From: Junio C Hamano <gitster@pobox.com>
Date: Sun, 14 Sep 2014 05:23:03 GMT
Subject: Re: [PATCH] help: ensure that common-cmds.h is only included by help.c
Message-ID: <CAPc5daXDCAi3eP3YmXfcO+9ncN8=b6tCGUFUxwKE=MuRBuXvEg@mail.gmail.com>
In-Reply-To: <5414f6b2.XwoOyI03zlXiZ3ds%perryh@pluto.rain.com>

On Sat, Sep 13, 2014 at 7:00 PM, Perry Hutchison <perryh@pluto.rain.com> wrote:
> David Aguilar <davvid@gmail.com> wrote:
>> Add a #ifndef guard to ensure that common-cmds.h can only
>> be included by help.c.
>
> This strikes me as a very peculiar, and sub-optimal, way of
> achieving the purpose.  If these definitions are intended to
> be private to help.c, why not put them there and eliminate
> common-cmds.h entirely?

Have you studied where common-cmds.h comes from?
After you have done so, would you make the same suggestion?

Having said that, I also do not think this is such a good idea.
Wouldn't the new "check" script added in this series a better
place? For example, it may want to make sure that git-compat-util.h
(or a couple of its equivalents) is the first file included in any mainline
C source file, and such an inclusion is done unconditionally.

Which would mean that the checker would scan *.c files with grep
or a Perl script. It would be trivial to enforce "nobody other than these
small selected C files is allowed to include common-cmds.h" rule.

Regarding the other patch that butchers many *.h files, I am not
still very enthused. Including cache.h at the beginning of branch.h,
for example, would mean git-compat-util.h ends up included at the
beginning of branch.h.