[pacman-dev] [arch-dev-public] [arch-general] WARNING: [testing] broken due to openssl and heimdal rebuilds

[Dan McGee]

On Wed, Apr 7, 2010 at 12:31 PM, Xavier Chantry
<chantry.xavier at gmail.com> wrote:
> On Wed, Apr 7, 2010 at 6:23 PM, Dan McGee <dpmcgee at gmail.com> wrote:
>> On Wed, Apr 7, 2010 at 10:16 AM, Allan McRae <allan at archlinux.org> wrote:
>>> On 08/04/10 00:54, Florian Pritz wrote:
>>>>
>>>> On 07.04.2010 16:10, Allan McRae wrote:
>>>>>
>>>>> This would really not help here.  Pacman does not directly link openssl,
>>>>> but does through libarchive and libfetch.  Adding versioned libarchive
>>>>> and libfetch to pacman's deps and using sodeps on openssl in those
>>>>> packages would prevent pacman's SyncFirst from working.
>>>>>
>>>>
>>>> Don't add depends=(libfetch.sp libarchive.so) to the pacman PKGBUILD and
>>>> it won't be a problem. (Also see #2 below)
>>>
>>> Adding those or not is the same as adding versioned deps or not.  Both cause
>>> bad issues.
>>>
>>>>> no versioned deps in pacman for lib{archive,fetch} = bad as between the
>>>>> openssl and lib{archive,fetch} updates vercmp is broken and so is
>>>>> install files.
>>>>
>>>> sodeps would ensure that pacman updates libfetch and libarchive directly
>>>> after openssl.
>>>
>>> Not directly after, but in the same transaction.   That is the same as the
>>> current issue we have.  Using sodeps is no improvement over versioned deps
>>> here.
>>
>> Allan and I talked a bit on IRC and we both came to some idea that a
>> vercmp without any deps would probably be a good solution. We can hack
>> this up for now, file attached to do so.
>>
>> If this goes into the pacman package dir in SVN and we just add a "gcc
>> -O2 -o vercmp vercmp.c" in there and put it in the right place, we can
>> avoid a lot of hassle with everything here.
>>
>
> That sounds fine, but what is the long term solution exactly ?

Make vercmp in pacman proper build clean (e.g. no library deps) via
including the vercmp code from alpm directly without doing a
copy-paste job. I have some local testing for this in the works.

-Dan