On Wed, Dec 20, 2006 at 08:56:39AM +0100, Paolo Bonzini wrote:
> >>If possible, I would like a proof of concept that it is possible to 
> >>migrate the building of the extra parts incrementally without breaking 
> >>non-converted targets, before committing (unless another maintainer 
> >>overrules me).
> >
> >OK, I'll do that (later).
> 
> Just to make it clearer, I'm ok with having to convert a directory at a 
> time (arm, rs6000, i386, ...).  So, if your proof-of-concept works only 
> with one target, but breaks only those targets that share the same cpu 
> directory, I'm fine with this.

I'd rather have it break nothing.  So that's what I did.  Of course,
the risk of collateral damage here is quite high; our configuration is
a terrible maze today.  So, I have not removed any information about
which startfiles to build from the gcc directory yet.  If information
is provided by both the gcc and libgcc configury, then they are
compared and there will be loud failure if they differ.

The attached two patches correct some omissions in my previous
config.host file, and move building of EXTRA_PARTS to the libgcc
subdirectory for all GNU/Linux targets.  It involves updating
config.host to indicate which parts, and adding the necessary
rules.  The rules are similar to the ones in gcc/ with only
minor adjustments (for automatic dependency tracking, using $(CC),
removal of $(T), et cetera).  It's easy to do other targets.

Both patches committed to the branch.  Look good?

-- 
Daniel Jacobowitz
CodeSourcery