Your point makes sense. We attempt to build as little as possible to get python messages generated (otherwise doc jobs will take forever), and the case of export flags requiring a package to be built wasn't one I came across in testing. I've made some modifications to the message generation code, not requiring export flags to pass in order to run message generation. Now, your job is stable:

On Tue, Nov 20, 2012 at 1:24 AM, Thomas Moulard <thomas.moulard@gmail.com> wrote:

Dear all,
since the new indexer has been released, vision_visp documentation is
in an unstable state[1].

>From my understanding, it is due to the fact that either the package
are not built anymore or that
rosmake is not called.

Here is the failure:
Failed to invoke /opt/ros/fuerte/bin/rospack
cflags-only-I;--deps-only visp_bridge
sh: 1: /home/rosbuild/hudson/workspace/doc-fuerte-vision_visp/doc_stacks/2012-11-19_15-04-56.298782/vision_visp/visp/install/bin/visp-config:
not found

This happens when cmake is invoked in the first package of the stack.
This behavior is expected as the dependencies are not built before
calling cmake on this package.
In particular, this package, visp_bridge, depends on another package
in this stack called visp.
The visp package uses a binary to retrieve its cflags/lflags:

(visp/manifest.xml)
<export>
<cpp
cflags="`${prefix}/install/bin/visp-config --cflags`"
lflags="`${prefix}/install/bin/visp-config --libs`"
/>
</export>

So, of course, one has to build first visp before asking for its export flags.
IMHO this seems an acceptable behavior in the sense that these flags
may vary depending on your
building flags and your system.

Is my diagnostic correct, and is there a way to solve this issue?

Thanks,

[1] http://jenkins.willowgarage.com:8080/job/doc-fuerte-vision_visp/lastBuild/console
--
Thomas Moulard
_______________________________________________
ros-users mailing list
ros-users@code.ros.org
https://code.ros.org/mailman/listinfo/ros-users

--
Eitan Marder-Eppstein

President, hiDOF Inc.

http://www.hidof.com