[gentoo-user] Unwanted dependecies

[gentoo-user] Unwanted dependecies

[Jure Varlec]

[-- Attachment #1: Type: text/plain, Size: 2621 bytes --]

Hello,

I apologize if this was previously discussed, but I couldn't find anything on 
gmane. And maybe I should post this on gentoo-devel, but I'm not going to 
molest developers until I understand portage better. Knowing there are users 
here with good knowledge of portage I thought I would ask here first.

For the past two weeks, I have been removing pieces of gnome from my system. I 
never used it, but some libraries and programs found their way in by being 
dependencies of some stuff I used (I'm especially angry with myself for not 
noticing I had openoffice compiled with USE="eds"). So I used 
second-to-the-next-best-thing-after-sliced-bread: equery. Great tool. A tad 
slow, but it gets the job done. When I asked it to show me which packages 
depended on a certain gnome lib, it dutifully presented me with the answer. 
The trick is, those packages were compiled with USE="-gnome". And looking at 
their ebuilds, dependency handling was correct and they did not depend on 
those libs. 

So portage never built this packages with gnome support. So why did they 
depend on gnome libs? I now believe that the libs were installed prior to the 
package in question, and the package's configure script detected that and 
compiled the package against the lib. This may be wrong; in that case, 
someone please explain what's going on. But if it's true, it raises two more 
questions.

Firstly, looking at /usr/bin/equery, it seems it uses portage and gentoolkit 
magic to extract package dependencies. I don't have enough time and python-fu 
to fight my way through all of the code, there's way too much stuff in all 
the modules imported. But I'd like to know: how, in principle, does equery 
(and, supposedly, portage) know that a certain package depends on a lib, when 
the decision to depend on it was made by the configure script independent of 
what portage was told to do via USE flags or hard dependencies?

And secondly: why the heck does portage allow configure autodetection? Why 
don't ebuilds specify explicitly, depending on USE flags, what to build and 
what not to build?

And to finish the story, I played around with equery long enough to find 
everything I deemed to be part of gnome, removed it, and rebuilt everything 
that was unjustly depending on it. Without changing the USE flags, of 
course -- they've never been set anyway. Emerge never pulled anything back 
in, and now everything works beautifully, telling me I am right at least in 
some respects. But I'd like to have the situation cleared up, so I'm turning 
to you.

Thank you
Jure

[-- Attachment #2: Type: application/pgp-signature, Size: 191 bytes --]

Re: [gentoo-user] Unwanted dependecies

[Bo Ørsted Andresen]

[-- Attachment #1: Type: text/plain, Size: 3251 bytes --]

On Friday 23 June 2006 14:35, Jure Varlec wrote:
> I apologize if this was previously discussed, but I couldn't find anything
> on gmane. And maybe I should post this on gentoo-devel, but I'm not going
> to molest developers until I understand portage better. Knowing there are
> users here with good knowledge of portage I thought I would ask here first.

Good call. :)

[SNIP]
> When I asked it to show me which
> packages depended on a certain gnome lib, it dutifully presented me with the
> answer. The trick is, those packages were compiled with USE="-gnome". And
> looking at their ebuilds, dependency handling was correct and they did not
> depend on those libs.

How did you determine that? Did the ebuild inherit any eclasses and did you 
look in those? Also make sure you have >=gentoolkit-0.2.2.

> So portage never built this packages with gnome support. So why did they
> depend on gnome libs? I now believe that the libs were installed prior to
> the package in question, and the package's configure script detected that
> and compiled the package against the lib. This may be wrong; in that case,
> someone please explain what's going on. But if it's true, it raises two
> more questions.

Remember USE flags are for *optional* dependencies. Mandatory dependencies 
will remain no matter what USE flags you set.

[SNIP]
> But I'd like to know: how, in principle, 
> does equery (and, supposedly, portage) know that a certain package depends
> on a lib, when the decision to depend on it was made by the configure
> script independent of what portage was told to do via USE flags or hard
> dependencies?

