> ä 2010å11æ25æ 01:20, Junio C Hamano åé:
> By the way, this codepath is shared by all forms of patches "diff --git"
> header, not just binary. Do you see a similar breakage with --no-prefix
> patches that are not binary, and if not why?
Some cases a git diff without "--- path/to/old" and "+++ path/to/new",
will make current git apply -p2 broken.
1. Prue file mode changes:
diff --git a/sub1/file1 b/sub2/file1
old mode 100644
new mode 100755
2. Remove empty file:
diff --git a/sub1/file1 b/sub1/file1
deleted file mode 100644
index e69de29..0000000
3. Add empty file:
diff --git a/sub2/file1 b/sub2/file1
new file mode 100644
index 0000000..e69de29
4. New binary file:
diff --git a/sub2/file1 b/sub2/file1
new file mode 100644
index 0000000000000000000000000000000000000000..4dd0c65a410e8aea1608d6b2dd70e74def0012c5
GIT binary patch
literal 8
Pcmb<-^>JfjWMlvU2^Ilc
literal 0
HcmV?d00001
5. Modify binary file:
diff --git a/sub1/file1 b/sub2/file1
index f96ec1a8358556543a73b3fb93a0d1c2aad8432a..4dd0c65a410e8aea1608d6b2dd70e74def0012c5 100644
GIT binary patch
literal 8
Pcmb<-^>JfjWMlvU2^Ilc
literal 7
Ocmb<-^>JfjWCQ>Qy8&AO
I think all these cases can use this testcase in t/t4120-apply-popt.sh:
+test_expect_success 'apply git diff with -p2 and use default name from header' '
+ sed -e "/^\(---\|+++\) / d" patch.file > patch.newheader &&
+ cp file1.saved file1 &&
+ git apply -p2 patch.newheader
+'
ä 2010å11æ25æ 09:46, Jiang Xin åé:
>
> ä 2010å11æ25æ 01:20, Junio C Hamano åé:
>> Jiang Xin <worldhello.net@xxxxxxxxx> writes:
>>
>>> When patch file generated against two non-git directories using
>>> 'git diff --binary --no-index' without '--no-prefix', the patch
>>> file has patch level greater then 1, and should be applied with
>>> '-p2' option. But it does not work if there are binary differ
>>> in the patch file, it is because in one case the patch level is
>>> not properly handled.
>>>
>>> Signed-off-by: Jiang Xin <jiangxin@xxxxxxxxx>
>>
>> Can you please add a testcase to protect your fix from getting broken by
>> later changes by other people, perhaps to t/t4120?
>>
>
> Yes, I add a testcase in t/t4120, and I will send it as a new PATCH.
> The new testcase is like following:
>
> +test_expect_success 'apply git diff with -p2 and use default name from header' '
> + sed -e "/^\(---\|+++\) / d" patch.file > patch.newheader &&
> + cp file1.saved file1 &&
> + git apply -p2 patch.newheader
> +'
> +
>
>> By the way, this codepath is shared by all forms of patches "diff --git"
>> header, not just binary. Do you see a similar breakage with --no-prefix
>> patches that are not binary, and if not why?
>>
>
> The breakage will appear under these circumstances:
> * the patch is in git style: header with "diff --git".
> * the header does not contain '--- path/to/old' and '+++ path/to/new'.
> * has a patch level greater than 1.
>
> When I do `git diff` against binary files, I find the patch header does not has
> '--- path/to/old' and '+++ path/to/new'. May be there exist other cases, but
> I'm not sure.
>
>>> ---
>>> builtin/apply.c | 3 ++-
>>> 1 files changed, 2 insertions(+), 1 deletions(-)
>>>
>>> diff --git a/builtin/apply.c b/builtin/apply.c
>>> index 23c18c5..d603e37 100644
>>> --- a/builtin/apply.c
>>> +++ b/builtin/apply.c
>>> @@ -1126,6 +1126,7 @@ static char *git_header_name(char *line, int llen)
>>> * form.
>>> */
>>> for (len = 0 ; ; len++) {
>>> + int nslash = p_value;
>>> switch (name[len]) {
>>> default:
>>> continue;
>>> @@ -1137,7 +1138,7 @@ static char *git_header_name(char *line, int llen)
>>> char c = *second++;
>>> if (c == '\n')
>>> return NULL;
>>> - if (c == '/')
>>> + if (c == '/' && --nslash <= 0)
>>> break;
>>> }
>>> if (second[len] == '\n' && !memcmp(name, second, len)) {
>>> --
>>> 1.7.3.2.245.g03276.dirty
>
>
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html