threads / patch / 24685

patchAvoid rename/add conflict when contents are identical

Subject: [PATCH] Avoid rename/add conflict when contents are identical

## tl;dr

8 messages between Aug 9, 2010 and Sep 3, 2010. Diffs are folded; open one to read it.

replies: 7people: 3as markdown or json

Schalk, Ken· Aug 9, 2010, 21:00 UTC · lore

Skip the entire rename/add conflict case if the file added on the other branch has the same contents as the file being renamed. This avoids giving the user an extra copy of the same file and presenting a conflict that is confusing and pointless.

Here's a simple sequence that generates this kind of conflict:
  git init
  echo content > fileA
  git add fileA
  git commit -m Initial
  git checkout -b abranch
  mv fileA fileB
  git add fileB
  rm fileA
  ln -s fileB fileA
  git add fileA
  git commit -m Linked
  git checkout master
  git mv fileA fileB
  git add fileB
  git commit -m Moved
  git merge --no-commit abranch
Signed-off-by: Ken Schalk <ken.schalk@intel.com>
---
 merge-recursive.c |   12 ++++++++++++
 1 files changed, 12 insertions(+), 0 deletions(-)
Show changes to merge-recursive.c +12 −1
diff --git a/merge-recursive.c b/merge-recursive.c
index fb6aa4a..57c7a85 100644
--- a/merge-recursive.c
+++ b/merge-recursive.c
@@ -955,6 +955,18 @@ static int process_renames(struct merge_options *o,
                                                        ren1->pair->two : NULL,
                                                        branch1 == o->branch1 ?
                                                        NULL : ren1->pair->two, 1);
+                       } else if ((dst_other.mode == ren1->pair->two->mode) &&
+                                  sha_eq(dst_other.sha1, ren1->pair->two->sha1)) {
+                               /* Added file on the other side
+                                  identical to the file being
+                                  renamed: clean merge */
+                               update_file(o, 1, ren1->pair->two->sha1, ren1->pair->two->mode, ren1_dst);
+                               if (!o->call_depth)
+                                       update_stages(ren1_dst, NULL,
+                                                       branch1 == o->branch1 ?
+                                                       ren1->pair->two : NULL,
+                                                       branch1 == o->branch1 ?
+                                                       NULL : ren1->pair->two, 1);
                        } else if (!sha_eq(dst_other.sha1, null_sha1)) {
                                const char *new_path;
                                clean_merge = 0;
--
1.7.0
Ævar Arnfjörð Bjarmason· Aug 9, 2010, 21:24 UTC · re: Schalk, Ken · lore

Re: [PATCH] Avoid rename/add conflict when contents are identical

On Mon, Aug 9, 2010 at 21:00, Schalk, Ken <ken.schalk@intel.com> wrote:
Show 18 quoted lines
> Here's a simple sequence that generates this kind of conflict:
>
>  git init
>  echo content > fileA
>  git add fileA
>  git commit -m Initial
>  git checkout -b abranch
>  mv fileA fileB
>  git add fileB
>  rm fileA
>  ln -s fileB fileA
>  git add fileA
>  git commit -m Linked
>  git checkout master
>  git mv fileA fileB
>  git add fileB
>  git commit -m Moved
>  git merge --no-commit abranch

I don't know merge well enough to review the code in this patch, but the patch would be much better if you turned this sequence of commands (slightly altered, see t/README) into a now-passing test as part of the patch.

t/t3030-merge-recursive.sh would be the right place to add it.
Junio C Hamano· Aug 13, 2010, 01:18 UTC · re: Schalk, Ken · lore

Re: [PATCH] Avoid rename/add conflict when contents are identical

"Schalk, Ken" <ken.schalk@intel.com> writes:
Show 23 quoted lines
> diff --git a/merge-recursive.c b/merge-recursive.c
> index fb6aa4a..57c7a85 100644
> --- a/merge-recursive.c
> +++ b/merge-recursive.c
> @@ -955,6 +955,18 @@ static int process_renames(struct merge_options *o,
>                                                         ren1->pair->two : NULL,
>                                                         branch1 == o->branch1 ?
>                                                         NULL : ren1->pair->two, 1);
> +                       } else if ((dst_other.mode == ren1->pair->two->mode) &&
> +                                  sha_eq(dst_other.sha1, ren1->pair->two->sha1)) {
> +                               /* Added file on the other side
> +                                  identical to the file being
> +                                  renamed: clean merge */
> +                               update_file(o, 1, ren1->pair->two->sha1, ren1->pair->two->mode, ren1_dst);
> +                               if (!o->call_depth)
> +                                       update_stages(ren1_dst, NULL,
> +                                                       branch1 == o->branch1 ?
> +                                                       ren1->pair->two : NULL,
> +                                                       branch1 == o->branch1 ?
> +                                                       NULL : ren1->pair->two, 1);
>                         } else if (!sha_eq(dst_other.sha1, null_sha1)) {
>                                 const char *new_path;
>                                 clean_merge = 0;

The logic seems to be clear enough. Could you write a test script instead of description in the commit log message, so that future changes to the codebase won't break this improvement?

