is external talloc and external tdb a requirement for building samba with ctdb

Matthieu Patou mat at samba.org
Sun Sep 16 12:36:13 MDT 2012


On 09/16/2012 10:53 AM, Andrew Bartlett wrote:
> On Sun, 2012-09-16 at 10:35 -0700, Matthieu Patou wrote:
>> On 09/16/2012 09:44 AM, Andrew Bartlett wrote:
>>> On Sun, 2012-09-16 at 08:57 +0200, Michael Adam wrote:
>>>> On 2012-09-15 at 23:20 -0700, Matthieu Patou wrote:
>>>>> Hello,
>>>>>
>>>>> I was trying to build samba with the ctdb support from the top waf build
>>>>> and it fails to detect my ctdb because the include for <tdb.h> and
>>>>> <talloc.h> failed.
>>>>> In theory if I'm building them internally the internal path should be
>>>>> added during the test for detection of ctdb no ?
>>>> I guess so.
>>>>
>>>> In the autobuild "samba-ctdb" target, I in fact compile, and
>>>> install tdb first (system talloc not being required by ctdb
>>>> master, apparently) and run samba configure with
>>>> "--bundled-libraries=!tdb".
>>>>
>>>> But we should find a way to fix that to work against built-in
>>>> paths.
>>> To avoid duplicate symbols, you would need to have both built against a
>>> 'system' talloc/tdb.
>> It seems that on sn-devel we have already a system talloc that's why we
>> might only need to specify --bunled-libraries=!tdb
>>>     Indeed, given how it will be used, ctdb should be
>>> the one insisting on this.
>> Can you develop this point ? what prevents samba to use the internal
>> library once we fix the waf code for the detection of ctdb.h ?
>> Because here ihmo the problem is that when we are building with the
>> bundled library we don't add the search path for the bundled lib into
>> waf's snippet.
> TDB and talloc both have bad failure modes (hopefully prevented/detected
> by our duplicate symbol detection code) if the static symbols end up
> twice in a single process.
>
> That is why with Openchange and CTDB, we can't use as 'internal' talloc
> or tdb.
Ok then I suspect that we should change the ctdb test to insure first 
that we have system tdb/talloc and if not to print an error message, as 
far as I'm concerned I was able to dig into the waf's tests to discover 
the need but a more explicit message can't be bad.

Matthieu.


-- 
Matthieu Patou
Samba Team
http://samba.org



More information about the samba-technical mailing list