FAQ
hi,
i write this mail after more people ask me in private. and since the
centos core devel team not share any info about it's build system,
status, there modifications, doesn't have any public revision system
and not really accept any help or advice for outside.
but i still believe these are useful info for many others so i try to
summarize our rhel-6 rebuild experience.

first question is the build host and base repo selection. it's advice to
choose both the build system and the base repo as close as to rhel-6 as
possible. unfortunately even with official rhel-6 isos you can't get the
optional packages in binary. so imho the best base system and base repo
is a union of rhel-6 rc2 and rhel-6 beta. both are available freely in
binary but only the beta include the optional packages in binary from.
after you finish you can rebuild packages on itself again.

another important note that most update packages (there're a few dozens)
have higher buildreq packages then in the base repo, but these buildreqs
are not versioned in the specs. so you've to add the updates pacakges to
the build system's base packages otherwise you'll get build errors.

i recommend mock-1.1.7 for building. unfortunately at rh it's not
required to be able to build src.rpm is mock, they use only rpmbuild (!)
which of course cause a few dozen of packages won't build:-(
imho it'd be useful something like this for rhel too:
http://fedoraproject.org/wiki/FTBFS

so let we see what's are the problematic packages (fortunately most of
the packages build clean). all of these packages bellow has a bugzilla
entry and most of them already contains the solutions. so just go and
search for it (i'd not like to link all bz entry:-()


- virt-top (missing ocaml-camomile-data)
- spice-client (missing pixman-devel)
these two packages can't be rebuild on rhel-6 since they has such
buildreq which is not shipped in rhel-6 (simple rh forget about it:-) so
you can either add to the base repo these packages from epel or from a
rebuild from fedora or simple drop them.

- bash
- zsh
can't be rebuild in mock-1.x since a mock tty bug. in this case imho the
easiest solution:
mock <params> --rebuild bash...
ctrl-c during build, then
mock <params> --shell
rpmbuild --rebuild /builddir/build/originals/bash...

- nss
can't be rebuild in mock since during %check try to connect to the
localhost to test ssl, but it's not working in mock. i suggest the above
solution plus ctrl-c during the rpmbuild when the build hang at the ssl
connection test and after the you'll got a successful build:-)

- kernel
unfortunately kernel spec is not updated to the new noarch subpackage
style. which means you've to build the arch and noarch packages
separately (as in rhel-5). but the new spec file even worst then the
rhel-5 was so eg:
a simple:
mock <params> --target "noarch,i686" --rebuild kernel...
no longer works, but
mock <params> --target "noarch" --rebuild kernel...
mock <params> --target "i686" --rebuild kernel...
works:-) imho the best would be to update the kernel spec to the new
noarch style subpackage, but afais rh so conservative about kernel spec
it won't be happened in the rhel-6.x series:-(

- opal
has a wrong rh patch which cause ekiga not compile, so you've to fix
opal first.

- java-1.6.0-openjdk-1.6.0.0-1.21.b17.el6
can't be build, unfortunately bz with fixes for this package are not
public, but there's an update which can be build (may be we'd have to
leave this package out).

and the remaining packages has different problems in the spec file, but
all has (mostly a simple) solution in bz, but doesn't have update package:
- guile
- openmpi
- foomatic
- hivex
- mod_auth_kerb
- gnome-doc-utils
- pilot-link
- linuxdoc-tools
- perl-Sys-Virt
- perl-Test-MockObject
- perl-Test-Memory-Cycle
- perl-GSSAPI
- perl-Makefile-Parser
- perl-CGI-Session
- perl-IPC-Run3
- python-lxml
- python-sqlalchemy
- rome
- epydoc
- kdepim-runtime

i hope it can help for a few poeple.

anyway happy xmas!

--
Levente "Si vis pacem para bellum!"

Search Discussions

  • Seth Vidal at Dec 24, 2010 at 10:30 am

    On Fri, 24 Dec 2010, Farkas Levente wrote:
    i recommend mock-1.1.7 for building. unfortunately at rh it's not
    required to be able to build src.rpm is mock, they use only rpmbuild (!)
    which of course cause a few dozen of packages won't build:-(
    imho it'd be useful something like this for rhel too:
    http://fedoraproject.org/wiki/FTBFS
    I don't know where you get your info but everything is built in mock.

    -sv
  • Farkas Levente at Dec 24, 2010 at 10:38 am

    On 12/24/2010 04:30 PM, Seth Vidal wrote:
    On Fri, 24 Dec 2010, Farkas Levente wrote:
    i recommend mock-1.1.7 for building. unfortunately at rh it's not
    required to be able to build src.rpm is mock, they use only rpmbuild (!)
    which of course cause a few dozen of packages won't build:-(
    imho it'd be useful something like this for rhel too:
    http://fedoraproject.org/wiki/FTBFS
    I don't know where you get your info but everything is built in mock.
    see comment 4:
    https://bugzilla.redhat.com/show_bug.cgi?ida3392#c4

    anyway if thwy would use mock then can't be so much problems and if rh
    try to rebuild the it's release on itself FTBFS (i assume they've got
    enough cpu cycle:-) then no package can be missing from the release.


