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

Re: [PATCH 02/10] bisect--helper: reimplement `bisect_next` and `bisect_auto_next` shell functions in C

From
Miriam R. <mirucam@gmail.com>
Date
Mar 6, 2020, 18:19 UTC
Message-ID
<CAN7CjDCb3Bo-reyCZBxMuxX7ecCRLo6XaZHUE5fKGnQav9g_xQ@mail.gmail.com>
In-Reply-To
<xmqqwo88f0do.fsf@gitster-ct.c.googlers.com>
Hi Junio,

El jue., 27 feb. 2020 a las 17:41, Junio C Hamano (<gitster@pobox.com>) escribió:

Show 54 quoted lines
>
> "Miriam R." <mirucam@gmail.com> writes:
>
> >> It would be surprising if this code were correct.  It may be that it
> >> happens to (appear to) work because parents of the commit hasn't
> >> been painted and calling the helper only clears the mark from the
> >> commit (so we won't see repeated "painting down to the root commit")
> >> in which case this might be an extremely expensive looking variant
> >> of
> >>
> >>         commit->object.flags &= ~ALL_REV_FLAGS;
> >>
> >> that only confuses the readers.
> >>
> >> Even then, I think by clearing bits like SEEN from commit, it breaks
> >> the revision traversal machinery---for example, doesn't this mean
> >> that the commit we just processed can be re-visited by
> >> get_revision() without deduping in a history with forks and merges?
> >>
> >> Has this been shown to any of your mentors before sending it to the
> >> list?
> >
> > Adding clear_commit_marks() was a suggestion of a previous review of this patch:
> > https://public-inbox.org/git/nycvar.QRO.7.76.6.2001301619340.46@tvgsbejvaqbjf.bet/
> >
> > And of course, my mentor always reviews my patch series before sending.
>
> OK, I just didn't know how you and your mentors work.  Some leave
> their door open for mentees and help them when asked but otherwise
> act as an ordinary reviewer who somehow prioritises reviewing the
> work by their mentees.  So your mentors may be a better source of
> information why this piece of code, which I still do not know how it
> could be correct, is supposed to work.  Good.
>
> After reading the above URL, I think you may have misread the
> suggestion you were given.  Resetting using clear_commit_marks() may
> be necessary, but you do so when you finished walking so that you
> can do unrelated revision walk later.  For that, you clear the flag
> bits after the while() loop that asks get_revision() to yield
> commits are done, using the initial set of commits that you used to
> start iteration.
>
> That is how bisect.c::check_ancestors() work, that is
>
>  - it initializes a rev_info 'revs' from an array of commit rev[]
>
>  - it lets bisect_common() use the 'revs', which is allowed to
>    smudge the flag bits of commit objects.
>
>  - it uses clear_commit_marks_many() to clear the flags of the
>    commits whose flag bits may have been smudged and their
>    ancestors, recursively.  In order to use as the starting points,
>    the original array of commit rev[] that started the revision
>    traversal is used.
Thank you for your explanation.

To my understanding, it looks like calling reset_revision_walk() after the while() loop should be enough. Am I right or am I missing something?

Previous: Junio C HamanoNext: Junio C Hamano
Message 9 of 20 in “Finish converting git bisect to C part 2”
  1. Miriam RubioFeb 26, 2020
  2. 01/10 bisect--helper: introduce new `write_in_file()` functionMiriam Rubio, Feb 26, 2020
  3. Junio C HamanoFeb 26, 2020
  4. 03/10 bisect--helper: finish porting `bisect_start()` to CMiriam Rubio, Feb 26, 2020
  5. 02/10 bisect--helper: reimplement `bisect_next` and `bisect_auto_next` shell functions in CMiriam Rubio, Feb 26, 2020
  6. Junio C HamanoFeb 26, 2020
  7. Miriam R.Feb 27, 2020
  8. Junio C HamanoFeb 27, 2020
  9. Miriam R.Mar 6, 2020
  10. Junio C HamanoMar 6, 2020
  11. Christian CouderMar 11, 2020
  12. 04/10 bisect--helper: retire `--bisect-clean-state` subcommandMiriam Rubio, Feb 26, 2020
  13. 05/10 bisect--helper: retire `--next-all` subcommandMiriam Rubio, Feb 26, 2020
  14. 06/10 bisect--helper: reimplement `bisect_autostart` shell function in CMiriam Rubio, Feb 26, 2020
  15. Johannes SchindelinFeb 27, 2020
  16. 07/10 bisect--helper: reimplement `bisect_state` & `bisect_head` shell functions in CMiriam Rubio, Feb 26, 2020
  17. Johannes SchindelinFeb 27, 2020
  18. 08/10 bisect--helper: retire `--check-expected-revs` subcommandMiriam Rubio, Feb 26, 2020
  19. 09/10 bisect--helper: retire `--write-terms` subcommandMiriam Rubio, Feb 26, 2020
  20. 10/10 bisect--helper: retire `--bisect-autostart` subcommandMiriam Rubio, Feb 26, 2020

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.