On Tue, Dec 7, 2010 at 1:50 PM, Brian Gerkey wrote: > On Mon, Dec 6, 2010 at 9:38 AM, Jack O'Quin wrote: >> On Mon, Dec 6, 2010 at 10:55 AM, Brian Gerkey wrote: >>> To address the problem of unwanted dependencies from testing, we've >>> taken to creating separate test packages (e.g., `test_rospack` has the >>> tests for `rospack`).  That solves things at a package level [*].  But >>> you still need somewhere to put, e.g., the `test_camera1394` package. >>> If it goes in the `camera_drivers` stack, then the stack still has the >>> test-induced dependencies.  This may or may not be acceptable.  If >>> it's not, you could create a separate stack to contain the tests; I >>> don't believe that we've done that yet. >> >> I would not want it in the camera_drivers stack for the same reason it >> should not go in the camera1394 package. > >> I don't see anything in ros-pkg/stacks that looks like a collection of >> tests. I am reluctant to add a test_camera stack at that level, but >> don't have a better idea, so suggestions are welcome. Would a >> ros-pkg/stacks/tests subdirectory be awkward? > > Just having a tests subdirectory won't be that useful, because we > would need to release that code, to ensure that tests are run, in > prerelease and postrelease builds.  Otherwise, we'd have to add some > special handling to tell Hudson, e.g., also check out the tests > subdirectory and run some tests from in there. > > As hesitant as I am to suggest a potential doubling of stacks, > creating a `test_camera_drivers` stack may be the best way to go. > Hopefully this situation (i.e., the tests bring unwanted dependencies > at the stack level) is uncommon; it feels like something that will > happen most often with driver-like things. I'll leave it to Blaise and Patrick to decide if a test_camera_drivers stack is a good idea. Unless there is more justification than just some tests for one package, I doubt it would be worth the effort. I suppose a test stack could have fairly broad scope, minimizing dependencies might not matter much. --  joq