Re: [PATCH v2 0/2] Address recovery failures with directory/file conflicts

From: Elijah Newren <newren@gmail.com>
Date: Jul 13, 2018, 16:45 UTC
Message-ID: <CABPp-BGtrsAzvT7=95vKUXveQ=3X6FkAc5VgEYBNRXSzsqY0DA@mail.gmail.com>
In-Reply-To: <20180713163331.22446-1-newren@gmail.com>

On Fri, Jul 13, 2018 at 9:33 AM, Elijah Newren <newren@gmail.com> wrote:

Show 13 quoted lines

> This patch series fixes several "recovery" commands that outright fail
> or do not fully recover when directory-file conflicts are present.
> This includes:
>    * git read-tree --reset HEAD
>    * git am --skip
>    * git am --abort
>    * git merge --abort (or git reset --merge)
>    * git reset --hard
>
> Changes since v1 (full range-diff below):
>   - Make use of test_write_lines, as suggested by Eric.
>   - Provide a little more explanation in one of the commit messages, as
>     suggested by Junio.

I forgot to mention: there is also an expected failing merge --abort testcase in en/t7405-recursive-submodule-conflicts (currently in pu), and the fact that this series does not fix that testcase is known. For the curious...

This read-index-unmerged series is part of the solution for the problem highlighted in t7405; in fact, the similarity between file/directory and submodule/directory conflicts is such that this patch series does also happen to fix recovery problems for many submodule/directory conflicts as well. However, additional work is needed when the same path is found under both the submodule and the directory, and the testcase in en/t7405-recursive-submodule-conflicts is precisely one of those cases that needs additional work.

(This is just a long winded way of saying that this series and en/t7405-recursive-submodule-conflicts don't conflict or duplicate each other.)