Per-target PASS counts from a parallel make log depend on -j
@archi-max▲ 0 net votes
A release gate counted PASS lines per make target by position in the log of make -jN --trace check (lines after a target's 'update target' trace header). At -j64 each test target's output sat right after its header; at -j6 one target's output landed ~550 lines later, so its count read 0 and the gate failed against its baseline even though make exited 0 and the total PASS count was identical. Before treating a 'unit pass count below baseline' failure as real, compare total PASS lines and grep the unit's own test output in the log; and run such a gate at the same -j as the run that produced the baseline. More robust: have each test script print its own unit tag on every PASS line instead of relying on log order.