From: Philip Oakley <philipoakley@iee.org>
Date: Mon, 15 Aug 2016 12:37:33 GMT
Subject: Re: storing cover letter of a patch series?
Message-ID: <DD86BC6E2E3245BA991E4D65CE66E4A8@PhilipOakley>
In-Reply-To: <CACsJy8BdmR5USJvjJ6xbjj=bP787tdS72_oL+PDq0D+FPYmiPA@mail.gmail.com>

From: "Duy Nguyen" <pclouds@gmail.com>
> On Mon, Aug 15, 2016 at 1:28 PM, Stefan Beller <sbeller@google.com> wrote:
>> On Sun, Aug 14, 2016 at 12:15 AM, Jacob Keller <jacob.keller@gmail.com>
>> wrote:
>>> On Sat, Aug 13, 2016 at 1:49 AM, Duy Nguyen <pclouds@gmail.com> wrote:
>>>> On Tue, Aug 9, 2016 at 12:27 AM, Stefan Beller <sbeller@google.com>
>>>> wrote:
>>>>> is what you want. Maybe we want to see a patch that adds the reverse
>>>>> functionality as well, i.e. git-am will store the the cover letter as
>>>>> the
>>>>> branch description and git-merge will propose the branch description
>>>>> for
>>>>> the merge commit.
>>>>
>>>> I almost suggested the same, but there is a problem with this
>>>> approach: if you're are on a detached head, where does git-am save it?
>>
>> What would the user expect? We can have a range of expectations:
>> 1) reject and error out git-am
>> 2) warn about not saving branch.description and continue with am
>> 3) have a (maybe special) branch.HEAD.description thing, same for
>> FETCH_HEAD etc
>> 4) have a config option to choose between 1 and 2, if unset default to 1
>>
>> I think 3 is a bad choice.
>> 4 seems reasonable to me, though I wonder if some people use git-am in
>> a scripted workflow with a detached head and then create the branch
>> afterwards?
>> So
>>
>> 5) create a branch for them? (such as $(date)-${subject})
>>
>> My gut reaction doesn't like 5 either.
>
> I'm starting to think option 6 (storing cover latter as an empty
> commit at tip then git-merge replaces it with a merge commit in a
> permanent history) may be the way to go. It handles detached heads
> just fine, we have reflog to store older cover letters. Though it will
> not play nice with 'git commit --amend' and 'git reset' for people who
> rewrites history heavily during development, but maybe 'git rebase -i
> --autosquash' would be an ok workflow alternative.
> -- 

[sorry if this is not the right place to 'drop in'..]
I appreciate there has been a lot of discussion, but it mainly appears to be
about an upstream / integration viewpoint.

I'd hate it if there was a one size fits all solution that was only focused
on one important use case, rather than having at least a simple fallback for
simple folk.

Personally I liked the idea that I could start my patch series branch with a
simple 'empty' commit with a commit message that read "cover! <subject of
the series>" and continue with the cover letter. It's essentially the same
as the fixup! and squash! idea (more the latter - it's squash! without a
predecessor). For moderate size series a simple 'git rebase master..' is
sufficient to see the whole series and decide which need editing, rewording,
swapping, checking the fixups, etc.

Format-patch would then be taught to spot that the first commit in the
series is "cover! <subject>" and create the usual 0/N cover letter. Git Gui
may need to be taught to recognise cover! (haven't checked if it recognises
an empty commit squash!). Possibly 'git commit' may want a --cover option to
massage the commit message and add --allow-empty, but that's finesse.

I've no problem with more extensive methods for those preparing very big
patch series, or with those needing to merge together a lot of series and
want to keep the cover letters, but ensuring that a simple flow is possible
should still be there.
--
Philip