From: Alejandro Colomar <alx@kernel.org>
Date: Sat, 03 Oct 2026 06:03:18 GMT
Subject: Re: git-bisect(1) next after finding the commit
Message-ID: <asCaIXYuZFdlFU0J@debian>
In-Reply-To: <xmqqece7vimq.fsf@gitster.g>

Hi Junio,

> Date: 2026-10-02 15:32:13-0700
> From: Junio C Hamano <gitster@pobox.com>
>
> Jeff King <peff@peff.net> writes:
> 
> > On Fri, Oct 02, 2026 at 11:49:01PM +0200, Alejandro Colomar wrote:
> >
> >> Is there a plumbing command for retrieving the bisected commit after
> >> a git-bisect(1) session has successfully found it (and of course before
> >> resetting the session)?
> >> 
> >> I expected `git bisect next` would bring me to it, and then I'd rev-list
> >> HEAD -1, but it doesn't bring me to it.
> >
> > I'm not 100% sure, but I think "git show bisect/bad" should work.
> >
> > As the bisection progresses, we advance a single refs/bisect/bad from
> > the bottom of the range (the top is multiple refs/bisect/good-* refs).
> > So at the end, it should point to the blamed commit.
> 
> The only code that gives "is the first .* commit" message is this bit
> in bisect.c:
> 
> 	if (oideq(bisect_rev, current_bad_oid)) {
> 		res = error_if_skipped_commits(tried, current_bad_oid);
> 		if (res)
> 			goto cleanup;
> 		printf("%s is the first '%s' commit\n", oid_to_hex(bisect_rev),
> 			term_bad);
> 
> it is fed bisect_rev only when it is the same as current_bad_oid,
> which was read from "refs/bisect/bad".  So I think you are right.

Thanks for confirming!  It certainly works for me (I've tested it).

> [Footnote]
> 
> There is a last-step optimization that made me double check the
> code, but the optimization is about not bothering to move HEAD and
> not moving refs/bisect/bad pointer.
> 
> We have a three-topic branch X, X is at the tip, X~1 at the middle
> and X-2 at the bottom.  X~3 is on the mainline.
> 
> ----- X~3 ---- X~2 ---- X~1 ---- X
> 
> We found X~2 to be good. X is bad.
> 
>   $ git bisect start X X~2
> 
> We are asked to test X~1.
> 
>  * It turns out to be bad.  We move bisect/bad to X~1.  And then we
>    report that X~1 is the first bad commit.
> 
>  * Or it turns out to be good.  Then we report that X is the first
>    bad commit.  But we do not move HEAD and checkout X.

Indeed, this optimization is what surprised me.  I expected it would
move the HEAD as with every other git-bisect(1) step, but the last one
didn't move it.  I can live with it, FWIW, since bisect/bad contains all
the information I need (and anyway, I will move the HEAD myself, so I
don't need the HEAD in any specific position).  It was just surprising.

:)


Have a lovely day!
Alex

-- 
<https://www.alejandro-colomar.es>