From: Vegard Nossum <vegard.nossum@oracle.com>
Date: Tue, 22 Oct 2019 12:11:22 GMT
Subject: Re: email as a bona fide git transport
Message-ID: <de6dd8b5-5c28-d0b2-d3fc-e72a6d643105@oracle.com>
In-Reply-To: <1a259d8d-b3d1-b64e-07c3-ba143b42c442@oracle.com>

On 10/20/19 8:28 AM, Vegard Nossum wrote:
> 
> On 10/20/19 5:17 AM, Willy Tarreau wrote:
>> On Fri, Oct 18, 2019 at 03:14:56PM -0400, Theodore Y. Ts'o wrote:
>>> On Fri, Oct 18, 2019 at 06:50:51PM +0200, Vegard Nossum wrote:
>>>> The problem I ran into with putting the metadata at the end was
>>>> detecting where the diff ends. A comment in 'git apply' suggested that
>>>> detecting the difference between "--" as a diff/signature separator and
>>>> as part of the diff is nontrivial in the sense that you need to 
>>>> actually
>>>> do some parsing and keep track of hunk sizes.
>>>
>>> Could we cheat by having "git format-patch" add a "Diff-size" in the
>>> header which gives the number of lines in the diff so git am can just
>>> count lines to find the Trailer section?
>>
>> Be careful with this, it starts like this and ends up with non-editable
>> patches. I'd rather have git-am use best-effort detection of the end.
> 
> Expect filesystem developers to come up with a format that uses extents ;-)
> 
>> Also when dealing with stable backports, I've done a lot of
>> "cat foo.diff >> bar.patch" to fixup some patches in which I just had
>> to move some parts around. Having to count lines and edit a counter
>> somewhere is going to become really painful.
> 
> I almost have some new patches ready for putting the metadata after the
> patch using a very bare-bones diff parser (it's actually not that bad),
> I just need to fix a few corner cases that are causing breakage in the
> git test suite.

I sent v2 of the patches (with metadata _after_ the diff) to the git
list here:

https://public-inbox.org/git/20191022114518.32055-1-vegard.nossum@oracle.com/T/#u

As I wrote in there, we could already today start using

   git am --message-id

when applying patches and this would provide something that a bot could
annotate with git notes pointing to lore/LKML/LWN/whatever. I think that
would already be a pretty nice improvement over today's situation.

Sadly, since the beginning of 2018, this was only used for a measly
~0.14% of all non-merge commits in the kernel:

$ git rev-list --count --no-merges --since='2018-01-01' --grep 
'Message-Id: ' linus/master
178

$ git rev-list --count --no-merges --since='2018-01-01' linus/master
130777

So how can we spread the word about --message-id and get maintainers to
actually use it? I don't suppose it's reasonable to change the 'git am'
default setting?


Vegard