9:26 a.m.
Hi all,
Today, I noticed that mock build root prepared by DNF is significantly
larger then prepared by YUM (see attached logs). Owners of packages
installed into minimal buildroot probably wants to review their
dependency chain.
I also reported the issue against DNF [1] in case DNF guys wants to
improve this situation.
Vít
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634
9:32 a.m.
Dne 18.8.2015 v 16:26 Vít Ondruch napsal(a):
Hi all,
Today, I noticed that mock build root prepared by DNF is significantly
larger then prepared by YUM (see attached logs). Owners of packages
installed into minimal buildroot probably wants to review their
dependency chain.
I also reported the issue against DNF [1] in case DNF guys wants to
improve this situation.
Vít
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634
Sorry, I attached wrong yum.log ... but you can find the correct one in
the bug referenced above.
Vít
9:52 a.m.
On 2015-08-18 16:32, Vít Ondruch wrote:
Dne 18.8.2015 v 16:26 Vít Ondruch napsal(a):
> Hi all,
>
> Today, I noticed that mock build root prepared by DNF is significantly
> larger then prepared by YUM (see attached logs). Owners of packages
> installed into minimal buildroot probably wants to review their
> dependency chain.
>
> I also reported the issue against DNF [1] in case DNF guys wants to
> improve this situation.
>
>
> Vít
>
>
>
> [1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634
>
Sorry, I attached wrong yum.log ... but you can find the correct one in
the bug referenced above.
Vít
The cause might be soft dependencies. dnf installs Recommends, while yum
ignores them. I think some packages in base buildroot are already using
them (gdb).
Michael
1:54 p.m.
On Tue, Aug 18, 2015 at 04:52:10PM +0200, Michael Šimáček wrote:
>>Today, I noticed that mock build root prepared by DNF is
significantly
>>larger then prepared by YUM (see attached logs). Owners of packages
[...]
The cause might be soft dependencies. dnf installs Recommends, while
yum ignores them. I think some packages in base buildroot are
already using them (gdb).
Hmmm. I think that for the buildroot case, we probably want DNF to
_not_ installer Recommended packages.
--
Matthew Miller
<mattdm(a)fedoraproject.org>
Fedora Project Leader
5:34 p.m.
Do we have a switch or configuration option for that yet?
On Aug 18, 2015 5:27 PM, "Heiko Adams" <ml(a)fedora-blog.de> wrote:
Am Dienstag, den 18.08.2015, 14:54 -0400 schrieb Matthew Miller:
> Hmmm. I think that for the buildroot case, we probably want DNF to
> _not_ installer Recommended packages.
That makes sense to me.
--
Regards,
Heiko Adams
--
devel mailing list
devel(a)lists.fedoraproject.org
https://admin.fedoraproject.org/mailman/listinfo/devel
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
3:28 a.m.
Dne 18.8.2015 v 16:26 Vít Ondruch napsal(a):
Hi all,
Today, I noticed that mock build root prepared by DNF is significantly
larger then prepared by YUM (see attached logs). Owners of packages
installed into minimal buildroot probably wants to review their
dependency chain.
I also reported the issue against DNF [1] in case DNF guys wants to
improve this situation.
Vít
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634
Please note that the issue was already resolved in mock upstream, but
you need to have DNF 1.1 to make it work, if I am not mistaken.
Vít
8:18 a.m.
Dne 18.8.2015 v 16:26 Vít Ondruch napsal(a):
Hi all,
Today, I noticed that mock build root prepared by DNF is significantly
larger then prepared by YUM (see attached logs). Owners of packages
installed into minimal buildroot probably wants to review their
dependency chain.
I also reported the issue against DNF [1] in case DNF guys wants to
improve this situation.
Vít
[1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634
This change does not look to be reflected on Koji, so I opened
https://fedorahosted.org/rel-eng/ticket/6330
Vít
10:03 a.m.
Dne 12.1.2016 v 15:18 Vít Ondruch napsal(a):
Dne 18.8.2015 v 16:26 Vít Ondruch napsal(a):
> Hi all,
>
> Today, I noticed that mock build root prepared by DNF is significantly
> larger then prepared by YUM (see attached logs). Owners of packages
> installed into minimal buildroot probably wants to review their
> dependency chain.
