LD_LIBRARY_PATH stripped when explicitly extending workspace

[mikepurvis]

System Info

• Operating System: Linux 3.16.0-69-generic x86_64 x86_64 x86_64 GNU/Linux
• Python Version: Python 2.7.6
• Version of catkin_tools: 0.4.2-1
• ROS Distro: indigo

Build / Run Issue

I'm building two catkin workspace inside a dpkg-buildpackage environment, one which extends the other. The first one (which starts with catkin) succeeds, and then the second one bails out right at the beginning, with the first thing it tries to build, which is catkin_tools_prebuild for a linked develspace or the first actual package with isolated develspace.

Error is:

ERROR: ld.so: object 'libfakeroot-sysv.so' from LD_PRELOAD cannot be preloaded (cannot open shared object file): ignored.

[jbohren]

@mikepurvis Does this happen if both workspaces have merged or isolated develspaces?

[mikepurvis]

Both workspaces are merged installspaces, and the second extends the first's installspace. I'm working on a MWE for you.

[mikepurvis]

The issue is that the --extend option is wiping out LD_LIBRARY_PATH. On the first build, I get:

echo ${LD_PRELOAD}
libfakeroot-sysv.so
echo ${LD_LIBRARY_PATH}
/usr/lib/x86_64-linux-gnu/libfakeroot:/usr/lib64/libfakeroot:/usr/lib32/libfakeroot:

After sourcing that, I still have LD_PRELOAD set, but now:

echo $LD_LIBRARY_PATH
/home/mpurvis/rosbundles/build/clearpath-1.7pre/clearpath-1.7pre_20160420205118/debian/tmp/opt/clearpath/1.7pre/demo1/lib

So I can no longer find the specified library. I can probably hack around this by mangling the path post-sourcing, but this is something catkin_tools needs to handle better.

[jbohren]

The issue is that the --extend option is wiping out LD_LIBRARY_PATH.

So if you build with an implicit underlay do you see the same problem?

[mikepurvis]

I believe so, yeah (but I don't see the problem with 0.3.x, with either explicit or implicit extension).

[jbohren]

but I don't see the problem with 0.3.x, with either explicit or implicit extension

Alright. It would be great to have a complete out-of-the-box repro, when you have time to put it together. I suspect the resultspace loading is cleaning out variables that should be making it through.

[mikepurvis]

Here you are:

git clone --recursive https://github.com/mikepurvis/catkin_tools_350.git
cd catkin_tools_350
./run  # happiness
git clean -fdx
fakeroot ./run  # sadness

EDIT: Interesting— if I change the overlay to extend implicitly, it actually does succeed. So that may be an acceptable workaround for now.