On Tue, Dec 28, 2010 at 11:19 PM, Patrick Bouffard
<bouffard@eecs.berkeley.edu> wrote:
> Thanks Ken. I wasn't aware of that usage of roscreate-stack. BTW, I
> noticed what I guess is a bug in unstable when the --show-deps flag is
> used:
>
> $ roscreate-stack --show-deps .
> Traceback (most recent call last):
>  File "/opt/ros/unstable/ros/bin/roscreate-stack", line 35, in <module>
>    roscreate.roscreatestack.roscreatestack_main()
>  File "/opt/ros/unstable/ros/tools/roscreate/src/roscreate/roscreatestack.py",
> line 219, in roscreatestack_main
>    create_stack(stack, stack_dir, stack_manifest, author, depends,
> licenses, options.show_deps)
> UnboundLocalError: local variable 'stack_manifest' referenced before assignment

Thanks for the report.  I didn't even know about the --show-deps flag.
 I learn something new about my own code every day :).  I fixed the
issue in r12666 and added documentation on the --show-deps option to
the wiki.

> Regarding extraneous (package) dependencies, thinking about it some
> more I guess there's no real way to find that out for sure aside from
> systematically removing one at a time from package's manifest and
> trying to rebuild that package. Does that make sense?

Pretty much.  As a quicker test, r-grepping for a package name in the
source is usually a good sniff test.  Also keep in mind that roslaunch
files create implicit dependencies, so 'make' is often not a
sufficient test -- though the fastidious can use the
rosbuild_add_roslaunch_check() macro:

http://www.ros.org/wiki/rosbuild/CMakeLists#rosbuild_add_roslaunch_check

And, of course, the only way to catch Python dependencies is to write
unit tests as well.

cheers,
Ken