Re: [kernelci-members] Working with the KernelCI project

["Nikolai Kondrashov" <Nikolai.Kondrashov@redhat.com>]

On 10/1/20 11:43 AM, Dmitry Vyukov wrote:
 > On Thu, Oct 1, 2020 at 10:30 AM Nikolai Kondrashov
 > <Nikolai.Kondrashov@redhat.com> wrote:
 >>
 >> On 9/30/20 10:07 AM, Dmitry Vyukov wrote:
 >>   > On Tue, Sep 29, 2020 at 7:13 PM Nikolai Kondrashov
 >>   > <Nikolai.Kondrashov@redhat.com> wrote:
 >>   >> Regarding bisection results I was hoping we could represent that data
 >>   >> implicitly. I.e. by having information about which 'revision' comes after
 >>   >> which, and executing a query to find a pair of adjacent revisions one of which
 >>   >> has the bug and another doesn't. In theory that should be possible, but in
 >>   >> practice... well, you know :) If that fails, we can attach that data
 >>   >> explicitly somewhere. Or we can simply start with that to have it faster.
 >>   >
 >>   > Humm... interesting. I guess it can work. We could mark the issue as
 >>   > happening on the cause commit and as not happening on the previous
 >>   > commit.
 >>   > However, one problem is that we don't keep info about the previous
 >>   > commit as of now, we only store the root cause commit in the database.
 >>
 >> Something has to maintain the information on history connectivity, and so far
 >> the plan was to let the submitting system do it. E.g. refer to a clone of the
 >> repo they're testing. There's definitely more research and design to be done.
 >>
 >> However, you should've been testing the previous commit to be able to find the
 >> root cause commit? Shouldn't you?
 >
 > We do. But that happens in a very different part of the system and
 > that info is not kept in a structured form because it's not useful for
 > anything in our system.
 >
 > Well, it actually turned out that it's not necessary to test all of
 > the parents of a commit during bisection to point at a particular
 > commit in some interesting git tree topologies.

Ah, I see I'll need to ask you more questions about bisections once we get to
that :)

 >>   > Another is that in git it's possible to have several preceding
 >>   > commits.
 >>
 >> Right, I didn't think through that yet. We might need a separate history link
 >> object, then. Do you know if a commit can have more than two parents?
 >
 > Octopus merge commits can have lots of parents.

Right, I forgot about those.

Thank you!
Nick