Re: [PATCH] format-patch: generate valid patch with diff.noprefix config

From: Đoàn Trần Công Danh <congdanhqx@gmail.com>
Date: Jun 4, 2020, 23:26 UTC
Message-ID: <20200604232619.GA6329@danh.dev>
In-Reply-To: <xmqq367aposu.fsf@gitster.c.googlers.com>

On 2020-06-04 13:18:57-0700, Junio C Hamano <gitster@pobox.com> wrote:

Show 12 quoted lines

> Laurent Arnoud <laurent@spkdev.net> writes:
> 
> > I don't know if its a bug but I founded strange that I needed to use an alias
> > "git -c diff.noprefix=false format-patch" to generate a patch that I can apply
> > directly with "git am".
> 
> The same thing can be said about the diff.relative option.
> 
> As I do not use either of these variables myself, I am somewhat
> indifferent, if those who set these variables find the consequences
> of doing so unpleasant.  As people often say, if it hurts, then...
> ;-)

FWIW, people need to dig into git-config(1) to know diff.noprefix We don't mention it in git-diff(1). Me neither until today.

Show 5 quoted lines

> Because the recipient of format-patch output who consumes it is
> typically different from the one who generates it, it probably does
> not make much sense to attempt linking diff.noprefix=true with the
> -p0 option (there isn't even a configuration for 'git am -p<num>',
> if I am not mistaken).

Could we extend the patch in someway that can hint git-am(1) to use "-p0" if patch was generated by --no-prefix? I couldn't come up with any idea right now, since all parts of the patch is very busy with a lot of information. An RFC 822 header, perhap?

Of course old git-am(1) couldn't understand the extension. And the new git-am may come up with false positive?

Show 23 quoted lines

> Depending on the project a user works with, allowing
> log/show/format-patch to honor certain diff.* configuration
> variables, without a way to countermand them with more specific
> configuration for log/show/format-patch, may smell like a bug.  
> 
> I am not sure where to draw the line, though.  Would we treat only
> format-patch and no other commands in the log family specially?
> Would we treat each commands in the log family specially and
> separately, so that you could say "diff and show uses noprefix, but
> 'log -p' and 'whatchanged' uses the standard a/ and b/ prefix and
> format-patch uses old/ and new/ prefix" independently?
> 
> > I didn't know the -p option but to send a patch to a
> > mailinglist it should have the prefix I guess ?
> 
> The participants in this project would certainly find it unusual
> when they see a prefix-less patch.
> 
> There probably are projects older than Git whose convention is to
> use .noprefix; we didn't want to force them to switch and instead
> accomodated their preference with .noprefix but in hindsight, it may
> have been a better idea to force one true way to everybody, which
> would have kept the world simpler.  I dunno.

I do use format-patch --no-prefix. It's unlikely that I will use diff.noprefix, though. I think other people also have different preferences.

-- 
Danh