark.greenend.org.uk> writes:
> 
>     Ian> (sending this because I got the release team address wrong) Ian
>     Ian> Jackson writes ("That merged-usr is mandatory is RC"):
>     >> In #923091, Guillem (with dpkg maintainer hat on) asks for a
>     >> base-installer option to allow installing buster without
>     >> merged-usr.
>     >> 
>     >> Guillem filed the bug as `wishlist' but given the controversy it
>     >> seems to me that it should be RC if for no other reasons than
>     >> social cohesion.
>     >> 
>     >> CCing the TC FYI (they have already been involved in merged-usr
>     >> debates via #914897) and the release team, in case they have an
>     >> opinion.  FAOD I am not a maintainer of base-files but AFAICT the
>     >> base-files maintainer has not expressed an opinion about
>     >> severity.
> 
> I've been debating doing this, but continue to believe that it's
> important after several days of pondering.  So, per constitution 5.1
> (2), I'd like to explicitly lend support to the idea that it would be
> really good if we provide our users a way to install buster without
> merged /usr.  I think that if we do not do so now, we need to be open to
> the possibility that if users are stymied in doing their work, we will
> need to do so in a buster point release even if we would not normally
> add something some might consider a feature in a point release.
> 
> I'm not speaking to whether I think it should be RC or even whether an
> expert only option is good enough.
> I am simply saying that with my DPL hat on, I think this issue is
> important enough it deserves real consideration.

I've posted
https://salsa.debian.org/installer-team/base-installer/merge_requests/1
to add a low-priority question for this, following Cyril's advice in
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=923091#10.  The text
may not be quite perfect, but I think it's a decent start.

I've deliberately avoided changing the default behaviour in this patch:
its effect is simply to make the behaviour configurable either via
expert mode or using preseeding (by setting base-installer/usrmerge to
false).  This is to maximise the chance of being able to get this change
into buster with a minimum of controversy of its own.  It is of course
simple to change the default behaviour and/or how prominently the
question is presented by way of follow-up changes, if the project so
chooses.

-- 
Colin Watson                                       [cjwatson@debian.org]

Message #54 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Ivo De Decker <ivodd@debian.org>

To: Sam Hartman <hartmans@debian.org>

Cc: Ian Jackson <ijackson@chiark.greenend.org.uk>, debian-release@lists.debian.org, 923091@bugs.debian.org, debian-ctte@lists.debian.org

Subject: Re: That merged-usr is mandatory is RC

Date: Wed, 29 May 2019 11:12:14 +0200

Hi,

Given that there is still discussion about the impact of merged /usr at this
very late point of the freeze, I think having merged /usr by default for new
installations should be reconsidered.

On Sun, May 19, 2019 at 07:22:08AM -0400, Sam Hartman wrote:
> I've been debating doing this, but continue to believe that it's
> important after several days of pondering.  So, per constitution 5.1
> (2), I'd like to explicitly lend support to the idea that it would be
> really good if we provide our users a way to install buster without
> merged /usr.  I think that if we do not do so now, we need to be open to
> the possibility that if users are stymied in doing their work, we will
> need to do so in a buster point release even if we would not normally
> add something some might consider a feature in a point release.
> 
> I'm not speaking to whether I think it should be RC or even whether an
> expert only option is good enough.
> I am simply saying that with my DPL hat on, I think this issue is
> important enough it deserves real consideration.
> 
> 
> I think that the TC's ruling and ongoing experience suggests we have
> carefully considered how we want to approach merged /usr for our own
> internal work developing Debian and come to a position that at least for
> the moment seems to be working.
> 
> What I'm most concerned about is people who use Debian to develop
> software they plan to use on Debian but who are not part of Debian.
> Examples of this include people within organizations who build programs
> to distribute within their organization.  People who build upstream
> programs using configure from source.  That sort of thing.
> 
> These people may not use packages.  These people may not use chroots.

People who develop software often do this on different machines than the one
the software runs on. When the production server gets upgraded, and a new
development machine is installed, one will have merged /usr and the other
doesn't. This probably isn't very good. Having an option to change this during
the install probably won't help these users.

In general, I think that if merged /usr is the default for new installations
for a Debian release, it should be the default on upgrades to that release as
well. This is not the case for buster. Obviously changing the default on
upgrades needs carefull planning and should be started at the beginning of a
release cycle.

> They are our users; they are our priority.  Even if we believe using
> chroots or containers would be better for them, I don't think we should
> force people into changing their build processes.
> 
> 
> I don't think we have a good idea how big the impact will be for these
> users, and so, I think we should be conservative.
> 
> If we don't choose to be conservative, I think we should be extra
> willing to revisit our decision if we find we are wrong.

Please note that there were a number of bugs triggered by merged /usr that
were discovered during the freeze. Most of them were actual bugs in the
packages, but they were (only) triggered with merged /usr. The fact that they
were only discovered late in the release cycle isn't a good sign.

> Again, all I'm saying is that I think this issue is important enough to
> consider seriously.  I am not in a position to balance this issue
> against other things before us.
> I'm speaking as the DPL because I'm trying to consider something that is
> a project level concern.  However, this statement has no actual force as
> clearly spelled out in the constitution.
> I'm speaking in the hopes of getting people to take a moment, think
> about this issue and come to their own conclusions.

Having an option to allow experienced user to change the default doesn't
really solve this. So the way forward is to change the default back to not
having merged /usr on new installs.

Thanks,

Ivo

Message #59 received at 923091@bugs.debian.org (full text, mbox, reply):

From: Sam Hartman <hartmans@debian.org>

To: Ivo De Decker <ivodd@debian.org>

Cc: Sam Hartman <hartmans@debian.org>, Ian Jackson <ijackson@chiark.greenend.org.uk>, debian-release@lists.debian.org, 923091@bugs.debian.org, debian-ctte@lists.debian.org

Subject: Re: That merged-usr is mandatory is RC

Date: Wed, 29 May 2019 07:31:16 -0400

>>>>> "Ivo" == Ivo De Decker <ivodd@debian.org> writes:

    Ivo> Hi, Given that there is still discussion about the impact of
    Ivo> merged /usr at this very late point of the freeze, I think
    Ivo> having merged /usr by default for new installations should be
    Ivo> reconsidered.

What discussion are you seeing other than this discussion here?
Things seem to have been fairly quiet on the merged /usr front since the
 TC decision.

What am I missing?

Send a report that this bug log contains spam.