The decision should not be made independent of what portage was told. If it is 
that would be a bug in the ebuild.

> And secondly: why the heck does portage allow configure autodetection? Why
> don't ebuilds specify explicitly, depending on USE flags, what to build and
> what not to build?

It does specify dependencies explicitly in the ebuilds depending on USE flags. 
If it does not it is a bug.

> And to finish the story, I played around with equery long enough to find
> everything I deemed to be part of gnome, removed it, and rebuilt everything
> that was unjustly depending on it. Without changing the USE flags, of
> course -- they've never been set anyway. Emerge never pulled anything back
> in, and now everything works beautifully, telling me I am right at least in
> some respects. But I'd like to have the situation cleared up, so I'm
> turning to you.

Portage cannot prevent an autoconfigure script from detecting stuff that the 
ebuild does not depend on. But as stated above it would be regarded as a bug 
when it does. The devs have two options to fix such a bug. Either patch the 
configure script to disable detection of additional stuff (when USE flag is 
disabled) or add the detected stuff as mandatory dependencies.

This also explains why applications that depend on gnome-python-extras like 
e.g. meld have ridiculous dependencies like totem and nautilus-burner. To get 
rid of those dependencies someone must patch the configure script of 
gnome-python-extras to allow USE flags to disable them with switches.

HtH

-- 
Bo Andresen

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Unwanted dependecies

[Jure Varlec]

[-- Attachment #1: Type: text/plain, Size: 4753 bytes --]

Thank you for replying. I see now that I should clarify further what is 
bothering me, as you didn't quite see my point. I guess it got lost in the 
noise. No more stories then :)

>Remember USE flags are for *optional* dependencies. Mandatory dependencies 
>will remain no matter what USE flags you set.

I understand. That is not the problem.

>How did you determine that? Did the ebuild inherit any eclasses and did you 
>look in those? Also make sure you have >=gentoolkit-0.2.2.

Gentoolkit is 0.2.2. After equery, I took a look at the ebuild.
As for the eclasses, I must admit I'm still not too comfortable with them, 
which is what I was actually referring to when I said I need to understand 
portage better. It's a shame really considering that I've been using gentoo 
for two-and-a-half years. I need to get spanked ;) . But as far as I 
understand it (and from what I can see looking at the eclasses) they are 
merely collections of useful shell functions which devs can use when writing 
ebuilds.

Anyway, to get rid of unnecessary confusion, I will give you an example 
package which I remember, and whose ebuild is simple enough so that it does 
not include any complicated inherits. In fact, it is short enough to be 
included in this mail.

> # Copyright 1999-2006 Gentoo Foundation
> # Distributed under the terms of the GNU General Public License v2
> # $Header: /var/cvsroot/gentoo-x86/media-gfx/gqview/gqview-2.0.1.ebuild,v
> 1.8 2006/02/21 21:31:07 mr_bones_ Exp $
>
> DESCRIPTION="A GTK-based image browser"
> HOMEPAGE="http://gqview.sourceforge.net/"
> SRC_URI="mirror://sourceforge/gqview/${P}.tar.gz"
>
> LICENSE="GPL-2"
> SLOT="0"
> KEYWORDS="alpha amd64 ia64 ppc ppc64 sparc x86"
> IUSE=""
>
> RDEPEND=">=x11-libs/gtk+-2.4.0
>         virtual/libintl"
>
> DEPEND="${RDEPEND}
>         sys-devel/gettext"
>
> src_install() {
>         make DESTDIR="${D}" install || die "make install failed"
>         # leave README uncompressed because the program reads it
>         dodoc AUTHORS ChangeLog TODO
>         rm -f "${D}/usr/share/doc/${PF}/COPYING"
> }

As you can see, it does not inherit anything, and does not depend on anything 
gnomish. However, when I checked with equery, it definitely depended on 
something which I didn't need/want (sorry, I can't remember what it was, 
probably gconf or some such), although the ebuild specifies no such 
dependency, not even indirectly (through gtk+ etc.)  After I removed the 
unwanted dep and remerged gqview, it did not pull the removed "dependency" 
back in, and no longer depends on it.

