git-mailinfo munges the patch?

I noticed that the new mailinfo when splitting a message into cover letter and the patch text seems to munge the patch text, applying the same "if content-type is not there then assume latin-1 and recode to utf-8" logic that is applied to the commit log message. That munging should not be done to the patch text, and it appears the current code botches it.

I am a bit too busy with day job today and haven't had a chance to look into this problem fully, but 1.5.0.3 does not seem to have this problem but post 87ab7992 mailinfo is problematic.

On Thu, Mar 29, 2007 at 01:18:51PM -0700, Junio C Hamano wrote:

Ok. I see what you are saying with the old code. Sorry about that. Do you have a sample file that I can play with to test my fix?

Cheers, Don

On Thu, 29 Mar 2007, Don Zickus wrote:

On that note - here's an unrelated simple case that the old mailinfo got right, but the new one seems to screw up: multiple Subject: lines.

The old one would make later Subject: lines override the earlier ones, and I depended on that when I fix up peoples emails to me manually (you wouldn't believe how bad explanations or subject lines people use for perfectly good patches ;)

The current mailinfo seems to just take the first one. 

		Linus

On Thu, Mar 29, 2007 at 02:19:26PM -0700, Linus Torvalds wrote:

I see what happened. The old code allowed rewriting of the mail headers but blocked rewriting of the inbody headers. For some reason I thought it was by accident that the mail headers were allow to be rewritten. Oops. :(

I wrote the code to have both header types block rewriting. I'll change that.

Cheers, Don

Don Zickus <dzickus@redhat.com> writes:

Attached are two files.

The first one is to be split with git-mailsplit and then processed with git-mailinfo; I just tried to make sure what I expect to see as the output is given by 1.5.0 but not with the current 'master'.

The second one is a bundle of the repository the test data comes from. You could use it like this:

	$ mkdir foo && cd foo && git init
        $ git pull ../sample.bdl master

The repository has one branch, 'master', with two commits on it. The first one creates an empty file 'a', and the second one adds a line to the file, which has a Spanish "tilde on top of N" in it, and the contents of the file is encoded in ISO 8859-1.

The commit message of the second one is plain ascii but it could be in UTF-8.

The point is that patch can have payload from different files that are possibly in different encodings, so shoud really be treated as binary data without mangling.

Thanks.

On Thu, Mar 29, 2007 at 02:24:39PM -0700, Junio C Hamano wrote:

Thanks.  I'll play with them later tonight.

Cheers, Don