In addition to solving the opencv issue, would it make sense for 
rosmake, rosdep or rospack to complain if a library appears two times in 
the exports of 2 different manifests?

On 11/04/10 12:35, Brian Gerkey wrote:
> hi Garratt,
>
> Package dependencies in the manifest are not meant to be ordered, and
> you shouldn't rely on ordering to produce a particular behavior.  In
> this case, as Radu says, the problem is that the ROS packages opencv2
> and flann conflict with each other.  That conflict should be resolved.
>
> 	brian.
>
> On Thu, Nov 4, 2010 at 11:36 AM, Radu Bogdan Rusu<rusu@willowgarage.com>  wrote:
>    
>> Garratt,
>>
>> This is a nasty problem, and we should fix the way FLANN is exposed from within OpenCV (I would argue that it shouldn't
>> really be exposed from within OpenCV in ROS, as long as we have a ROS package for FLANN). In general, this practice of
>> just "eating up libraries" -- that is, include a library (FLANN) into another library (OpenCV) should be avoided for
>> many reasons. As long as FLANN remains standalone, it should be included as a dependency through the normal
>> rosdep/depend mechanisms.
>>
>> The OpenCV case is a bit different though, because OpenCV itself is not a ROS package but an external "3rd party
>> dependency", wrapped in a ROS meta-package (opencv2). That means that it was easier for OpenCV to just copy FLANN inside
>> it rather than using some CMake/Makefile logic to search for it on the system. The latter is even more complicated as we
>> wouldn't really install FLANN on our systems globally, but use a ROS meta-package for it, so OpenCV would need to be
>> able to search through other ROS meta-packages...
>>
>>
>> Cheers,
>> Radu.
>>
>>
>> On 11/03/2010 10:44 PM, garratt wrote:
>>      
>>> Hi all,
>>>
>>> I'm using the flann library in PCL.  I recently came across a rather
>>> annoying phenomenon:  opencv also includes a version of flann that is
>>> similar enough to not generate compile errors, but causes segfaults when
>>> I run code compiled with it.  When tracking down how I started using the
>>> wrong flann, I realized that the cmake voodoo that is rosmake seems to
>>> include a library by it's _last_ mention, rather than the first:
>>>
>>> in my manifest file:
>>>
>>> <depend package="pcl"/>
>>> <depend package="depends_on_opencv"/>
>>> <depend package="pcl"/>
>>>
>>> fails, whereas:
>>>
>>> <depend package="pcl"/>
>>> <depend package="pcl"/>
>>> <depend package="depends_on_opencv"/>
>>>
>>> does not.
>>>
>>> I could solve this problem in my case, because I owned all the manifest
>>> files.  However, I can image a scenario where I need to include a
>>> package that, through the dependency ordering causes my package to
>>> include the wrong library.
>>>
>>> Is my perception of the rosmake dependency inclusion process correct? If
>>> so, maybe changing to a 'first inclusion' paradigm might be appropriate.
>>>
>>> thanks,
>>> Garratt
>>>
>>>
>>> _______________________________________________
>>> ros-users mailing list
>>> ros-users@code.ros.org
>>> https://code.ros.org/mailman/listinfo/ros-users
>>>        
>> _______________________________________________
>> ros-users mailing list
>> ros-users@code.ros.org
>> https://code.ros.org/mailman/listinfo/ros-users
>>
>>      
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>    

-- 
Dr. Cedric Pradalier
http://www.asl.ethz.ch/people/cedricp