Re: [PATCH] gitweb: fill in gitweb configuration by Makefile

From: Junio C Hamano <junkio@cox.net>
Date: Jul 31, 2006, 06:41 UTC
Message-ID: <7vslki9sy9.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <20060731035737.24181.qmail@web31803.mail.mud.yahoo.com>

- Luben Tuikov <ltuikov@yahoo.com> writes:

Show 13 quoted lines

> -- Martin Waitz <tali@admingilde.org> wrote:
>> Generate gitweb/gitweb.cgi to reduce the need to patch gitweb.cgi by
>> the end user.
>> The GIT installation directory and version are already known by the
>> Makefile, they can be inserted directly into gitweb.
>> All other gitweb configuration parameters can now be specified
>> by providing GITWEB_* variables while building GIT.
>> These are described in gitweb/README.
>
> NACK.
>
> I don't like it.  While this method works, it is too much effort
> to have to run make to do this, plus it pollutes your tree.

I do not quite agree with this reasoning. With that definition of "polluting your tree", all build of git-*.{sh,perl,py} are polluting your tree. If compile/install time customization is handier I do not have problem doing it that way.

> Instead, what you can do is make gitweb.cgi read a text file
> from . which has those variables defined.

I think there was a discussion on having a configuration file for gitweb (not per-repository .git/config but installation wide configuration file). If I recall correctly, the killer argument against it was that you would need to bootstrap the process by somehow telling gitweb.cgi where to read from that configuration file, and if you are going to customize gitweb.cgi before installation for the site that way, then giving a bit more configuration in gitweb.cgi (like Martin did in the patch we are discussion) is simpler.

Show 5 quoted lines

>   - no need to run make to build "gitweb.cgi" or "gitweb.pl" whatever
>     you call it,
>   - no need to pollute your tree with site defined variables,
>   - simple copy (cp) would install a working version, instead of
>     the current cp + patch with local settings method.

While I agree all of these "no need to" makes things slightly simpler, I do not think it is such a big deal -- we are building git-*.{sh,perl,py,c} anyway.