Re: [PATCH] commit & merge: modularize the empty message validator

From: Kaartic Sivaraam <kaarticsivaraam91196@gmail.com>
Date: Jul 14, 2017, 17:49 UTC
Message-ID: <1500054552.18990.8.camel@gmail.com>
In-Reply-To: <xmqqiniww37i.fsf@gitster.mtv.corp.google.com>

On Thu, 2017-07-13 at 12:23 -0700, Junio C Hamano wrote:

Show 5 quoted lines

> All good points; if it bothers you that "commit" and "merge" define
> "emptyness" of the buffer differently too much, I think you could
> persuade me to unify them to "the buffer _must_ contain no bytes",
> i.e. not special-casing sign-off lines only "commit".
>

Intereseting, let me give it a try.

To persuade you with this, I have to convince you that the current behaviour (special-casing of sign-off lines) is defective and/or biased. It really is for quite a few reasons,

            * Though it's not apparent, it indirectly seems to be hindering
            (to some extent) the idea of including the sign-off (or) other
            trailers which *can't be modified* by the user.

            IOW, the current behaviour seems make the contributors/users
            falsely believe (at least to some extent) that git *does* have
            trailers for commit messages and thus preventing them from coming
            up with ideas that could make "untouchable trailers" a reality.

            Thus, consider "the buffer _must_ contain no bytes" hoping this
            would initiate a "Butterfly effect" :)

        * Looking from an implementation perspective, it's biased in that
        it checks only for sign-offs. Making it work in general is
        difficult as there's no standard definition for the term
        <trailer>. That's because it varies with respect to usage, I
        think. Different people/projects may consider different lines to
        be trailer lines. A few examples are,

            * Bug:
            * Fixes:
            * Change-id:
            * Helped-by:

        Moreover, some people may wish to have commit messages that only
        have such trailers (e.g. "Fixes:"). So, it's difficult to do a
        generalized implementation that aborts when the message is empty
        or consists only of trailers.

        Thus, consider "the buffer _must_ contain no bytes" because it's
        not easy to define what a <trailer> means and special casing
        "sign-off" is biased.

        * Imagine a hypothetical version of git that aborts when the
        <message> is empty though a <trailer> is present. This would
        quite possibly instigate controversies as the "hypothetical git"
        reduces the "valid commit messages" and would quite possibly
        reject a commit message as "empty" (which is uncommunicative)
        though a previous version (which did not have this change)
        accepted a similar message.

        SO, bringing in the Occam's razor, let's choose the option that's
        the simplest and makes the fewest assumptions.

Thus, I conclude that the considering a commit message consisting only of <trailer>s as empty isn't a good idea and should be dropped.

-- 
Kaartic