>
> I also reported the issue against DNF [1] in case DNF guys wants to
> improve this situation.
>
>
> Vít
>
>
>
> [1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634
>
This change does not look to be reflected on Koji, so I opened
https://fedorahosted.org/rel-eng/ticket/6330
Vít
So this request was rejected:
Changes (by ausil):
* status: new => closed
* resolution: => wontfix
Comment:
after thinking about this some I think that all deps should be installed
into the buildroot. I am going to close this as won't fix. This really
needs to have more thought. and discussion before anything is done. there
is work, tradeoffs and unexpected results regardless of what is done.
So it appears this thread was probably not enough. Which keeps us with
interesting state where mock by default does not install weak
dependencies where Koji installs them. It causes interesting issues already.
Not sure if I should prepare my packages to be ready or not ....
Vít
10:05 a.m.
Dne 25.1.2016 v 17:03 Vít Ondruch napsal(a):
Dne 12.1.2016 v 15:18 Vít Ondruch napsal(a):
> Dne 18.8.2015 v 16:26 Vít Ondruch napsal(a):
>> Hi all,
>>
>> Today, I noticed that mock build root prepared by DNF is significantly
>> larger then prepared by YUM (see attached logs). Owners of packages
>> installed into minimal buildroot probably wants to review their
>> dependency chain.
>>
>> I also reported the issue against DNF [1] in case DNF guys wants to
>> improve this situation.
>>
>>
>> Vít
>>
>>
>>
>> [1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634
>>
> This change does not look to be reflected on Koji, so I opened
>
> https://fedorahosted.org/rel-eng/ticket/6330
>
>
> Vít
So this request was rejected:
Changes (by ausil):
* status: new => closed
* resolution: => wontfix
Comment:
after thinking about this some I think that all deps should be installed
into the buildroot. I am going to close this as won't fix. This really
needs to have more thought. and discussion before anything is done. there
is work, tradeoffs and unexpected results regardless of what is done.
So it appears this thread was probably not enough. Which keeps us with
interesting state where mock by default does not install weak
dependencies where Koji installs them. It causes interesting issues already.
Not sure if I should prepare my packages to be ready or not ....
Vít
Actually there is the upstream Koji ticket opened:
https://fedorahosted.org/koji/ticket/326#comment:4
So there is still some hope :)
Vít
11:09 a.m.
W dniu 25.01.2016 o 17:03, Vít Ondruch pisze:
So it appears this thread was probably not enough. Which keeps us
with
interesting state where mock by default does not install weak
dependencies where Koji installs them. It causes interesting issues already.
mock/koji not installing weak dependencies == anything wanting ruby
being broken.
Reason: "ruby" suggests "rubypick" which suggests "ruby".
Packages buildrequire "ruby" but do not get "rubypick" installed (or
if
they are lucky they get) so are unable to find Ruby because there is no
"/bin/ruby" executable.
https://bugzilla.redhat.com/show_bug.cgi?id=1300669 shows how it look on
aarch64.
Not sure if I should prepare my packages to be ready or not ....
"rubypick" is yours ;D
11:58 a.m.
mån 2016-01-25 klockan 18:09 +0100 skrev Marcin Juszkiewicz:
W dniu 25.01.2016 o 17:03, Vít Ondruch pisze:
> So it appears this thread was probably not enough. Which keeps us with
> interesting state where mock by default does not install weak
> dependencies where Koji installs them. It causes interesting issues already.
mock/koji not installing weak dependencies == anything wanting ruby
being broken.
Reason: "ruby" suggests "rubypick" which suggests "ruby".
Packages buildrequire "ruby" but do not get "rubypick" installed (or
if
they are lucky they get) so are unable to find Ruby because there is no
"/bin/ruby" executable.
If ruby needs ruby-pick to work, then ruby-pick must not be a weak
dependency of ruby, but a hard one.
