On 03/06/2025 1:42 pm, Anthony PERARD wrote:
> diff --git a/automation/scripts/qubes-x86-64.sh
> b/automation/scripts/qubes-x86-64.sh
> index 2750d24eba..046137a4a6 100755
> --- a/automation/scripts/qubes-x86-64.sh
> +++ b/automation/scripts/qubes-x86-64.sh
> @@ -135,10 +135,11 @@ done
> ### tests: tools-tests-pv, tools-tests-pvh
> "tools-tests-pv"|"tools-tests-pvh")
> retrieve_xml=1
> - passed="test passed"
> + passed="run-tools-test over"
> domU_check=""
> dom0_check="
> -/root/run-tools-tests /usr/lib/xen/tests /tmp/tests-junit.xml && echo
> \"${passed}\"
> +/root/run-tools-tests /usr/lib/xen/tests /tmp/tests-junit.xml ||:
> +echo \"${passed}\"
> nc -l -p 8080 < /tmp/tests-junit.xml >/dev/null &
> "
> if [ "${test_variant}" = "tools-tests-pvh" ]; then
I noticed this too while hacking on XTF support. Also of note, I'm not
sure we want to be saying done to the expect script before sending the
results.
The underlying problem is that ${passed} isn't really a pass message;
it's a done+pass message, and there is no way to say "failed" to the
expect script, leaving timeout as the only option of signalling a failure.
I don't think we want to be ignoring errors from run-tools-tests
specifically, but arranging this is hard. You always need something at
an outer level judging whether there was a good test result.
For XTF, the way I did this was to declare that anything didn't match
"^(Success|Error|Crash|Fail).*" on the final line of the console log to
be deemed to be Crash (because it's usually an uncaught exception).
For this, the best I can think of is to have a fixed outer script which
runs "./test/script && echo $success || echo $failure". The internal
set -e will cause most unexpected conditions to be an error.
~Andrew
