On 02/07/2012 05:22 PM, Jan Kiszka wrote:
> On 2012-02-07 17:19, Gilles Chanteperdrix wrote:
>> On 02/01/2012 09:21 PM, Gilles Chanteperdrix wrote:
>>> On 02/01/2012 04:50 PM, Jan Kiszka wrote:
>>>> On 2012-02-01 16:38, Gilles Chanteperdrix wrote:
>>>>> On 02/01/2012 04:25 PM, Jan Kiszka wrote:
>>>>>> On 2012-02-01 16:17, Gilles Chanteperdrix wrote:
>>>>>>> On 02/01/2012 03:37 PM, Jan Kiszka wrote:
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> don't remember anymore: Is there any subtle reason that prevent a
>>>>>>>> change like
>>>>>>>>
>>>>>>>> diff --git a/src/skins/native/Makefile.am
>>>>>>>> b/src/skins/native/Makefile.am
>>>>>>>> index 39eaaed..4cc8859 100644
>>>>>>>> --- a/src/skins/native/Makefile.am
>>>>>>>> +++ b/src/skins/native/Makefile.am
>>>>>>>> @@ -22,6 +22,9 @@ libnative_la_SOURCES = \
>>>>>>>> wrappers.c \
>>>>>>>> wrappers.h
>>>>>>>>
>>>>>>>> +libnative_la_LIBADD = \
>>>>>>>> + ../common/libxenomai.la
>>>>>>>> +
>>>>>>>> libnative_la_CPPFLAGS = \
>>>>>>>> @XENO_USER_CFLAGS@ \
>>>>>>>> -I$(top_srcdir)/include
>>>>>>>> diff --git a/src/skins/rtdm/Makefile.am b/src/skins/rtdm/Makefile.am
>>>>>>>> index 8ad74be..2dc0a90 100644
>>>>>>>> --- a/src/skins/rtdm/Makefile.am
>>>>>>>> +++ b/src/skins/rtdm/Makefile.am
>>>>>>>> @@ -8,6 +8,9 @@ librtdm_la_SOURCES = \
>>>>>>>> core.c \
>>>>>>>> init.c
>>>>>>>>
>>>>>>>> +librtdm_la_LIBADD = \
>>>>>>>> + ../common/libxenomai.la
>>>>>>>> +
>>>>>>>> librtdm_la_CPPFLAGS = \
>>>>>>>> @XENO_USER_CFLAGS@ \
>>>>>>>> -I$(top_srcdir)/include
>>>>>>>> diff --git a/src/testsuite/latency/Makefile.am
>>>>>>>> b/src/testsuite/latency/Makefile.am
>>>>>>>> index c772c26..6534df5 100644
>>>>>>>> --- a/src/testsuite/latency/Makefile.am
>>>>>>>> +++ b/src/testsuite/latency/Makefile.am
>>>>>>>> @@ -14,5 +14,4 @@ latency_LDFLAGS = @XENO_USER_LDFLAGS@
>>>>>>>> latency_LDADD = \
>>>>>>>> ../../skins/native/libnative.la \
>>>>>>>> ../../skins/rtdm/librtdm.la \
>>>>>>>> - ../../skins/common/libxenomai.la \
>>>>>>>> -lpthread -lm
>>>>>>>>
>>>>>>>> i.e. that we let the skin libraries depend on libxenomai and then
>>>>>>>> remove
>>>>>>>> the explicit dependency from our binaries and the xeno-config output?
>>>>>>>> Is there some ordering issue again (we have -l<skin> before -lxenomai
>>>>>>>> in the ldflags).
>>>>>>>>
>>>>>>>> If possible, this would allow for things like dlopen("libnative.so").
>>>>>>>
>>>>>>> It allows xeno-config result to work both with dynamic and static
>>>>>>> libraries. Static libraries have no dependency system, so, when linking
>>>>>>> a program whith libnative.a for instance, without libtool, you still
>>>>>>> have to link it with libxenomai.a.
>>>>>>
>>>>>> OK, part two could stay, but the dependencies should still be added to
>>>>>> the skin libs - if possible.
>>>>>>
>>>>>>>
>>>>>>> How come you can not dlopen libnative.so, dlopening libxenomai.so before
>>>>>>> does not work?
>>>>>>
>>>>>> Dependencies of libnative on libxenomai are not resolved when you open
>>>>>> the former even if the latter is already loaded. Maybe you can do this
>>>>>> by pulling in all required symbols one by one manually, haven't tried
>>>>>> yet. But that would at least be unhandy.
>>>>>
>>>>> What about RTLD_GLOBAL ?
>>>>>
>>>>
>>>> Was possibly the reason, need to check back.
>>>>
>>>> Still, what prevents stating the existing dependency of lib<skin> on
>>>> libxenomai? The dance above would than be obsolete.
>>>
>>> The change is merged. I took the chance to check that static build still
>>> built.
>>>
>>
>> Unfortunately this change also causes a build failure when
>> cross-compiling for ARM, so, change reverted, too.
>>
>
> Indeed unfortunate. Any pointers to logs?
This causes a -L/usr/lib to be added on the link-edit command line,
which causes the link to fail by finding /usr/lib/libpthread.so instead
of the cross-compiler one, and fail.
--
Gilles.
_______________________________________________
Xenomai-core mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-core
