git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [External] Re: [PATCH v2 2/2] negotiator/skipping: fix some problems in mark_common()

From
Han Xin <hanxin.hx@bytedance.com>
Date
Apr 26, 2023, 11:55 UTC
Message-ID
<CAKgqsWXU0ZCuQtCPA+O+g-36SkF7w2vsgPxS4iQ7gu5V+XCHhQ@mail.gmail.com>
In-Reply-To
<41273b5d-f4f8-2dce-94d1-37a9b56ed1ea@github.com>
On Wed, Apr 26, 2023 at 7:09 PM Derrick Stolee <derrickstolee@github.com> wrote:
Show 12 quoted lines
>
> On 4/26/2023 12:05 AM, Han Xin wrote:
> > Fixed the following problems:
>
> This might be a good time to reference the change from recursive to
> iterative:
>
>   The mark_common() method in negotiator/skipping.c was converted
>   from recursive to iterative in 4654134976f (negotiator/skipping:
>   avoid stack overflow, 2022-10-25), but there is some more work
>   to do:
>
Make sense.
Show 22 quoted lines
> >       while ((c = prio_queue_get(&queue))) {
> >               struct commit_list *p;
> >               if (c->object.flags & COMMON)
> > -                     return;
> > +                     continue;
> >               c->object.flags |= COMMON;
> >               if (!(c->object.flags & POPPED))
> >                       data->non_common_revs--;
> >
> >               if (!c->object.parsed)
> > -                     return;
> > +                     continue;
> >               for (p = c->parents; p; p = p->next) {
> > -                     if (p->item->object.flags & SEEN)
> > +                     if (p->item->object.flags & SEEN || p->item->object.flags & COMMON)
> >                               prio_queue_put(&queue, p->item);
>
> This is the incorrect check for the COMMON bit, because it is
> a positive check (we add the common bit after we pop a commit
> from the queue) _and_ because we could add a commit multiple
> times before it is first popped and that bit is added.
>
Yes, I introduced a silly thing.
Show 19 quoted lines
> Instead, we need
>
>                         if ((p->item->object.flags & SEEN) &&
>                             !(p->item->object.flags & COMMON)) {
>                                 p->item->object.flags |= COMMON;
>                                 prio_queue_put(&queue, p->item);
>                         }
>
> and at the start of the loop we need to add the COMMON bit to
> the starting commit. We also need to remove this bit from the
> main section of the loop:
>
>                 if (c->object.flags & COMMON)
>                         continue;
>                 c->object.flags |= COMMON;
>
> because it does nothing if the COMMON bit is added before
> being added to the queue.
>

Make sense. And with this, we should do return before loop:

                if (seen_commit->object.flags & COMMON)
                        return;
                prio_queue_put(&queue, seen_commit);
                while ((c = prio_queue_get(&queue))) {
Show 7 quoted lines
> I'm very suspicious that this change did not trigger a test
> failure, since the behavior is quite different from the previous
> version. Of course, the recursive-to-iterative change was first
> to change the behavior, so I'm not surprised that it isn't caught
> by tests. What kind of tests can we introduce to harden our
> coverage here?
>

With "p->item->object.flags & COMMON", it takes more meaningless walking, but doesn't seem to introduce any errors. I haven't found any good way to avoid similar problems.

Thanks -Han Xin

Previous: Derrick StoleeNext: Han Xin
Message 8 of 20 in “negotiator/default.c: avoid stack overflow”
  1. negotiator/default.c: avoid stack overflowHan Xin, Apr 24, 2023
  2. Derrick StoleeApr 24, 2023
  3. Han XinApr 25, 2023
  4. Derrick StoleeApr 25, 2023
  5. 0/2 negotiator/default: avoid stack overflowHan Xin, Apr 26, 2023
  6. 2/2 negotiator/skipping: fix some problems in mark_common()Han Xin, Apr 26, 2023
  7. Derrick StoleeApr 26, 2023
  8. Han XinApr 26, 2023
  9. 1/2 negotiator/default: avoid stack overflowHan Xin, Apr 26, 2023
  10. Derrick StoleeApr 26, 2023
  11. Han XinApr 26, 2023
  12. 0/2 negotiator/default: avoid stack overflowHan Xin, Apr 26, 2023
  13. 1/2 negotiator/default: avoid stack overflowHan Xin, Apr 26, 2023
  14. Junio C HamanoApr 26, 2023
  15. Derrick StoleeApr 26, 2023
  16. Junio C HamanoApr 26, 2023
  17. 2/2 negotiator/skipping: fix some problems in mark_common()Han Xin, Apr 26, 2023
  18. Junio C HamanoMay 1, 2023
  19. Derrick StoleeMay 2, 2023
  20. Junio C HamanoMay 2, 2023

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.