The same applies for dependencies which were disabled through use flags, such 
as this snip (from openoffice):

>         gnome? ( >=x11-libs/gtk+-2.4
>
>                 >=gnome-base/gnome-vfs-2.6
>                 >=gnome-base/gconf-2.0 )

In this case, It depended on gnome-vfs and gconf, even when the gnome USE flag 
was disabled. After I removed the offending packages and recompiled 
openoffice, the dependecy is gone.
Note: you might remember I wrote that oo.org used to be compiled with 
USE="eds", which pulls these in along with other stuff indirectly through 
gnome-extra/evolution-data-server. It was the reason these libs were 
installed in the first place. The dependency problem I'm describing, however, 
is independent from that and was present even after I recompiled oo.org 
without "eds".

I hope this helped you better understand what is bothering me. It's not that 
I'm particularly against having those libs installed and wouldn't mind to 
keep them if they were hard deps of something I needed, but I'm kind of 
frustrated when packages depend on things I told them not to depend on 
through USE flags. It means I cant test a package with a hard dep on a lib 
because in case I emerge something while that lib is installed, I'll never 
know if it will choose to depend on that lib, which in turn would mean that 
after I remove the lib, I need to recompile stuff, once more. In the past two 
weeks, I recompiled openoffice 4 times, which is 3 times more than I'm 
normally willing to do it. This is not to say I would be less frustrated if 
only small packages were affected :)

Looking at oo.org ebuild again and I see that it uses use_enable function in 
the call to ./configure. So why the heck did it depend on gconf and gnome-vfs 
when it was obviously told not to? And even more confusing: how did equery 
know about this dependency?

*Sigh*
I hope I'm missing something here. Because if these are bugs, there's quite a 
few. Sadly, I removed all the temporary lists of packages I removed and 
rebuilt, so I have nothing to report now. Stupid me.

Jure



[-- Attachment #2: Type: application/pgp-signature, Size: 191 bytes --]

Re: [gentoo-user] Unwanted dependecies

[Neil Bothwick]

[-- Attachment #1: Type: text/plain, Size: 716 bytes --]

On Fri, 23 Jun 2006 17:51:16 +0200, Jure Varlec wrote:

> Note: you might remember I wrote that oo.org used to be compiled with 
> USE="eds", which pulls these in along with other stuff indirectly
> through gnome-extra/evolution-data-server. It was the reason these libs
> were installed in the first place. The dependency problem I'm
> describing, however, is independent from that and was present even
> after I recompiled oo.org without "eds".

How did you disable that USE flag? AIUI equery works with global USE
flags, so if emerge --info still shows eds, equery will think OOo depends
on it, no matter what you have in /etc/portage.


-- 
Neil Bothwick

This screen intentionally left blank.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Unwanted dependecies

[Bo Ørsted Andresen]

[-- Attachment #1: Type: text/plain, Size: 443 bytes --]

On Friday 23 June 2006 18:34, Neil Bothwick wrote:
> AIUI equery works with global USE
> flags, so if emerge --info still shows eds, equery will think OOo depends
> on it, no matter what you have in /etc/portage.

I am pretty sure that it uses /var/db/pkg/${category}/${name}/USE to determine 
the USE flags that were used when the package was compiled. Also `equery 
uses` is affected by /etc/portage/package.use.

-- 
Bo Andresen

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Unwanted dependencies

[Benno Schulenberg]

Bo Ørsted Andresen wrote:
> On Friday 23 June 2006 18:34, Neil Bothwick wrote:
> > AIUI equery works with global USE
> > flags, so if emerge --info still shows eds, equery will think
> > OOo depends on it, no matter what you have in /etc/portage.
>
> I am pretty sure that it uses /var/db/pkg/${category}/${name}/USE
> to determine the USE flags that were used when the package was
> compiled.