    --
    Levente "Si vis pacem para bellum!"
  • Josh Boyer at Dec 24, 2010 at 10:50 am

    On Fri, Dec 24, 2010 at 10:38 AM, Farkas Levente wrote:
    On 12/24/2010 04:30 PM, Seth Vidal wrote:

    On Fri, 24 Dec 2010, Farkas Levente wrote:
    i recommend mock-1.1.7 for building. unfortunately at rh it's not
    required to be able to build src.rpm is mock, they use only rpmbuild (!)
    which of course cause a few dozen of packages won't build:-(
    imho it'd be useful something like this for rhel too:
    http://fedoraproject.org/wiki/FTBFS
    I don't know where you get your info but everything is built in mock.
    see comment 4:
    https://bugzilla.redhat.com/show_bug.cgi?ida3392#c4

    anyway if thwy would use mock then can't be so much problems and if rh
    try to rebuild the it's release on itself FTBFS (i assume they've got
    enough cpu cycle:-) then no package can be missing from the release.
    There is nothing that says all BuildRequires packages have to be
    shipped in the release. It's only important to them that they can
    build RHEL, not anyone else.

    josh
  • R P Herrold at Dec 24, 2010 at 11:26 am

    On Fri, 24 Dec 2010, Josh Boyer wrote:

    enough cpu cycle:-) then no package can be missing from the release.
    There is nothing that says all BuildRequires packages have
    to be shipped in the release. It's only important to them
    that they can build RHEL, not anyone else.
    True enough -- closed and complete 'self-hosting' has never
    been a goal upstream in their enterprise product; I understand
    the wish of those here (and even more strongly in EPEL which
    this piece was crossposted to as well ...) to have it, however

    include my stock IAAL disclaimer, and also not sent from a
    centos email address

    For those components of the upstream RHEL sources that are
    GPLv2 licensed (not all of them, by a growing margin) and to
    the people to whom binary content is distributed, this is
    possibly not the case

    GPLv2, section 3
    a) Accompany it with the complete corresponding
    machine-readable source code

    [later un-numbered para portion] ... For an executable
    work, complete source code means all the source code for all
    modules it contains, plus any associated interface definition
    files, plus the scripts used to control compilation and
    installation of the executable

    Assuming, arguendo that there are 'tweaks' from Release
    Engineering, or such to make mock stand up and dance, on the
    offending 'make check' absence of PTY [Clark William's comment
    on one], or such [or a need to manually 'walk a build through'
    outside of mock -- I make no such representation, and my
    approach on distribution buildsystems is not CentOS' present
    one], ...

    ... the build system deviations from what is stock 'mock'
    would appear to constitute either part of the 'complete source
    code' to attain a binary, or alternatively part of 'the
    scripts used to control compilation' ...

    and be part of the releasables in proper and a somewhat
    narrow set of cases

    * shrug *

    and GPLv3 does not have such an argument chain available in it
    that I can see presently

    The bugs referenced in this thread here and on the EPEL
    mailing list, seem to refer to Artistic licensed code, however

    As to builders, I looked, and upstream released its v 6
    product beta in late July, and the gold release drop in early
    November. This new major release, with the changes I've noted
    before on one of the mailing lists on the difficulty of doing
    a yum migration from 5 to 6 should be taken to confirm that
    it is indeed challenging to solve it all

    I've privately solved subsets of the upstream's 6 sources, not
    for CentOS but for other purposes, and it can be tricky ;) I
    guess it is time to write my annual buildsystems piece. If
    bugs in this regard are filed in the CentOS tracker, or in the
    upstream tracker on such problems, and you do no see me 'cc
    into them, please feel free to send a private email to me at
    herrold owlriver com
    and I'll look in on them

    -- Russ herrold
  • Hubert Bahr at Dec 27, 2010 at 2:03 am

    Farkas Levente wrote:
    hi,
    i write this mail after more people ask me in private. and since the
    centos core devel team not share any info about it's build system,
    status, there modifications, doesn't have any public revision system
    and not really accept any help or advice for outside.
    but i still believe these are useful info for many others so i try to
    summarize our rhel-6 rebuild experience.
    Thanks for the summary of your experience. In my case I created a build
    environment much as you laid out but I am still doing everything as
    rpmbuild --rebuild rather than a Mock environment. I ended up with 36
    packages not building and did have to import a couple of fedora packages
    to get it to that number. In comparing my list to yours, you have
    solutions to 11 of those on my list. I will be attempting those when I
    have the time. My question is this. Are these the only packages you
    were unable to rebuild, or do you still have some left to resolve? I
    noticed a few of mine failed with unpackaged elements so I assume those
    may be resolved with a change in build options. I used to believe that
    if I just found the right environment all of RedHat's sources would
    rebuild. I now have found out that, this belief was very naive.

