Real example 02 · Mission harness · open PR #109
The behavior accepted a two-second burn duration and appeared to honor it, but the helm remained blocked for roughly four seconds because the completion callback could trigger the burn twice.
The upstream problem
BHV_TestFailure intentionally stalls the helm to exercise failure handling. Its configurable burn time looked complete, but onCompleteState() could be invoked more than once and each invocation started a fresh busy wait.
The initial implementation measured elapsed wall time inside onCompleteState() and returned after the requested duration. It had no one-shot guard, so a two-second configuration could burn twice and produce a four-second operational gap.
One-shot guard and monotonic timer in the corrected PR
void BHV_TestFailure::onCompleteState()
{
+ if(m_failure_triggered)
+ return;
+ m_failure_triggered = true;
+
if(m_failure_crash)
assert(0);
else if(m_failure_burn) {
- MBTimer timer;
- timer.start();
- bool done = false;
- while(!done) {
- double elapsed_time = timer.get_float_wall_time();
- if(elapsed_time > m_failure_burn_time)
- done = true;
- }
- timer.stop();
+ const auto start_time = std::chrono::steady_clock::now();
+ while(std::chrono::duration<double>(
+ std::chrono::steady_clock::now() - start_time).count()
+ <= m_failure_burn_time) {}
}
}The test
cd harnesses/testfailure_behavior_harnesses/H01-testfailure_behavior_unit
./zlaunch.sh --case=burn_gap_detected_pass \
--port_base=45500 --log=minimal 10At the incomplete revision
Configured burn: 2 seconds
Observed normalized helm gap: approximately 4 seconds
Expected gap band: 1.6–2.4 seconds
Result: FAIL — burn lasted about twice as long as requestedThe correction
The strengthened mission harness measures the gap between helm iterations and checks duration-specific bands. That distinguishes a two-second burn from a three-second burn instead of merely proving that some delay occurred.
The proposed correction makes failure triggering one-shot and measures the burn with std::chrono::steady_clock. The duration cases then track their configured values across time warps 5, 10, and 20.
All nine harness cases pass on corrected commit 7e6e6129. PR #109 is still open, so this result is corrected and reproducible but not yet merged.
Next example