On Fri, Jan 31, 2025 at 03:49:22PM -0800, Junio C Hamano wrote:

To help the idea of catching before things hit next, it probably
would make the most sense to test the tip of the 'jch' branch, which
is somewhere between the 'master' and the 'seen' branches and
contains a bit more topics than the 'next' branch does. The branch
is usually what I use for my work every day, so even though it may
have acquired new leaks and UBs that would not cause troubles in
practice, it should functionally be a lot more stable and usable
than the tip of 'seen'.

I've switched to basing my branch on 'jch', which did turn up a few new
hits (lots of uninteresting ones, but a few that may be important; I've
responded in the relevant threads).

What I'm most worried about is that it does have leaks or UB. My daily
workflow is usually to build my personal branch (previously 'next' now
'jch', plus my own topics) and stop everything to investigate and fix if
CI fails. If I start getting CI failures for other peoples' topics, then
I'll have even less time to do my own backlogged ones. ;)

Right now you are shouldering a lot of that during your integration
runs. And if it were even putting some of your load on me, that might be
a good tradeoff. But I have a feeling it is just putting the same on
both of us as we see the same CI failures and poke at them
independently. I dunno. I'll try it for a while and see how it goes.

-Peff