RFC patch[git-subtree.sh] Use raw subject and body modifier "%B" instead of "%s%n%n%b" for commit

I don't know if it is the right place to post this patch, I have sended an email to the original author apenwarr and have no response. According to <https://github.com/apenwarr/git-subtree/blob/master/THIS-REPO-IS-OBSOLETE>, this is the place, but <https://github.com/git/git/blob/master/contrib/README> says different, which is really confusing. Anyway, here I am.

Recently, I imported a foreign git project as a sub directory into a main repo which I intend to maintain as primary.

Due to the project I imported has its own remote repo which hosted on the github, I expected after a 'git-subtree.sh split' the newly generated subtree branch would be exactly identical to the original branch. Unfortunately, it is not. I have fixed the committer date and make everything looks the same with the original branch, but they just did not end up with same commit sha1 hash. Then, I used `git cat-file -p` to view the raw output of the both commits and found that the commit generate by git-subtree has a extra 'new-line' character appended at the end of the subject which causes the problem.

I checked the source and found "%s%n%n%b" were used to generate the commit message, this works the fine when a commit has a subject as well as a body, but most of my commits only have a subject under which condition a extra 'new-line' character is appended.

Instead, a raw subject and body message modifier '%B' should be used.

Though I think this patch should be applied by default, but the mistake has been there for a long time, applying this patch may cause the patched git-subtree generate a different branch for those whose subtree branch has already been generated using the old git-subtree. Maybe this should be explained in the help or man page, and add a condition check or a compatible mode somehow.

Techlive Zheng (1):
  subtree.sh: Use raw subject and body modifier "%B" instead of "%s%n%n%b"

 contrib/subtree/git-subtree.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

"%s%n%n%b" is not always equal to "%B". If the commit msg does not have a body, this will append an extra new-line character to the msg title which would cause the splited commit has a new sha1 hash. In most cases, this does not matter, but for a project which did not merged using this script initially, the 'split' command would not genereate the same commits as the orginal which may cause conflicts.

Signed-off-by: Techlive Zheng <techlivezheng@gmail.com>
---
 contrib/subtree/git-subtree.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Techlive Zheng <techlivezheng@gmail.com> writes:

This looks good to me. I assume this passes all the tests. Can you add a test for this bug so we don't regress? Junio, I am good with this patch as soon as we get a test for the problem.

Thanks!

                             -Dave

greened@obbligato.org writes:

I've applied this patch to my working copy but I'm not finding that I can recreate the original problem when the patch is disabled.

I assumed the scenario you're trying to fix is:

- Make some commit C to project A with a one-line message
- work, commit, work...
- Add project A as a subproject
- work, commit, work...
- Split project A off into a separate repository

After this, commit C with the one-line message in the split-off projet should have the same hash it had before project A was incorporated as a subproject.

As I understad it, you saw the post-split commit having a different hash?

Is that right?  I am not seeing that problem even without your patch.

I want to make sure I understand what the problem is so I can test for it.

Thanks!

                      -David

greened@obbligato.org writes:

     ^ not

                          -David

Techlive Zheng <techlivezheng@gmail.com> writes:

This is the place.

Ok.

I would have thought so too.

Hmm.

Ah.  Yes, we should fix this.

Ok.

The problem is in the split code? I'm not sure this is a big issue. I can run some experiments.

                      -Dave