On 12/12/13 12:00, Yann Rouillard wrote:
No significant answer from Oracle for 3 weeks now.
I am asking for update on a weekly basis but I don't have matter to increase the priority of this issue as there is not production impact and an easy workaround.

I do think we will eventually get an answer.

Yann



2013/12/12 Dagobert Michelsen via buildfarm <[email protected] <mailto:[email protected]>>

    Hi Jake,

    Am 12.12.2013 um 19:53 schrieb Jake Goerzen via buildfarm:
    > On 11/13/13 06:16, Dagobert Michelsen via buildfarm wrote:
    >> Am 13.11.2013 um 10:03 schrieb Laurent Blume via buildfarm
    <[email protected] <mailto:[email protected]>>:
    >>> Regularly, I'm having silly issues with linking on the
    buildfarm with different behaviour on x86 and sparc.
    >>> This time, in krb5-lib: with the same recipe, some binaries
    get linked to libintl.so on unstable10s, and they don't on
    unstable10x.
    >>> On my home system, x86, they do get linked.
    >>>
    >>> I'm noticing that ld on the buildfarm is not at all consistent:
    >>>
    >>> At home:
    >>> -rwxr-xr-x   1 root     bin        10300 janv 14  2013
    /usr/ccs/bin/ld
    >>>
    >>> unstable10s:
    >>> $ ls -l /usr/ccs/bin/ld
    >>> -rwxr-xr-x   1 root     bin        10788 Jan 16  2013
    /usr/ccs/bin/ld
    >>>
    >>> unstable10x:
    >>> $ ls -l /usr/ccs/bin/ld
    >>> -rwxr-xr-x   1 root     bin        10172 Jul  4  2011
    /usr/ccs/bin/ld
    >>>
    >>>
    >>> Since it's part of the kernel patch, I gather that unstable10x
    was kept back for some reason, as its kernel is older.
    >>>
    >>> Can unstable10x be upgraded? I am reasonably sure it would fix
    some of the linking issues I'm hitting right now.
    >>
    >> I would prefer not to unless we fully understand the issue as
    discussed on irc.
    >
    > Hi Dago,
    >
    > Has there been any update on the issue of ld being inconsistent
    on the buildfarm yet?  I have been putting off working on some
    things until a resolution has been found.

    Yann has a case open at Oracle, but I doubt we get anything useful
    out of it.
    For now I recommend just adding the extra deps and unconditionally
    overriding
    them for i386. For mid-term William told me he will get some T5220
    and he would
    be willing to give one to the project. This would allow me another
    build-only
    machine which is not going to be updated. Then we could also
    really stick to
    u8 (or u5?) for all packaging zones. But don't expect this before
    q2 2014.

    Sorry for the inconvenience

      -- Dago



Thank you all for your efforts and contributions! I will use the work around as suggested. Great idea about setting up the T5220 build-only host!

Best Regards,
/Jake

Reply via email to