Thanks.
Schalk, Ken· Aug 27, 2010, 22:14 UTC · re: Junio C Hamano · lore

RE: [PATCH] Avoid rename/add conflict when contents are identical

>The logic seems to be clear enough.  Could you write a test script instead
>of description in the commit log message, so that future changes to the
>codebase won't break this improvement?
Here's a revised patch which fixes a problem with the earlier one and includes a test (where Ævar Bjarmason suggested it):

Skip the entire rename/add conflict case if the file added on the other branch has the same contents as the file being renamed. This avoids giving the user an extra copy of the same file and presenting a conflict that is confusing and pointless.

A simple test of this case has been added in t/t3030-merge-recursive.sh.

Signed-off-by: Ken Schalk <ken.schalk@intel.com>
---
 merge-recursive.c          |    6 ++++++
 t/t3030-merge-recursive.sh |   40 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+), 0 deletions(-)
Show changes to 2 files +46 −1

merge-recursive.c, t/t3030-merge-recursive.sh

diff --git a/merge-recursive.c b/merge-recursive.c
index fb6aa4a..a2fba84 100644
--- a/merge-recursive.c
+++ b/merge-recursive.c
@@ -955,6 +955,12 @@ static int process_renames(struct merge_options *o,
                                                        ren1->pair->two : NULL,
                                                        branch1 == o->branch1 ?
                                                        NULL : ren1->pair->two, 1);
+                       } else if ((dst_other.mode == ren1->pair->two->mode) &&
+                                  sha_eq(dst_other.sha1, ren1->pair->two->sha1)) {
+                               /* Added file on the other side
+                                  identical to the file being
+                                  renamed: clean merge */
+                               update_file(o, 1, ren1->pair->two->sha1, ren1->pair->two->mode, ren1_dst);
                        } else if (!sha_eq(dst_other.sha1, null_sha1)) {
                                const char *new_path;
                                clean_merge = 0;
diff --git a/t/t3030-merge-recursive.sh b/t/t3030-merge-recursive.sh
index d541544..6436582 100755
--- a/t/t3030-merge-recursive.sh
+++ b/t/t3030-merge-recursive.sh
@@ -23,6 +23,8 @@ test_expect_success 'setup 1' '
        git branch df-3 &&
        git branch remove &&
        git branch submod &&
+       git branch mv &&
+       git branch mv-ln &&

        echo hello >>a &&
        cp a d/e &&
@@ -248,6 +250,25 @@ test_expect_success 'setup 7' '
        git commit -m "make d/ a submodule"
 '

+test_expect_success 'setup 8' '
+
+       git checkout mv-ln &&
+       git mv a e &&
+       ln -s e a &&
+       git add a e &&
+       test_tick &&
+       git commit -m "rename a->e, symlink a->e"
+'
+
+test_expect_success 'setup 9' '
+
+       git checkout mv &&
+       git mv a e &&
+       git add e &&
+       test_tick &&
+       git commit -m "rename a->e"
+'
+
 test_expect_success 'merge-recursive simple' '

        rm -fr [abcd] &&
@@ -580,4 +601,23 @@ test_expect_failure 'merge-recursive simple w/submodule result' '
        test_cmp expected actual
 '

+test_expect_success 'merge-recursive rename vs. rename/symlink' '
+
+       git checkout -f mv &&
+       git merge mv-ln &&
+       ( git ls-tree -r HEAD ; git ls-files -s ) >actual &&
+       (
+               echo "100644 blob $o0   b"
+               echo "100644 blob $o0   c"
+               echo "100644 blob $o0   d/e"
+               echo "100644 blob $o0   e"
+               echo "100644 $o0 0      b"
+               echo "100644 $o0 0      c"
+               echo "100644 $o0 0      d/e"
+               echo "100644 $o0 0      e"
+       ) >expected &&
+       test_cmp expected actual
+'
+
+
 test_done
--
1.7.0
Ævar Arnfjörð Bjarmason· Aug 28, 2010, 10:12 UTC · re: Schalk, Ken · lore

Re: [PATCH] Avoid rename/add conflict when contents are identical

On Fri, Aug 27, 2010 at 22:14, Schalk, Ken <ken.schalk@intel.com> wrote:
> +       ln -s e a &&

Due to this this (and maybe all the tests) need to depend on the SYMLINKS prereq.