    Thanks again
    Hubert
  • Farkas Levente at Dec 27, 2010 at 4:01 pm

    On Mon, Dec 27, 2010 at 08:03, Hubert Bahr wrote:
    Farkas Levente wrote:
    hi,
    i write this mail after more people ask me in private. and since the
    centos core devel team not share any info about it's build system,
    status, there modifications, doesn't have any public revision system
    and not really accept any help or advice for outside.
    but i still believe these are useful info for many others so i try to
    summarize our rhel-6 rebuild experience.
    Thanks for the summary of your experience. ?In my case I created a build
    environment much as you laid out but I am still doing everything as
    rpmbuild --rebuild rather than a Mock environment. ?I ended up with 36
    packages not building and did have to import a couple of fedora packages
    to get it to that number. ?In comparing my list to yours, you have
    solutions to 11 of those on my list. ?I will be attempting those when I
    have the time. My question is this. ?Are these the only packages you
    were unable to rebuild, or do you still have some left to resolve? ?I
    noticed a few of mine failed with unpackaged elements so I assume those
    may be resolved with a change in build options. ?I used to believe that
    if I just found the right environment all of RedHat's sources would
    rebuild. ?I now have found out that, this belief was very naive.
    everything else build, except that i've to find the reason why yum
    wouldn't like to install 32 bit packages into a 64bit system in mock
    (but this is bot an rh problem).


    --
    ? Levente? ? ? ? ? ? ? ? ? ? ? ? ? ? ?? "Si vis pacem para bellum!"
  • Leonard den Ottolander at Dec 27, 2010 at 4:13 pm
    Hello Farkas,
    On Mon, 2010-12-27 at 22:01 +0100, Farkas Levente wrote:
    everything else build, except that i've to find the reason why yum
    wouldn't like to install 32 bit packages into a 64bit system in mock
    (but this is bot an rh problem).
    Perhaps you are referring to some packages that have conflicting
    documentation files for the i686 and x86_64 versions? That is a problem
    with doxygen creating slightly different doc files for both versions.

    I couldn't tell you how to fix that for mock, but on a plain build
    system you can safely force install those rpms (after you verified it's
    only the doc files that are clashing of course).

    Regards,
    Leonard.

    --
    mount -t life -o ro /dev/dna /genetic/research
  • Hubert Bahr at Jan 1, 2011 at 1:58 am

    Farkas Levente wrote:
    hi,
    i write this mail after more people ask me in private. and since the
    centos core devel team not share any info about it's build system,
    status, there modifications, doesn't have any public revision system
    and not really accept any help or advice for outside.
    but i still believe these are useful info for many others so i try to
    summarize our rhel-6 rebuild experience.

    - nss
    can't be rebuild in mock since during %check try to connect to the
    localhost to test ssl, but it's not working in mock. i suggest the above
    solution plus ctrl-c during the rpmbuild when the build hang at the ssl
    connection test and after the you'll got a successful build:-)
    I tried this with both versions of nss currently out and do not get the
    hang on the ssl connection. I get a report of 5 errors and then rebuild
    ends since errors are greater than 0. I do not understand the
    significance of 5 errors out of 5000+tests.
    - opal
    has a wrong rh patch which cause ekiga not compile, so you've to fix
    opal first.


    - epydoc
    My interpretation of the above two is not that the don't build, but they
    build with an error that cause other packages not to build. Is it your
    intent that we use these patched packages to replace the defective ones
    in our repository and then build ekiga or python-nss.
    i hope it can help for a few poeple.

    anyway happy xmas!
    Thanks for your info I have now reduced my unbuilt packages down to 10.
    I haven't used all your patches yet so I expect to reduce them still
    further.
    Hubert
  • Akemi Yagi at Jan 1, 2011 at 3:33 am

    On Fri, Dec 31, 2010 at 10:58 PM, Hubert Bahr wrote:
    Farkas Levente wrote:
    - opal
    has a wrong rh patch which cause ekiga not compile, so you've to fix
    opal first.
    My interpretation of the above two is not that the don't build, but they
    build with an error that cause other packages not to build. ?Is it your
    intent that we use these patched packages to replace the defective ones
    in our repository and then build ekiga or python-nss.
    This is the bugzilla Levente filed:

    https://bugzilla.redhat.com/show_bug.cgi?idf1769

    Akemi
  • Hubert Bahr at Jan 3, 2011 at 2:33 am

    Farkas Levente wrote:
    hi,
    i write this mail after more people ask me in private. and since the
    centos core devel team not share any info about it's build system,
    status, there modifications, doesn't have any public revision system
    and not really accept any help or advice for outside.
    but i still believe these are useful info for many others so i try to
    summarize our rhel-6 rebuild experience.

