TestExec is one of our more troublesome tests, but I have not seen this particular failure mode before. It looks like it happened only once:
which will make this hard to debug.
Should it be disabled or replaced with a more reliable subset?
We know what the issue is (not with this failure which we’ve never seen, but generally) . We are waiting for some feedback from the kernel/runtime teams on how to dress it.
If it is holding up Swift *compiler *pull requests then yes, but on the LLDB bots themselves we really need the signal even if there is some noise. Unfortunately bugs like this one that are on the interface between kernel and debug server are notoriously hard to reproduce and debug, but we are working on it. We do have a horrible mechanism in the LLDB test suite that allows us to mark specific tests as flaky which will re-run them if they failed, but there's a real risk that enabling this will drown useful signal, so we're cautious about it.