[Libreoffice-qa] Correct use of bibisect to identify a particular commit

Xisco Fauli xiscofauli at libreoffice.org
Wed Jun 20 13:34:52 UTC 2018

Hello Alex,

El 20/06/18 a les 13:47, Alex Kempshall ha escrit:
> Hi Xisco
> Still don't understand what the  technique is to get from a bibisect
> log to a bisect log.
> If you remember I identified, by using bibisect, when a regression
> occurred. Then you went on to identify the actual commit see
> https://bugs.documentfoundation.org/show_bug.cgi?id=118220#c3
So, taking this as example.

I used bibisect-linux64-6.2 [1] as mentioned in the comment. This is a
'max' repository which means there's one build per commit, thus, if I
bisect that bug, I get:

670b6459ca52221d8c2a314dd55133c8dada35d4 is the first bad commit
commit 670b6459ca52221d8c2a314dd55133c8dada35d4
Author: Jenkins Build User <tdf at pollux.tdf>
Date:   Tue Jun 5 00:24:31 2018 +0200

    source sha:b69457c1a5cb31c8b0fb3d047e2826bc3d61fea4
    source sha:b69457c1a5cb31c8b0fb3d047e2826bc3d61fea4

which points directly to the commit in source sha.

OTOH, as you used a gdbutil daily repo to bibisect it, you got:

# good: [072a924b4800a972e66f4f2c043d91b79749a19c] 2018-05-29:
git bisect good 072a924b4800a972e66f4f2c043d91b79749a19c
# first bad commit: [26517de6f6c3dc7472175ca36e5a959d7b9bbf21]
2018-05-30: source-hash-7725eff80a7bc3f3e422e56e1cd7bfa531f19a70

which returns you a range of commits [2]. At this point you could check
manually with commit introduced the problem, but it's much easier to use
the 'max' repositories instead.


[1] https://gerrit.libreoffice.org/#/admin/projects/bibisect-linux-64-6.2

Xisco Faulí
Libreoffice QA Team
IRC: x1sc0

More information about the Libreoffice-qa mailing list