What a warning is worth

A kernel maintainer may face dozens of similar proposed patches on one mailing list. Of the 21,500 mainline changes Google tracked with Sashiko, 0.9 per cent of those that passed cleanly needed a fix within 60 days. The rate was 7.6 per cent for changes merged despite a critical warning. For a maintainer deciding between two patches, the gap offers a clue about which one deserves a second look first.[1]

Changing the order could save maintainers time. If critical warnings tend to land on difficult patches, a machine's first pass could reduce the time spent reading every change from scratch. The patches that draw warnings may also be more complicated to begin with. The gap persists after Google's adjustment for patch size. Size alone cannot capture complexity, and the two fix rates tell us little about how many hours maintainers saved.[1]

Who handles the false alarms?

Sashiko follows its search for possible faults with a separate verification stage. Once a report reaches a mailing list, though, a developer who spots a false alarm may spend time explaining the error. Google says more than 1,500 developers have replied to the reports on the lists. Objections to false alarms helped the team revise its instructions and verification stage.[1]

Google runs the review service. Maintainers who opt in have a concrete measure available to them as well. For each genuine fault found before a patch lands, how many minutes do developers spend clearing false alarms? Tracking that time alongside how long patches sit in the queue would show which work the tool removes and which work comes back to the list. Maintainers who choose whether to receive the reports should be able to see that accounting, too.[1]