On May 14, 2005, at 4:58 AM, Helge Jensen wrote:
>
>> What is fundamentally wrong is that autoconf is needed at all.
>>
>
> While that may be true, so is the state of the world. Refusing to
> accept that will not improve things.
>
Well explained, Helge.
Anyone can create a simple standalone Makefile that will run on
nearly every distribution of Linux. But try having that Makefile
work on all flavors of SunOS, HP-UX, AIX, IRIX, Tru64, FreeBSD,
NetBSD, Darwin... and work with every possible C compiler, both free
and commercial, installed in different locations on each syste.m
Heck, even the basic C libraries have different quirky behaviors on
each system!
At my previous job, they refused to use autoconf, and my job *was* to
do exactly this. Our product had to compile on 7 flavors of Unix,
and my job was to write a Makefile / build system. It was the
biggest nightmare ever... huge, complex, impossible to maintain.
Contrast that against a few macro calls in configure.in: "find a C
compiler", "does the system have X feature? how about this other
feature?". And it *works*, because autoconf is a gigantic library of
knowledge, a compendium of OS-specific quirks and workarounds
gathered from thousands of programmers over many years. When you use
it, you're standing on the shoulders of everyone's collective
experience.
I think the problem is that Scott is grimacing at the the ugliness of
the cure, having never seen the horror of the original disease. :-)
That said, the world is slowly changing; Linux and BSD are pushing
Unix back towards unification again. It's gotten to the point where
non-GNU systems (like SunOS) are now considered 'eclectic'
platforms. Who'da thunk it?
OK, have we wandered off-topic enough yet?
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Sat May 14 17:13:40 2005