On Wed, Feb 27, 2013 at 12:44 PM, Vincent Rabaud
<vrabaud@willowgarage.com>wrote:

> ok, not sure but this is my guess:
> - the error is happening when parsing your manifest.xml: recursively, a
> dependency is not found (rostest)
> - now, this is really weird because none of your dependencies (that are
> wet/catkinized btw) depends on rostest
>

Thanks, Vincent. I totally misunderstood the problem.

I looked into your log and added debug comments to rospack and it seems
> that the bug is from
> https://github.com/ros/rospkg/blob/master/src/rospkg/manifest.py#L383:
> dependencies are considered to be an aggregation of build+runtime. In your
> case, none of the depencies run_depend on rostest (but they build_depend).
> Therefore it does not get installed but it's sought for. Therefore, it
> crashes.
>
> BTW, you should depend on rostest too, as you call the macros yourself :)
>

So it should. Not sure why that was working in Fuerte.

I have a guess as to why it suddenly started failing in Groovy. It seems
you recently released a new image_common, version 1.10.3, which
includes camera_info_manager changes to eliminate a rostest error message
and get those unit tests working again under catkin. In image_common
1.10.0, camera_info_manager did have a rostest run_depend (which was
causing an error message).

If that is correct, why did the pre-release tests not detect that
camera1394 had broken as a side-effect?

Don't the pre-release tests still build and test all dependent packages,
like they used to?
-- 
 joq