@@ -149,4 +149,34 @@ test_expect_success 'diff --no-index allows external diff' '
test_cmp expect actual
'
+test_expect_success 'diff --no-index normalizes mode: no changes' '
+ echo foo >x &&
+ cp x y &&
+ git diff --no-index x y >out &&
+ test_must_be_empty out
+'
+
+test_expect_success 'diff --no-index normalizes mode: chmod +x' '
+ chmod +x y &&
+ cat >expected <<-\EOF &&
+ diff --git a/x b/y
+ old mode 100644
+ new mode 100755
+ EOF
+ test_expect_code 1 git diff --no-index x y >actual &&
+ test_cmp expected actual
+'
+
+test_expect_success 'diff --no-index normalizes: mode not like git mode' '
+ chmod 666 x &&
+ chmod 777 y &&
+ cat >expected <<-\EOF &&
+ diff --git a/x b/y
+ old mode 100644
+ new mode 100755
+ EOF
+ test_expect_code 1 git diff --no-index x y >actual &&
+ test_cmp expected actual
+'
+
test_done
When "git diff --no-index X Y" is run the modes of the files being differ are normalized by canon_mode() in fill_filespec(). I recently broke that behavior in a patch of mine[1] which would pass all tests, or not, depending on the umask of the git.git checkout. Let's test for this explicitly. Arguably this should not be the behavior of "git diff --no-index". We aren't diffing our own objects or the index, so it might be useful to show mode differences between files. On the other hand diff(1) does not do that, and it would be needlessly distracting when e.g. diffing an extracted tar archive whose contents is the same, but whose file modes are different. 1. https://lore.kernel.org/git/20210316155829.31242-2-avarab@gmail.com/ Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> --- t/t4053-diff-no-index.sh | 30 ++++++++++++++++++++++++++++++ 1 file changed, 30 insertions(+)