Segmentation fault when subscribing to gazebo topics

[ljburtz]

Hello and thanks for this great program.
I have an issue plotting data from gazebo9.

Issue A: Seg fault when subscribing to the live gazebo simulation

Segmentation fault (Address not mapped to object [(nil)])

Subscribing to gazebo topics leads very quickly (few seconds) to a segmentation fault / core dumped : attached log
plot_juggler_fromSource.log
plot_juggler_debianPackage.log
(the logs are lauching Plotjuggler with a layout but Plot juggler also crashes if I just start it without any cli arguments and then manually choose the topics)

Issue B: Seg fault when subscribing to a rosbag made from the gazebo simulation:

• I recorded the ROS messages : recording started before launching plotjuggler, recording stopped after plotjugler crashed. (attached as .bag).

segfault_2020-07-02-12-08-07bag.zip

• Then shutdown gazebo.
• Then restart roscore.
• Then rosbag play -l segfault_2020-07-02-12-08-07.bag.
• Then launch plotjuggler: (for example with this layout: plotjuggler_groundruth_xml.zip)

There is no crash like in issue A.
But after a while (several minutes) there is the same seg fault and error output as reported in #93
(I am also using XY plot of messages in the same topic).
log attached:
plot_juggler_XYsame_topic.log
throw std::runtime_error("X and Y axis don't share the same time axis");

I could not reproduce the issues with datasamples/example.bag
I wonder if there is a race condition caused by the messages being published at very high hz and bw by gazebo.

Environment:

• Using v2.8.1 compiled from source / master branch (latest commit: 6364f73)
• Ubuntu 18 / ros melodic / gazebo9 from debian packages
• Debian package ros-melodic-ros-type-introspection is already the newest version (2.0.4-1bionic.20200530.015709)

Please let me know if this is a known issue.

Otherwise I'll be happy to provide more info / explore more simplified configurations to make it easier to reproduce. I'm not sure whether it's related to gazebo.

[ljburtz]

update about issue B:
I could also reproduce it with the .bag from issue #93
(after ~5 minutes in rosbag play loop mode)
The layout xml of issue #93 is from a version of Plotjuggler that's too old. So I used this one (xy_layout_new.xml) with random topics plotted in XY
reproduceIssue93.zip

[ljburtz]

update about issue A :
It works great with the AppImage ! No seg fault with the topics from gazebo or any that I tested so far.

So maybe some library dependency is skewed on my system?

[facontidavide]

I will investigate

[facontidavide]

I think I solved issue A. can you try the last version on master?

EDIT: I haven't been able to reproduce B, but the more I think about it, more I believe that A and B were caused by the same issue that the last commit should have solved (thread safety).

And INDEED, it is the same problem as #93 (plugin were rewritten)

[ljburtz]

Fantastic thanks for the fast fix. I've tested the latest version on master: no more issue. Happy to close the ticket!