build directly to install space

[mathias-luedtke]

This PR picks some install space related commits from #137
The current master (still) builds all packages, cleans and rebuild to install space (unless `NOT_TEST_INSTALL=true).

These patches avoid rebuilding and fix the retry behavior.

[mathias-luedtke]

Since the mockup tests are not considered (fixed in #137), I have prepared a test brach in ros_canopen: https://travis-ci.org/ipa-mdl/ros_canopen/builds/213134116

In addtion there is a cob_robots branch: https://travis-ci.org/ipa320/cob_robots/builds/213121299
This repository is the main reason for fast-forwarding the install-space patches: Without them the travis build gets stopped because it exceeds 50 minutes.

[130s]

Is this comment correct? Isn't this line skipping those specified in $PKGS_DOWNSTREAM (which are already tested in catkin_run_tests section)?

[mathias-luedtke]

No, it just skips packages without a directory in install/share. AFAIK this is only the case for meta-packages.

[130s]

Seems like this works great!

I'm not sure if I understand correctly, but looking at https://travis-ci.org/ipa320/cob_robots/jobs/213121300, seems like in catkin_run_tests section all tests are run, and catkin_install_run_tests finished within a second (not sure why that short. Are all tests done are from downstream?). I assume with the workspace configured for install, catkin run_tests does all what we wanted?

If so, is keeping those 2 sections still worthy?
Maybe for other build tools that don't handle tests in install space (e.g. catkin_make) we can still keep the logic in catkin_install_run_tests.

[mathias-luedtke]

I don't know why, but the catkin_install_run_tests section runs all *.test files from target and downstream packages found in the install space.

We don't install *.test for the cob packages by default, so this block is a no-op.

Maybe for other build tools that don't handle install space (e.g. catkin_make)

catkin_make install would build to install space

[mathias-luedtke]

We might consider skipping catkin_install_run_tests if we don't find any *.test files.

[130s]

We might consider skipping catkin_install_run_tests if we don't find any *.test files.

I think that's the way we might want to do as long as tests get to run in catkin_run_tests.

[mathias-luedtke]

We might consider skipping catkin_install_run_tests if we don't find any *.test files.

I think that's the way we might want to do as long as tests get to run in catkin_run_tests.

Let's create another issue/PR for this and not mix it.

[130s]

#54 (comment) helped me a lot to understand what this PR does.

The current behaviour is really odd:

1. build to build/devel space
2. run build test
3. clean
4. build to build/devel space and install to install space
5. run rostest on install space

So in step 4 we are rebuilding everything from step 1, which doubles the time the build takes.

Ok, with this PR we're skipping the build in 4, is that correct?
What I was afraid was to skipping step 1 and 2 (and 3 obvisously), which could cause errors on the repos where no install rules are not defined. But looks like I was wrong and worrying about nothing.
Looking at the example ipa320/cob_robots, it still goes through all the steps with the only change is skipping the rebuild I think.

Merging. Thanks!

[130s]

#150 (review)

The current behaviour is really odd:

1. build to build/devel space
2. run build test
3. clean
4. build to build/devel space and install to install space
5. run rostest on install space

So in step 4 we are rebuilding everything from step 1, which doubles the time the build takes.

Ok, with this PR we're skipping the build in 4, is that correct?
What I was afraid was to skipping step 1 and 2 (and 3 obvisously), which could cause errors on the repos where no install rules are not defined. But looks like I was wrong and worrying about nothing.

I overlooked adding this line. With this line, users have no way to avoid testing against install space. E.g. ros-industrial-consortium/descartes#198 (comment) where the tests depend on the files that are intentionally not deployed.