Schalk, Ken· Aug 31, 2010, 21:05 UTC · re: Ævar Arnfjörð Bjarmason · lore

RE: [PATCH] Avoid rename/add conflict when contents are identical

>On Fri, Aug 27, 2010 at 22:14, Schalk, Ken <ken.schalk@intel.com> wrote:
>> +       ln -s e a &&
>Due to this this (and maybe all the tests) need to depend on the
>SYMLINKS prereq.
I used a symlink here to try and duplicate the original problem case that a git user at Intel brought to me.  I admit I didn't consider the problem of testing on platforms without symlink support.
I'm having a little difficulty generating a rename/add conflict inside this test without using a symlink, which seems odd to me as the symlink doesn't seem to be necessary in an experiment I did by hand.  I'll either re-submit with this portion of the test conditional on symlink support, or with an alternate test that doesn't use symlinks (if I can get that to work in the test).
--Ken
Schalk, Ken· Sep 1, 2010, 20:15 UTC · re: Ævar Arnfjörð Bjarmason · lore

RE: [PATCH] Avoid rename/add conflict when contents are identical

>Due to this this (and maybe all the tests) need to depend on the
>SYMLINKS prereq.
Here's a third attempt with no use of symlinks in the test:

Skip the entire rename/add conflict case if the file added on the other branch has the same contents as the file being renamed. This avoids giving the user an extra copy of the same file and presenting a conflict that is confusing and pointless.

A simple test of this case has been added in t/t3030-merge-recursive.sh.

Signed-off-by: Ken Schalk <ken.schalk@intel.com>
---
 merge-recursive.c          |    6 ++++++
 t/t3030-merge-recursive.sh |   39 +++++++++++++++++++++++++++++++++++++++
 2 files changed, 45 insertions(+), 0 deletions(-)
Show changes to 2 files +45 −1

merge-recursive.c, t/t3030-merge-recursive.sh

diff --git a/merge-recursive.c b/merge-recursive.c
index fb6aa4a..a2fba84 100644
--- a/merge-recursive.c
+++ b/merge-recursive.c
@@ -955,6 +955,12 @@ static int process_renames(struct merge_options *o,
                                                        ren1->pair->two : NULL,
                                                        branch1 == o->branch1 ?
                                                        NULL : ren1->pair->two, 1);
+                       } else if ((dst_other.mode == ren1->pair->two->mode) &&
+                                  sha_eq(dst_other.sha1, ren1->pair->two->sha1)) {
+                               /* Added file on the other side
+                                  identical to the file being
+                                  renamed: clean merge */
+                               update_file(o, 1, ren1->pair->two->sha1, ren1->pair->two->mode, ren1_dst);
                        } else if (!sha_eq(dst_other.sha1, null_sha1)) {
                                const char *new_path;
                                clean_merge = 0;
diff --git a/t/t3030-merge-recursive.sh b/t/t3030-merge-recursive.sh
index d541544..b23bd9f 100755
--- a/t/t3030-merge-recursive.sh
+++ b/t/t3030-merge-recursive.sh
@@ -23,6 +23,8 @@ test_expect_success 'setup 1' '
        git branch df-3 &&
        git branch remove &&
        git branch submod &&
+       git branch copy &&
+       git branch rename &&

        echo hello >>a &&
        cp a d/e &&
@@ -248,6 +250,24 @@ test_expect_success 'setup 7' '
        git commit -m "make d/ a submodule"
 '

+test_expect_success 'setup 8' '
+
+       git checkout rename &&
+       git mv a e &&
+       git add e &&
+       test_tick &&
+       git commit -m "rename a->e"
+'
+
+test_expect_success 'setup 9' '
+
+       git checkout copy &&
+       cp a e &&
+       git add e &&
+       test_tick &&
+       git commit -m "copy a->e"
+'
+
 test_expect_success 'merge-recursive simple' '

        rm -fr [abcd] &&
@@ -580,4 +600,23 @@ test_expect_failure 'merge-recursive simple w/submodule result' '
        test_cmp expected actual
 '

+test_expect_success 'merge-recursive copy vs. rename' '
+
+       git checkout -f copy &&
+       git merge rename &&
+       ( git ls-tree -r HEAD ; git ls-files -s ) >actual &&
+       (
+               echo "100644 blob $o0   b"
+               echo "100644 blob $o0   c"
+               echo "100644 blob $o0   d/e"
+               echo "100644 blob $o0   e"
+               echo "100644 $o0 0      b"
+               echo "100644 $o0 0      c"
+               echo "100644 $o0 0      d/e"
+               echo "100644 $o0 0      e"
+       ) >expected &&
+       test_cmp expected actual
+'
+
+
 test_done
--
1.7.0

← back to recent threads