configure problems

[gregory.menke at gsfc.nasa.gov]

I'm working up at the cvs head using gcc-3.3.3.  I'm running bootstrap
with

autoconf-2.59 with autoconf-2.59-quoting-20040817-1.diff
automake-1.92

and it seems to complete OK.  My standard ./configure completes OK
too.  However, when I make, a number of MIPS targets fail to build
down in libcpu, among other places.  

A representative location is, when linking isr-entries.o, the Makefile
is looking for interrupts/isr-entries.o, but the file is actually
located in the parent directory.  The failure generally occurs when a
series of makes in recursive directories generate their .o's in the
parent directory, but the Makefile is expecting them to be in the
subdirectories and so fails when doing ld -r to aggregate a bunch of
.o's.

The other possibly related problem is the configure pass occasionally
leaves $() macros in the makefile (libcpu as well), causing syntax
errors.

Simply editing the Makefile to look in the right spot or to comment
out the spurious macro allows the make to proceed to completion.
Built images run fine.

However, the mcp750 and mtx603e both build fine.

THis behavior has persisted for a month or so now, despite tree-wide
cvs updates and re-bootstrapping.  I imagine its not something broken
in the RTEMS source, the likelihood being something is broken locally.

Thanks for any hints,

Gregm