From: Junio C Hamano <gitster@pobox.com>
Date: Wed, 08 May 2019 02:56:43 GMT
Subject: Re: [PATCH 2/2] format-patch: Make --base patch-id output stable
Message-ID: <xmqq1s19k61g.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <155725118838.14659.6891235578536334187@swboyd.mtv.corp.google.com>

Stephen Boyd <sboyd@kernel.org> writes:

> Quoting Junio C Hamano (2019-05-06 21:38:24)
>> Stephen Boyd <sboyd@kernel.org> writes:
>> 
>> > I wonder if we need to make some other sort of form of
>> > "prerequisite-patch-id:" here and let that be a legacy form of the
>> > patch-id so that users know that they have a fixed version of this code?
>> > Maybe "prerequisite-stable-patch-id:"? Or we don't have to care because
>> > it's been broken for anything besides the most trivial type of patches
>> > and presumably users aren't able to use it with 'patch-id --stable'?
>> 
>> Do projects actively use -O<orderfile> when generating the patches?
>> I had an impression that not many do, and without -O<orderfile> in
>> the picture, --unstable/--stable would not matter, no?
>> 
>> So, I am not sure if this matters very much in practice.
>> 
>
> I'm not really concerned with projects using -O<orderfile> for patch
> generation. 

I think I misunderstood, then.  I have been assuming that the order
of target file paths was the primary thing that contributes to the
differences between --[un]stable modes, but apparently I forgot
about that 30e12b92 ("patch-id: make it stable against hunk
reordering", 2014-04-27) affects even a patch that touches a single
path.

If we advise "--stable" in the documentation to those who wants to
interpret "--base", then I agree with the goal of this series to
make sure that is what actually is happening.

Thanks for working on this.