Bisecting through the middle of a big merge?

Bisecting through the middle of a big merge?

[walt]

My problem is that I lack a good mental model of what git does
when it commits a big merge, and so I get lost when I try to
bisect back through a merge.  (I'm a tester, not a developer.)

This specific recent example is causing me problems:

Merge: b3c9dd1 85a0f7b
Author: Linus Torvalds
Date:   Sun Jan 15 12:48:41 2012 -0800
    Merge branch 'for-3.3/drivers' of git://git.kernel.dk/linux-block

There are many individual commits from Tejun Heo et al included
in that one big commit from Linus.  Unfortunately for me, some of
those commits cause other problems that I'm not trying to bisect;
other problems that evidently get fixed by other commits in the
same big merge.

So I do 'git bisect skip' six or eight times until the 'false' bug
goes away, and that leaves me at the end of the bisect without finding
the individual commit that's causing my 'real' bug.

How do you experts handle this kind of problem?

Thanks for any clues.

Re: Bisecting through the middle of a big merge?

[Andreas Schwab]

walt <w41ter@gmail.com> writes:

> There are many individual commits from Tejun Heo et al included
> in that one big commit from Linus.  Unfortunately for me, some of
> those commits cause other problems that I'm not trying to bisect;
> other problems that evidently get fixed by other commits in the
> same big merge.
>
> So I do 'git bisect skip' six or eight times until the 'false' bug
> goes away, and that leaves me at the end of the bisect without finding
> the individual commit that's causing my 'real' bug.
>
> How do you experts handle this kind of problem?

If you can identify the commit that fixes the unrelated problem you can
try to cherry-pick it during the bisect.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Bisecting through the middle of a big merge?

[walt]

On 01/28/2012 12:55 AM, Andreas Schwab wrote:
> walt <w41ter@gmail.com> writes:
> 
>> There are many individual commits from Tejun Heo et al included
>> in that one big commit from Linus.  Unfortunately for me, some of
>> those commits cause other problems that I'm not trying to bisect;
>> other problems that evidently get fixed by other commits in the
>> same big merge.
>>
>> So I do 'git bisect skip' six or eight times until the 'false' bug
>> goes away, and that leaves me at the end of the bisect without finding
>> the individual commit that's causing my 'real' bug.
>>
>> How do you experts handle this kind of problem?
> 
> If you can identify the commit that fixes the unrelated problem you can
> try to cherry-pick it during the bisect.

Thanks Andreas.  With an eye to doing that, is there an easy way to
get a list of all the commits included in Linus's merge?  (I mean a
more accurate list than Linus casually mentions in his commit message.)

Trying to build that mental model I mentioned:  All the commits from
Tejun Heo are dated mid-December but Linus didn't commit them until
mid-January.  When I'm bisecting through that merge, git builds the
kernels with names like vmlinuz-3.2.0-rc5-foo, i.e. names a month
older than Linus's current kernel version.  Where does git get those
older names during the bisect?  And does my working tree exclude all
of Linus's commits made later than 3.2.0-rc5-foo?

Many thanks.

Re: Bisecting through the middle of a big merge?

[Andreas Schwab]

walt <w41ter@gmail.com> writes:

> With an eye to doing that, is there an easy way to get a list of all
> the commits included in Linus's merge?

$ git log merge-commit^..merge-commit

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Bisecting through the middle of a big merge?

[walt]

On 01/28/2012 11:47 AM, Andreas Schwab wrote:
> walt <w41ter@gmail.com> writes:
> 
>> With an eye to doing that, is there an easy way to get a list of all
>> the commits included in Linus's merge?
> 
> $ git log merge-commit^..merge-commit

Amazing. And obvious now that you've already told me the answer.
I'm not thinking like a developer yet, but I'm working on it :)