It should, but it doesn't do that.  For example, on my system I have 
no hal:

$ eix -e hal | grep Installed
     Installed:           none

But:

$ equery depends hal
[ Searching for packages depending on hal... ]
gnome-base/gnome-vfs-2.12.2
app-cdr/k3b-0.12.15

It apparently produces any package that either depends on hal, or 
_conditionally_ depends on hal.  It clearly doesn't take USE flags 
into account.

$ emerge --info | grep hal
[nothing]

$ equery --version
equery(0.1.4) - Gentoo Package Query Tool
Author(s): Karl Trygve Kalleberg

The previous version didn't list any conditional dependency, even 
when the USE flag was set and the package installed, the current 
version lists these conditional deps always.

These bugs look relevant:
https://bugs.gentoo.org/show_bug.cgi?id=81012
https://bugs.gentoo.org/show_bug.cgi?id=124053

Benno
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Unwanted dependencies

[Jure Varlec]

[-- Attachment #1: Type: text/plain, Size: 302 bytes --]

On Friday 23 June 2006 23:16, Benno Schulenberg wrote:
>
> These bugs look relevant:
> https://bugs.gentoo.org/show_bug.cgi?id=81012
> https://bugs.gentoo.org/show_bug.cgi?id=124053
>
> Benno

Ah, thank you. So people are aware of this. I need to search more thoroughly 
next time :/

Jure

[-- Attachment #2: Type: application/pgp-signature, Size: 191 bytes --]

Re: [gentoo-user] Unwanted dependecies

[Jure Varlec]

[-- Attachment #1: Type: text/plain, Size: 1012 bytes --]

On Friday 23 June 2006 18:34, Neil Bothwick wrote:
> On Fri, 23 Jun 2006 17:51:16 +0200, Jure Varlec wrote:
> > Note: you might remember I wrote that oo.org used to be compiled with
> > USE="eds", which pulls these in along with other stuff indirectly
> > through gnome-extra/evolution-data-server. It was the reason these libs
> > were installed in the first place. The dependency problem I'm
> > describing, however, is independent from that and was present even
> > after I recompiled oo.org without "eds".
>
> How did you disable that USE flag? AIUI equery works with global USE
> flags, so if emerge --info still shows eds, equery will think OOo depends
> on it, no matter what you have in /etc/portage.

It's disabled in make.conf. I have a policy to disable global flags there, and 
local flags in package.use. It's just that I missed eds when I went through 
global flags, and I obviously didn't pay attention when I emerged oo.org. Not 
that it matters for the problem at hand.

Jure

[-- Attachment #2: Type: application/pgp-signature, Size: 191 bytes --]

Re: [gentoo-user] Unwanted dependecies

[Bo Ørsted Andresen]

[-- Attachment #1: Type: text/plain, Size: 3205 bytes --]

On Friday 23 June 2006 17:51, Jure Varlec wrote:
> Thank you for replying. I see now that I should clarify further what is
> bothering me, as you didn't quite see my point. I guess it got lost in the
> noise. No more stories then :)

It was not lost. Just questioned...

[SNIP]

> Gentoolkit is 0.2.2. After equery, I took a look at the ebuild.
> As for the eclasses, I must admit I'm still not too comfortable with them,
> which is what I was actually referring to when I said I need to understand
> portage better. It's a shame really considering that I've been using gentoo
> for two-and-a-half years. I need to get spanked ;) . But as far as I
> understand it (and from what I can see looking at the eclasses) they are
> merely collections of useful shell functions which devs can use when
> writing ebuilds.

ebuilds inherit dependencies and USE flags from the eclasses too.

[SNIP]

> As you can see, it does not inherit anything, and does not depend on
> anything gnomish. However, when I checked with equery, it definitely
> depended on something which I didn't need/want (sorry, I can't remember
> what it was, probably gconf or some such), although the ebuild specifies no
> such dependency, not even indirectly (through gtk+ etc.)  After I removed
> the unwanted dep and remerged gqview, it did not pull the removed
> "dependency" back in, and no longer depends on it.

