correctly initialise Ogre types and check return value of 'transform'

[christian-rauch]

Description

In some sections of the code, the Ogre types where not correctly initialized and return values have not been checked. This PR resolves these issues to prevent crashes due to invalid marker scales from uninitialized memory.

[rhaschke]

Thanks for this contribution. However, I think it will be sufficient to return on failure. Can you confirm?

[rhaschke]

Wouldn't it suffice to have this line fixed, i.e. just return on failure?
All the uninitialized variables wouldn't be used in this case.

[christian-rauch]

Yes, in this case it would be sufficient to return and keep the values uninitialised.
But uninitialised values are the root of evil. I usually initialise values on their creating as a good habit and the performance costs of initialising 3-element vectors is neglectable.

[rhaschke]

@christian-rauch, could you provide a marker message causing rviz to fail on this?
From my understanding, transform() shouldn't fail here, since the validity of the transform is checked in advance by a TF message filter. So, I'm curious, which message causes this issue.

[christian-rauch]

The issue was related to the transforms, not the marker per se. Due to a misconfiguration from my side, I had two roots in the TF tree and transform returned false at

rviz/src/rviz/default_plugin/markers/marker_base.cpp

Line 93 in f9b357f

if (!context_->getFrameManager()->transform(message->header.frame_id, stamp, message->pose, pos, orient))

. This caused the function to return without setting scale.

[rhaschke]

My point is, that if the transform is missing, the message filter should bail out before already.
If that's not happening, we have another issue.

[christian-rauch]

I understand. But I am unable to reproduce this issue since I fixed the original problem in the URDF and a couple of other places.

[rhaschke]

Superseded by #1400. @christian-rauch, please validate.