Message Headers missing dependency information

[stonier]

It appears that published messages are missing the message definition for the child types which is used for message introspection.

ros/common_msgs#20

[valentinp]

Has there been any progress on this bug? This is particularly annoying for collecting rosbags from Android devices.

[stonier]

Rosjava is mostly following a scratch that itch development cycle at the moment. Those of us who are scratching itches right now are working at companies and can commit time to fixes where we need it or can justify it. We don't have a need for this one at our company (yet), so it's on the todo...unless someone else gets a terrific urge to scratch this itch ;)

It's probably not too hard to fix - I think you'll need to compare the way gencpp/genpy does things with the code in message_generation.

[stonier]

To summarise the discussion in the issue posted on common_msgs, a likely source is that the message_definition field specified in TCPROS Specification is not providing the dependencies definition as well as the msg definition (should exactly match a rosrun roslib gendeps --cat _msg_file_).

[gavanderhoorn]

Not a proper fix at all, but at least a work-around: as I had quite some bags that contained messages published by rosjava publishers where I ran into the exception reported in ros/common_msgs#20, I went for the nuclear option and wrote something that replaces the msg defs in connection headers in bag files with completely expanded ones.

The message definitions obviously need to be available (and identical to what was used by the publisher(s)) on the machine running the script, but that is typically the case.

See gavanderhoorn/rosbag_fixer.

---

Edit: just solved a pretty serious oversight in rosbag_fixer that caused it to not properly close the bag it writes the "fixed" connection headers to. Smaller bags would suffer data loss which sometimes caused rosbag to not be able to read any messages after running the fixer script. This should now be resolved.

[ramonss]

Is there any update on this issue? A permanent fix would be greatly appreciated. Currently, I've been having to use the rosbag_fixer quite often (strangely I do not always have the problem) when recording topics with different types of nodes (rosjava and non-rosjava nodes).

[gavanderhoorn]

I don't believe there has been any development to fix this. As @stonier writes in his comment above, rosjava itself must be changed for this.

[ramonss]

Thanks for the reply. Is there any planned update for rosjava itself? I think the current bug compromises the compatibility with non-rosjava nodes, since recording ros topics into bag files is a very basic ros feature and extensively used for testing purposes.

[gavanderhoorn]

I'm just a user, like you :)

Dependent on the rosjava gods.

[jubeira]

hi @ramonss,

Rosjava is currently developed mainly by its community; there are no developers working full time or part time on it that I know of. If you need this feature, feel free to submit a PR; I'll review it as soon as I can. If it's something that adds new features and doesn't have too much risk of breaking the existing codebase it will be merged rather fast. I've been pretty packed up lately and I don't have the time to implement this right now.

Dependent on the rosjava gods.

That certainly made me laugh 😅

[krixkrix]

@gavanderhoorn: thanks for the rosbag_fixer script, it works for me on melodic.
What about adding an option to rosbag record to use local defs? I would like to do it, but not sure if I am capable. Any hints?

[gavanderhoorn]

What about adding an option to rosbag record to use local defs?

Well, that would be adding a work-around for an issue local to rosjava to a tool that sees most of its usage outside rosjava contexts.

Unless there are other use-cases, I doubt the rosbag maintainers are going to accept something like that -- and I think I would agree with them.