Hard to comment on since you don't remember..

> The same applies for dependencies which were disabled through use flags,
> such
>
> as this snip (from openoffice):
> >         gnome? ( >=x11-libs/gtk+-2.4
> >
> >                 >=gnome-base/gnome-vfs-2.6
> >                 >=gnome-base/gconf-2.0 )
>
> In this case, It depended on gnome-vfs and gconf, even when the gnome USE
> flag was disabled. After I removed the offending packages and recompiled
> openoffice, the dependecy is gone.
> Note: you might remember I wrote that oo.org used to be compiled with
> USE="eds", which pulls these in along with other stuff indirectly through
> gnome-extra/evolution-data-server. It was the reason these libs were
> installed in the first place. The dependency problem I'm describing,
> however, is independent from that and was present even after I recompiled
> oo.org without "eds".

I do have openoffice and gnome-vfs too. equery depends gnome-vfs does report 
openoffice for me despite the fact that the gnome use flag is disabled. So I 
tried removing gnome-vfs from /var/db/pkg/app-office/openoffice*/{R,}DEPEND. 
That stopped equery from showing openoffice with equery depends gnome-vfs. I 
also tried removing gnome-vfs and running revdep-rebuild -p. This did not 
report any broken dependencies from openoffice. So I think it is very 
possible that there are still bugs in equery... I don't know though how 
equery works though so it's just a guess.

[SNIP]

> *Sigh*
> I hope I'm missing something here. Because if these are bugs, there's quite
> a few. Sadly, I removed all the temporary lists of packages I removed and
> rebuilt, so I have nothing to report now. Stupid me.

While I think those are bugs they are really minor.

-- 
Bo Andresen

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Unwanted dependecies

[Jure Varlec]

[-- Attachment #1: Type: text/plain, Size: 1255 bytes --]

On Friday 23 June 2006 19:40, Bo Ørsted Andresen wrote:
> I do have openoffice and gnome-vfs too. equery depends gnome-vfs does
> report openoffice for me despite the fact that the gnome use flag is
> disabled. So I tried removing gnome-vfs from
> /var/db/pkg/app-office/openoffice*/{R,}DEPEND. That stopped equery from
> showing openoffice with equery depends gnome-vfs. I also tried removing
> gnome-vfs and running revdep-rebuild -p. This did not report any broken
> dependencies from openoffice. So I think it is very possible that there are
> still bugs in equery... I don't know though how equery works though so it's
> just a guess.

Hm. I guess it's not hard to imagine equery fails to read USE flags correctly 
and that would make it a bug. But this still doesn't explain gqview, which 
doesn't have neither hard nor USE dependencies on anything I removed.

Obviously I have to do more research, especially since I haven't any data to 
support me. And I thought cleaning out unneeded libs would be a fun thing to 
do in between the exams :) . Meh. Unless someone picks this up, I'll have 
more time in a month or so. I'm definitely not going to file any bugs or some 
such until I know more.

Thank you for your input
Jure

[-- Attachment #2: Type: application/pgp-signature, Size: 191 bytes --]

Re: [gentoo-user] Unwanted dependecies

[Bo Ørsted Andresen]

[-- Attachment #1: Type: text/plain, Size: 587 bytes --]

On Friday 23 June 2006 20:41, Jure Varlec wrote:
> Hm. I guess it's not hard to imagine equery fails to read USE flags
> correctly and that would make it a bug.

Benno has now provided references that shows that `equery depends` was not 
designed to consider use flags. Enhancement bugs has already been filed.

> But this still doesn't explain 
> gqview, which doesn't have neither hard nor USE dependencies on anything I
> removed.

By default equery shows only installed packages. So if you said you removed 
those dependencies that might explain it.

-- 
Bo Andresen

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]