Re: [GIT PULL] MM updates for 7.2-rc1

[Konstantin Ryabitsev]

On Thu, Jun 18, 2026 at 04:37:52PM -0400, Konstantin Ryabitsev wrote:
> On Thu, Jun 18, 2026 at 09:02:06AM -0700, Linus Torvalds wrote:
> > This is very wrong. That merge (2f9f5887b427) is merging Jens'
> > io_uring epoll cleanup. Not Andrew's MM tree, which I haven't gotten
> > to yet.
> > 
> > I wonder why pr-tracker-bot got this wrong.
> > 
> > [ Goes off and looks ]
> > 
> > Oh. Andrew's pull request is confused. It says
> > 
> >    The following changes since commit e771677...
> >      ...
> >    for you to fetch changes up to e771677....
> > 
> > and note how the "since" commit and "up to" commits are the same (and
> > is my iommufd merge). So then pr-tracker bot thinks it's merged when
> > the "since" commit was merged.
> 
> Let me at least handle this situation -- it's clearly an oversight.

The bot should now properly recognize when the start and end of the range are
the same sha and ignore the message (it seemed a safer logic than trying to do
anything else).

> > That doesn't explain why pr-tracker-bot thinks it got merged in what
> > was me merging Jens' tree, but whatever.

It's "bad assumptions snowballing." The merge commit in the notification isn't
taken from the pull request at all -- it's computed. Once the bot believes a
commit is merged, it runs:

    git rev-list <commit>.. --ancestry-path

and then takes the oldest commit in that range. With <commit> being e771677,
that is simply the next merge you made after the iommufd pull:

    $ git rev-list e771677c937d.. --ancestry-path | tail -1
    2f9f5887b42711595e768b9dc0582dccfdf60c3b

...which is your merge of Jens' io_uring epoll tree. So the bot wasn't
associating Andrew's MM tree with Jens' work; it just tried to do its best
with bad data, with predictably bad outcomes. :)

-K