Subject: Re: More on the #define spaghetti
To:, Richard Earnshaw <>
From: Chris Gilbert <>
List: port-arm
Date: 06/11/2001 11:11:02
On Friday 08 June 2001  4:29 pm, Richard Earnshaw wrote:
> I've been thinking further about the current defopt mess, where every
> platform we build has to defopt several things even though it will never
> make use of them, simply because some source file it needs to build
> includes opt_<foo> for some reason.
> An idea for one way to solve some of this mess is for the basic hardware
> options that an ARM platform needs to support to be defopt-ed in a single
> header file, eg opt_hwconf.h, we could then use this for footbridge, ofw
> vidc iomd etc (ie the basic hw build options for an ARM) and save the need
> to include large numbers of opt_foo files at the top of a file.  The
> defopt for each of the supported hw configurations would then just live in
> arm/conf/files.arm

I have to say this sounds great (having just had someone point out that cats 
GENERIC doesn't build because I removed the defopt NETWINDER line which 
locore.S needs *sigh* (and I didn't use a clean build dir to do a build, so 
I've got lots of clutter in it))

> Thinking further on this, we could even use this principal to enable us to
> have a single consinit() function in a single source file in the same sort
> of way that i386 has.  The code would then look something like
> #include "opt_hwconfig.h"
> void
> consinit ()
> {
> 	/* Glue to allow switching console at boot time */
> #ifdef COM_CONSOLE
> 	/* com-console code */
> #endif
> #ifdef VIDC
> 	/* vidc console init code */
> #endif
> 	/* Footbridge console init code */
> #endif
> ...
> }
> Comments?

sounds good, although part of me is inclined to say that we should put inside 
each of those blocks a call to other the relevant init function, which are 
only there when that define is available.  But that's just semantics :)