I just ran into a second instance of of signal han...
# proj-agent-canvass
straight-dog-12443
08/01/2026, 7:12 PMI just ran into a second instance of of signal handling gone wild. In this case, I think I killed the multiplexer that had the launcher in it. That issued a SIGHUP to the launcher but the shutdown handler didn't catch HUPs
I could sweep for that class of bug and probably more offer PRs if that would be wanted.
> Previous fix: github.com/OpenHands/OpenHands/pull/16152
> Current issue: github.com/OpenHands/OpenHands/issues/16238
> Draft fix: github.com/OpenHands/OpenHands/pull/16239
gratitude thank you 2
✅ 1
c
calm-artist-7206
08/01/2026, 7:17 PMThank you, of course!
🙏 1
s
straight-dog-12443
08/02/2026, 11:15 PMI have the next 3 small patches to ship but I'm gonna wait until 16239 passes review bc it's more of the same
✅ 1
c
calm-artist-7206
08/03/2026, 11:43 AMMerged, thank you! My agent also make a related issue, because it noticed the code around is a bit fragile
calm-artist-7206
08/03/2026, 11:43 AMI'd love to know what you think 🙏
s
straight-dog-12443
08/04/2026, 1:45 AMYes, I noticed as well. I had a branch that was largely the same work as the fix for issue 16257. I stopped when it started to get involved because too big of PR from a newcomer is a headache.
I'm examining the differences now to see if there's anything worth raising.
straight-dog-12443
08/04/2026, 10:58 AMhad a look. the issue's right, and it shows up in the listener counts: importing dev-static on main gives SIGINT 2 / SIGTERM 2 / SIGHUP 1, its own pair plus the three it inherits. the duplicate-handler race is real.
swept the rest of the class while i was in there. five files spawn detached, the four launchers plus the shared helper. dev-with-automation got SIGHUP in 16239, dev-static is covered by 16262, and dev-safe and dev-extra-backend never had it -- those two are up as 16279 and 16280. ingress spawns nothing so nothing to orphan there, and bin/electron inherit from main().
so once those land the class is closed.
c
calm-artist-7206
08/04/2026, 11:29 AMtoo big of PR from a newcomer is a headacheI hear that, though I wonder. On one side, we are, obviously, trying to use agents for as much as possible for good old code review. On the other side, I started to feel many months ago that humans can’t and won’t do most of code review anymore, but that’s because code review is changing into code design review. If the design is right, maybe the size doesn’t matter all that much?
calm-artist-7206
08/04/2026, 11:30 AMSorry, just wanted to note because your sentence reminded me - this is still in flux TBH
s
straight-dog-12443
08/04/2026, 11:47 AMI definitely agree with that thinking but there's still gotta be some kind of introductory vetting.
You've gotta have some way of concluding "I've seen this person exercise good judgment, risk is less".
Otherwise, you're going to get people who put in major rewrites. Even if it is good content, if you don't know the submitter then it has to get set aside for a more careful investigation. Before long it becomes a major feature against an old version.
Either way, that was definitely my reasoning. Submit some small stuff, establish competence before considering any bigger conversations.
AI makes it too easy to write messages that convey understanding without actually having that understanding.
I think usually that only lasts for a few messages,
nod 1