During a recent internal meeting, we (the IOL team) were discussing bringing new test cases into the CI environment. We found that many of the test cases which recompile DPDK seem to break testpmd for all of the tests after them. As far as we can tell, this is because some of these tests leave testpmd in a location different than the default once they are compiled, resulting in the executable not being able to be found. We wanted to reach out and ask if the skip setup flag ('-s', '--skip-setup') was intended to stop recompilation of dpdk during the time that test cases are running. If it is, then it seems many of the tests which recompile dpdk do not follow this convention, and if it is not then those test cases might need to be fixed. It is also possible that there is a misconfiguration or misunderstanding of the root cause on our side. If it is not one of those cases and this is both behavior that is expected given proper configuration but not behavior which is desired, then I think I have a relatively painless solution. This solution would be to add a check for the flag in the build_install_dpdk function and, if it is present, assumes that everything is compiled correctly and exits the function early. This would only involve adding a few lines for the check and then whatever is required to access the flags deeper into dts.

Where this becomes a possible addition to bug 515, is that doing this could break many of the tests which edit dpdk source directly, since the compilation they expect would not occur. As such, this would either need to be its own bug to be fixed after 515, or tacked on to 515.