Re: non-regression bug in 'git merge' (really, xdiff/xmerge, I think)

From: Elijah Newren <newren@gmail.com>
Date: Mar 6, 2020, 15:48 UTC
Message-ID: <CABPp-BGtez4qjbtFT1hQoREfcJPmk9MzjhY5eEq1QhXT23tFOw@mail.gmail.com>
In-Reply-To: <xmqqy2si9z4x.fsf_-_@gitster-ct.c.googlers.com>

On Mon, Mar 2, 2020 at 2:21 PM Junio C Hamano <gitster@pobox.com> wrote:

Show 63 quoted lines

>
> Thomas Gummerer <t.gummerer@gmail.com> writes:
>
> > diff --git a/Makefile b/Makefile
> > index 9804a0758b..096c6d1fbb 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -609,7 +609,6 @@ SCRIPT_SH += git-merge-one-file.sh
> >  SCRIPT_SH += git-merge-resolve.sh
> >  SCRIPT_SH += git-mergetool.sh
> >  SCRIPT_SH += git-quiltimport.sh
> > -SCRIPT_SH += git-legacy-stash.sh
> >  SCRIPT_SH += git-request-pull.sh
> >  SCRIPT_SH += git-submodule.sh
> >  SCRIPT_SH += git-web--browse.sh
>
> Merging this topic to 'pu', which has dl/merge-autostash topic
> already merged, breaks the build and it is not fault of Thomas's
> topic.  dl/merge-autostash has the following bit:
>
> commit 4b981afaa03c00524f1d4986271a3f4cc119f4cd
> Merge: 06928448f7 de4db86ff2
> Author: Junio C Hamano <gitster@pobox.com>
> Date:   Mon Mar 2 13:23:26 2020 -0800
>
>     Merge branch 'dl/merge-autostash' into pu
>
>     "git merge" learns the "--autostash" option.
>
>     * dl/merge-autostash:
>       ...
>       Makefile: alphabetically sort += lists
>
> diff --git a/Makefile b/Makefile
> index a5961113d8..c0793ac75b 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -604,12 +604,12 @@ unexport CDPATH
>  SCRIPT_SH += git-bisect.sh
>  SCRIPT_SH += git-difftool--helper.sh
>  SCRIPT_SH += git-filter-branch.sh
> +SCRIPT_SH += git-legacy-stash.sh
>  SCRIPT_SH += git-merge-octopus.sh
>  SCRIPT_SH += git-merge-one-file.sh
>  SCRIPT_SH += git-merge-resolve.sh
>  SCRIPT_SH += git-mergetool.sh
>  SCRIPT_SH += git-quiltimport.sh
> -SCRIPT_SH += git-legacy-stash.sh
>  SCRIPT_SH += git-request-pull.sh
>  SCRIPT_SH += git-submodule.sh
>  SCRIPT_SH += git-web--browse.sh
> @@ -617,8 +617,8 @@ SCRIPT_SH += git-web--browse.sh
>
> which makes it appear to the xmerge code that legacy-status that
> used to exist in between quiltimport and request-pull has already
> been removed.
>
> As Thomas's above patch exactly wants to do that (i.e. we want to
> remove legacy-stash between quiltimport and request-pull), the
> xmerge code seems to think incorrectly that the requested change has
> already been applied.  And as a result of that, there remains
> legacy-stash between fliter-branch and merge-octopus, breaking the
> build by having one unbuildable build target X-<.

Thanks for the heads up; I agree that it looks like an xdiff thing, and in particular it reminds me of https://lore.kernel.org/git/20190816184051.GB13894@sigill.intra.peff.net/ and https://lore.kernel.org/git/CABPp-BHvJHpSJT7sdFwfNcPn_sOXwJi3=o14qjZS3M8Rzcxe2A@mail.gmail.com/

I'm collecting these cases because I'd like to look into it at some point, though for now I'm more focused on merge-ort (my merge-recursive replacement).