The koji buildroot really should only install hard dependencies. The
buildroot is supposed to be the minimal possible set needed to build
the package. If a package that would be installed as a weak dependency
of one of the build dependencies is needed to build the package, that
packa is a build dependency too.
Mattias
2:19 a.m.
Dne 25.1.2016 v 18:58 Mattias Ellert napsal(a):
mån 2016-01-25 klockan 18:09 +0100 skrev Marcin Juszkiewicz:
> W dniu 25.01.2016 o 17:03, Vít Ondruch pisze:
>> So it appears this thread was probably not enough. Which keeps us with
>> interesting state where mock by default does not install weak
>> dependencies where Koji installs them. It causes interesting issues already.
> mock/koji not installing weak dependencies == anything wanting ruby
> being broken.
Yep, I know. Believe or not, this is precisely how I found out that Koji
installs the weak dependencies while mock does not.
>
> Reason: "ruby" suggests "rubypick" which suggests
"ruby".
>
> Packages buildrequire "ruby" but do not get "rubypick" installed
(or if
> they are lucky they get) so are unable to find Ruby because there is no
> "/bin/ruby" executable.
If ruby needs ruby-pick to work, then ruby-pick must not be a weak
dependency of ruby, but a hard one.
The thing is that ruby does not need rubypick to work, but you have to
call ruby-mri instead of ruby in this case. This of course allows you to
avoid rubypick on your system in case you really know what you are doing.
For some moment, I was thinking to add rubypick as dependency of
ruby-devel, but decided against it (at least for the time being), since
there are several reasons for the situation and multiple ways how to fix
it. Just several examples which comes to my mind:
1) There are chances that the build does not need the ruby executable at
all, since it just want to have headers for build.
2) The configure script can be configured what Ruby executable should be
called.
3) There can be explicit BR: /usr/bin/ruby
I'd love to take this opportunity to gain knowledge how all these
projects are actually using Ruby.
Vít
4:18 a.m.
> > So it appears this thread was probably not enough. Which
keeps us with
> > interesting state where mock by default does not install weak
> > dependencies where Koji installs them. It causes interesting issues already.
>
> mock/koji not installing weak dependencies == anything wanting ruby
> being broken.
>
> Reason: "ruby" suggests "rubypick" which suggests
"ruby".
>
> Packages buildrequire "ruby" but do not get "rubypick" installed
(or if
> they are lucky they get) so are unable to find Ruby because there is no
> "/bin/ruby" executable.
If ruby needs ruby-pick to work, then ruby-pick must not be a weak
dependency of ruby, but a hard one.
The koji buildroot really should only install hard dependencies. The
buildroot is supposed to be the minimal possible set needed to build
the package. If a package that would be installed as a weak dependency
of one of the build dependencies is needed to build the package, that
packa is a build dependency too.
That's incorrect. EG a package has optional Ruby bindings, it doesn't
need ruby to build but if ruby isn't present the ruby binding sub
package is empty. The buildroot should have all that is needed to
build the desired functionality, nothing more nothing less.
Peter
4:35 a.m.
On Tue, 26 Jan 2016 10:18:08 +0000
Peter Robinson <pbrobinson(a)gmail.com> wrote:
>> > So it appears this thread was probably not enough.
Which keeps
>> > us with interesting state where mock by default does not install
>> > weak dependencies where Koji installs them. It causes
>> > interesting issues already.
>>
>> mock/koji not installing weak dependencies == anything wanting ruby
>> being broken.
>>
>> Reason: "ruby" suggests "rubypick" which suggests
"ruby".
>>
>> Packages buildrequire "ruby" but do not get "rubypick"
installed
>> (or if they are lucky they get) so are unable to find Ruby because
>> there is no "/bin/ruby" executable.
>
> If ruby needs ruby-pick to work, then ruby-pick must not be a weak
> dependency of ruby, but a hard one.
>
> The koji buildroot really should only install hard dependencies. The
> buildroot is supposed to be the minimal possible set needed to build
> the package. If a package that would be installed as a weak
> dependency of one of the build dependencies is needed to build the
> package, that packa is a build dependency too.