    - nss
    can't be rebuild in mock since during %check try to connect to the
    localhost to test ssl, but it's not working in mock. i suggest the above
    solution plus ctrl-c during the rpmbuild when the build hang at the ssl
    connection test and after the you'll got a successful build:-)
    In an earlier reply I indicated that instead of a hang the build
    indicated 5 errors. I changed the spec file to allow up to 5 errors and
    it completed the build.
    so now I am down to one unbuilt src.rpm
    and the remaining packages has different problems in the spec file, but
    all has (mostly a simple) solution in bz, but doesn't have update package:

    - linuxdoc-tools
    This is it. I am afraid I am extremely rusty on spec files and while I
    tried to follow the bz instructions, I don't know whether I improved
    things or made them worse. While I managed to get rid of the not found
    errors, I ended up with a long list of unpackaged files. Could I get a
    copy of your spec file for linuxdoc-tools or a diff/patch file from the
    distribution spec.

    Thanks
    Hubert
  • Leonard den Ottolander at Jan 3, 2011 at 4:35 am
    Hello Hubert,
    On Mon, 2011-01-03 at 01:33 -0600, Hubert Bahr wrote:
    I changed the spec file to allow up to 5 errors and
    it completed the build.
    That sounds ugly. If you do this kind of patching you should patch out
    the specific test failures, not set a number of random tests that are
    allowed to fail. Just a thought.

    Regards,
    Leonard.

    --
    mount -t life -o ro /dev/dna /genetic/research
  • Hubert Bahr at Jan 4, 2011 at 1:44 am

    Leonard den Ottolander wrote:
    Hello Hubert,
    On Mon, 2011-01-03 at 01:33 -0600, Hubert Bahr wrote:

    I changed the spec file to allow up to 5 errors and
    it completed the build.
    That sounds ugly. If you do this kind of patching you should patch out
    the specific test failures, not set a number of random tests that are
    allowed to fail. Just a thought.
    All a matter of perspective. This is one of over 2000 source packages
    to build binary's and it passes 99.9% of the test cases, and this 5
    character change to a specfile allowed it to build. The previous
    alternative was to kill it during a test and then completing the build.
    The other one was to skip the testing in total. There are 433 bugs
    filed against nss. Some were just expired certificates causing
    problems. I chose not worry about it unless I run into further specific
    problems that point me back to nss. In this case I just wanted to have
    a baseline should I have need to modify nss source (unlikely unless it
    is for branding purposes). If I make an error that increases the count
    above current value, I would still be alerted. My intuition says it is
    probably a problem with the test case, not a problem with the source or
    build technique. I would like to hear from anybody who has built
    rhel-6 nss from RedHat's source rpms, on their results and what
    technique they applied.

    Hubert
    Regards,
    Leonard.
  • Farkas Levente at Jan 3, 2011 at 6:44 am

    On 01/03/2011 08:33 AM, Hubert Bahr wrote:
    - linuxdoc-tools
    This is it. I am afraid I am extremely rusty on spec files and while I
    tried to follow the bz instructions, I don't know whether I improved
    things or made them worse. While I managed to get rid of the not found
    errors, I ended up with a long list of unpackaged files. Could I get a
    copy of your spec file for linuxdoc-tools or a diff/patch file from the
    distribution spec.

    --
    Levente "Si vis pacem para bellum!"
    -------------- next part --------------
    An embedded and charset-unspecified text was scrubbed...
    Name: linuxdoc-tools.spec
    Url: http://lists.centos.org/pipermail/centos-devel/attachments/20110103/801c2910/attachment.pl
  • Hubert Bahr at Jan 4, 2011 at 2:10 am

    Farkas Levente wrote:
    On 01/03/2011 08:33 AM, Hubert Bahr wrote:

    Thank you for the Spec file. I was a lot closer than I thought I
    was. It solved my problem. I believe I have now built all of Red
    Hat's source rpms. I will now attempt to scrip everything, put
    to-geather a repository based on these builds and rebuild everything
    again just to verify I have remembered everything. In my case I use
    rpmbuild rebuild for about 99% of the build and mock for the
    remainder. This speeds up the process signifantly on my cluster.
    Thanks again
    Hubert

Related Discussions

Discussion Navigation
viewthread | post
Discussion Overview
groupcentos-devel @
categoriescentos
postedDec 24, '10 at 9:38a
activeJan 4, '11 at 2:10a
posts15
users7
websitecentos.org
irc#centos

People

Translate

site design / logo © 2022 Grokbase