Harness constantly prints boost::filesystem errors and sometimes aborts

[dobey]

terminate called after throwing an instance of 'boost::filesystem::filesystem_error'
  what(): boost::filesystem::equivalent: No such file or directory: "/tmp/tmpvb08fgci/.local/share/unity-scopes/unconfined/client/logs", "/tmp/tmp9ha9gfc6/.local/share/unity-scopes/unconfined/client/logs"
scoperegistry: no remote registry configured, only local scopes will be available
Aborted

When running the tests locally, many similar messages to this are repeated, but often without aborting. On the jenkins armhf builder, it seems to cause an abort fairly consistently, due to this unhandled exception. This also causes the python3 process to hang, which causes further errors to show up in the logs, as the chroot apparently cannot unmount /run/shm when this happens.

[Michi Henning (michihenning)]

Any chance of a stack trace? That would be really useful. How can I reproduce this?

Looking at the code, I think the issue here may be that the registry is running in a non-standard location and there is no unity-scopes directory under .local/share (or .local/share itself doesn't exist).

Pawel ran into something similar yesterday when running in a chroot. I think I have a fair idea of what's going on, but would still appreciate a stack trace, if possible. In particular, I'm interested in termination due to the uncaught exception. That clearly is a bug in our code but, without a stack trace, it might be difficult to identify the exact source.

I'll see what I can learn from the code.

[Michi Henning (michihenning)]

I believe the linked branch should fix this. Could you give it a spin please?

I still would very much appreciate seeing a stack trace, particularly for the terminate() problem.

[Michi Henning (michihenning)]

Rodney, did the branch fix the problem for you?

[dobey (dobey)]

As the branch was targeted and landed in /devel, I haven't been able to
properly test it (as the failure is only happening in Jenkins), but
given that the failure is about the logs directory missing, and the
branch creates them, I'm confident it will solve the problem. It just
needs to get into the archive that CI builds against, to be usable
there, and unblock my branch (though due to final freeze, I expect it to
be blocked a bit longer now anyway).

On Thu, 2015-04-16 at 08:47 +0000, Michi Henning wrote:
> Rodney, did the branch fix the problem for you?
>