That's incorrect. EG a package has optional Ruby bindings, it doesn't
need ruby to build but if ruby isn't present the ruby binding sub
package is empty. The buildroot should have all that is needed to
build the desired functionality, nothing more nothing less.
+1, distribution buildsystem should include all needed deps, including
weak, to build full featured result, but the users then can decide if
they want the bare minimum or to include some the weak deps.
Dan
1:27 p.m.
tis 2016-01-26 klockan 10:18 +0000 skrev Peter Robinson:
> > > So it appears this thread was probably not enough.
Which keeps us with
> > > interesting state where mock by default does not install weak
> > > dependencies where Koji installs them. It causes interesting issues
already.
> >
> > mock/koji not installing weak dependencies == anything wanting ruby
> > being broken.
> >
> > Reason: "ruby" suggests "rubypick" which suggests
"ruby".
> >
> > Packages buildrequire "ruby" but do not get "rubypick"
installed (or if
> > they are lucky they get) so are unable to find Ruby because there is no
> > "/bin/ruby" executable.
>
> If ruby needs ruby-pick to work, then ruby-pick must not be a weak
> dependency of ruby, but a hard one.
>
> The koji buildroot really should only install hard dependencies. The
> buildroot is supposed to be the minimal possible set needed to build
> the package. If a package that would be installed as a weak dependency
> of one of the build dependencies is needed to build the package, that
> packa is a build dependency too.
That's incorrect. EG a package has optional Ruby bindings, it doesn't
need ruby to build but if ruby isn't present the ruby binding sub
package is empty. The buildroot should have all that is needed to
build the desired functionality, nothing more nothing less.
Peter
I agree that it is a reasonable expectation that installing "ruby"
should make a /usr/bin/ruby binary available. The fact that it
currently doesn't when weak dependencies are not installed is a
packaging bug in ruby, which should be addressed. It is not a reason to
bloat every koji build root for every build of every package in Fedora
with every weak dependency of every build dependency of the package
being built.
Mattias
7:13 a.m.
Dne 26.1.2016 v 20:27 Mattias Ellert napsal(a):
tis 2016-01-26 klockan 10:18 +0000 skrev Peter Robinson:
>>>> So it appears this thread was probably not enough. Which keeps us with
>>>> interesting state where mock by default does not install weak
>>>> dependencies where Koji installs them. It causes interesting issues
already.
>>> mock/koji not installing weak dependencies == anything wanting ruby
>>> being broken.
>>>
>>> Reason: "ruby" suggests "rubypick" which suggests
"ruby".
>>>
>>> Packages buildrequire "ruby" but do not get "rubypick"
installed (or if
>>> they are lucky they get) so are unable to find Ruby because there is no
>>> "/bin/ruby" executable.
>> If ruby needs ruby-pick to work, then ruby-pick must not be a weak
>> dependency of ruby, but a hard one.
>>
>> The koji buildroot really should only install hard dependencies. The
>> buildroot is supposed to be the minimal possible set needed to build
>> the package. If a package that would be installed as a weak dependency
>> of one of the build dependencies is needed to build the package, that
>> packa is a build dependency too.
> That's incorrect. EG a package has optional Ruby bindings, it doesn't
> need ruby to build but if ruby isn't present the ruby binding sub
> package is empty. The buildroot should have all that is needed to
> build the desired functionality, nothing more nothing less.
>
> Peter
I agree that it is a reasonable expectation that installing "ruby"
should make a /usr/bin/ruby binary available. The fact that it
currently doesn't when weak dependencies are not installed is a
packaging bug in ruby, which should be addressed. It is not a reason to
bloat every koji build root for every build of every package in Fedora
with every weak dependency of every build dependency of the package
being built.
When you install "ruby" package on your system, the "/usr/bin/ruby"
command is indeed installed on your system. The question is what should
be installed when you install ruby-devel package.
Vít
3010
days inactive
3172
days old
17 comments
10 participants
participants (10)
-
Dan Horák -
Heiko Adams -
Jan Zelený -
Marcin Juszkiewicz -
Matthew Miller -
Mattias Ellert -
Michael Šimáček -
Neal Gompa -
Peter Robinson -
Vít Ondruch