Error message with 'bisect'

[Nicolas Pitre]

On Fri, 9 Jul 2010, Mathieu Poirier wrote:

> Andy,
> 
> I started investigating the "bisect" command as you suggested on Friday.
> After having a few successful runs with the tool, I isolated my problem
> to a check in between rc3 and rc4... But there is 425 of them.  
> 
> So what better tool than "bisect" to sort through all this !  I checkout
> a fresh upstream tree and set out to bisect but got an error message:
> 
> mpoirier at black:~/canonical/linus$ git bisect start
> mpoirier at black:~/canonical/linus$ git bisect good v2.6.35-rc4
> mpoirier at black:~/canonical/linus$ git bisect bad v2.6.35-rc3
> Some good revs are not ancestor of the bad rev.
> git bisect cannot work properly in this case.
> Maybe you mistake good and bad revs?
> mpoirier at black:~/canonical/linus$
> 
> 
> Tim tried on his own tree and got the same issue.  Would you know what
> is going on ?

The 'git bisect' command is meant to let you find when a bug was 
_introduced_, and not when it was _fixed_.  What the message above is 
telling you is that the good revision you provided (v2.6.35-rc4) is not 
an ancestor of the bad revision (v2.6.35-rc3).  Obviously, v2.6.35-rc4 
cannot have happened before v2.6.35-rc3.

If what you really want is to find out when a bug was fixed then you can 
use 'git bisect' as well, but then you need to invert all the logic, 
i.e. mark fixed commits as "bad" and non fixed ones as "good".  In any 
case, the "good" commits are expected to exist before the "bad" ones.


Nicolas