git rerere and diff3

> i have noticed that merge.conflictstyle has an impact on the rerere
> resolution. looking briefly at the source code, it seems that git
> tries to discard the common ancestor diff3 bits, but what I am seeing
> is that if i do the following then it fails:
>
> 1. from a clean rr-cache state, with merge.conflictsytle=diff3, git
> merge <branch with conflict>, resolve the conflicts, then commit
> 2. undo the previous merge, remove merge.conflictstyle=diff3 (disable
> diff3) and merge the *same* branch, then rerere won't fix the
> conflicts

>
> Nicolas Dechesne <nicolas.dechesne@linaro.org> writes:
>
> > i have noticed that merge.conflictstyle has an impact on the rerere
> > resolution. looking briefly at the source code, it seems that git
> > tries to discard the common ancestor diff3 bits, but what I am seeing
> > is that if i do the following then it fails:
> >
> > 1. from a clean rr-cache state, with merge.conflictsytle=diff3, git
> > merge <branch with conflict>, resolve the conflicts, then commit
> > 2. undo the previous merge, remove merge.conflictstyle=diff3 (disable
> > diff3) and merge the *same* branch, then rerere won't fix the
> > conflicts
>
> It is possible that the conflict left when making the same merge are
> actually different when using these two conflict styles.  IOW, if
> the merge produces
>
>         <<<
>         side A
>         |||
>         common
>         ===
>         side B
>         >>>
>
> when diff3 style is chosen, but if the same merge results in
>
>         <<<
>         side A'
>         ===
>         side B'
>         >>>
>
> where side A' is not identical to side A (or B' and B are not
> identical), then we will fail to find the previously recorded
> resolution.