Re: [PATCH 0/4] More tests for hand-written configure (resend)

From: Junio C Hamano <junkio@cox.net>
Date: Jul 7, 2006, 19:40 UTC
Message-ID: <7vhd1tw6dd.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <20060707162513.25746.57374.stgit@leonov.stosberg.net>

Dennis Stosberg <dennis@stosberg.net> writes:

> I noticed that the autoconf-based solution has replaced Pasky's
> scripts in the pu branch.  Has a final decision been made?

My preference has been to see both sides battle it out without forcing me to decide, but...

> I must admit that I'm less convinced today that a hand-written
> configuration script is better than I was yesterday when I started
> to write the tests.

... I started to share the same feeling after Pavel Roskin made a good point in "git on HP-UX" thread,

    http://thread.gmane.org/gmane.comp.version-control.git/23380/focus=23393

and then after seeing the messages in response to your patch that used `which` from yesterday.

Shell scripts generated by autoconf are almost unreadable, but the way how they detect features have been polished in the field for portability for a long time, and there is no point for us to spend time reinventing the wheel. The configure.ac files are often quite readable even when generated configure scripts are not.

So, I would not veto the use of autoconf, as long as configure stays as an _optional_ mechanism to manage config.mak.gen that is used by the main Makefile. The users for whom the configure script breaks for whatever reason can work it around by simply not using it, instead of having to debug either the unreadable configure or having to install autoconf and debug configure.ac just to build git.

The _optional_ is really the key word here. So "make clean" to clean autoconf intermediate files is good, "make realclean" to remove "configure" script generated from "configure.ac" is also good, but if "make rpm" by default runs "configure", then that is BAD and I would be very unhappy.

I could probably live with "make rpm-using-configure", though.