On Jun 23, 2009, at 9:06 AM, Hyrum K. Wright wrote:
> On Jun 22, 2009, at 12:44 PM, Hyrum K. Wright wrote:
>
>> On Jun 21, 2009, at 5:48 PM, Arfrever Frehtes Taifersar Arahesis
>> wrote:
>>
>>> 2009-06-21 23:25:15 Greg Stein napisał(a):
>>>> On Sun, Jun 21, 2009 at 03:40, Arfrever Frehtes Taifersar
>>>> Arahesis<Arfrever.FTA_at_gmail.com> wrote:
>>>>> ...
>>>>>> If expat was being found just fine before r38120
>>>>>
>>>>> It was being found only because APR-Util was accidentally linked
>>>>> against Expat
>>>>> and apu-1-config was providing path to Expat directory which was
>>>>> used during
>>>>> building of APR-Util.
>>>>
>>>> That is NOT an "accident". That is entirely by design.
>>>
>>> So the design is broken.
>>> Even libaprutil-1.so isn't linked against libraries printed by
>>> `apu-1-config --libs`, so linking against these libraries would slow
>>> down
>>> Subversion at run-time (ld.so would have to load attitional
>>> libraries).
>>>
>>> $ apu-1-config --libs
>>> -lldap -llber -llber -ldb-4.7 -lgdbm -lexpat
>>> $ scanelf -qF '%F: %n' /usr/lib64/libaprutil-1.so
>>> /usr/lib64/libaprutil-1.so: libexpat.so.1,libapr-1.so.0,libcrypt.so.
>>> 1,libpthread.so.0,libc.so.6
>>>
>>>> APRUTIL is *supposed* to be linked against Expat on the system, and
>>>> SVN is *supposed* to use that fact to find Expat for its own usage.
>>>>
>>>>> ...
>>>>>> Seems like a regression in functionality to me...
>>>>>
>>>>> Actually it's an enhancement :) .
>>>>
>>>> If I'm required to pass more stuff on the command line, then it
>>>> is a
>>>> *regression*. Period. That is the very definition of regression.
>>>>
>>>> And if that regression causes a build break, then you better get
>>>> this
>>>> fixed before somebody decides to revert it. The last time you broke
>>>> our build, we said "it is fair to revert when a change breaks the
>>>> build". You've now broken the build, so it is just a small matter
>>>> of
>>>> time before somebody reverts this change. You have until then to
>>>> fix
>>>> it.
>>>
>>> Fixed in r38132.
>>
>> The buildbot is still failing to configure. r38120 and friends have
>> broken the build system, and will be reverted.
>
> The buildbot is still failing, this time in search of zlib.
Per our policy of reverting breakage to the build system, I reverted
r38152, r38133, r38132, and r38120 in r38164. If there is a real bug
to be solved here, hopefully we can still fix it without causing
regressions.
-Hyrum
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2364575
Received on 2009-06-23 19:45:44 CEST