incremental building in 4.2

[Rask Ingemann Lambertsen]

On Tue, Jun 19, 2007 at 06:15:54PM -0700, Mike Stump wrote:
> This triggers after two reconfigurations of the toplevel of a cross  
> compiler that has things like ar, lipo, nm, ranlib and strip in the  
> path instead of the build tree.  After this has been done once, all  
> further builds are hosed.  I do this all the time.  The complete set  
> of variables so hosed are AR_FOR_TARGET, LIPO_FOR_TARGET,  
> NM_FOR_TARGET, RANLIB_FOR_TARGET and STRIP_FOR_TARGET.
> 
> I've not seen the build problems in the 4.3 tree yet, I have reason to  
> believe they don't happen there.  A quick check turns up the fact we  
> don't use the cache at the top-level, so, they don't cause problems in  
> 4.3.  Upon further checking, the problem is latent there as well.

   I've run into this problem several times in 4.3 building my 16-bit x86
port. The symptom would be this:

1) svn update. Works fine.
2) Configure. Works fine.
3) make. Works fine.
4) svn update. Works fine.
5) make. Fails running "rc libgcc.a ..." where it should be running
"ia16-elf-ar rc libgcc.a ...".

The workaround would be to run configure again or override AR_FOR_TARGET etc
when calling make. It's been a few months since I last forgot to use a
workaround.

-- 
Rask Ingemann Lambertsen