egcs failures (consistent)

[by-tor@by-tor.dyn.ml.org]

For a little more than a week I have experienced the following build
failure with the egcs cvs tree (updated and tried at least once/day):

stage1/xgcc -Bstage1/  -DIN_GCC    -W -Wall -O2 -O6 -mpentium -march=pentium -D_GNU_SOURCE  -DHAVE_CONFIG_H   -o gencodes \
 gencodes.o rtl.o bitmap.o print-rtl.o ` case "obstack.o" in ?*) echo obstack.o ;; esac ` ` case "" in ?*) echo  ;; esac ` ` case "" in ?*) echo  ;; esac ` ` case "" in ?*) echo  ;; esac ` ` case "" in ?*) echo  ;; esac ` 
./gencodes ../../egcs/gcc/config/i386/i386.md > tmp-codes.h
switch format wrong in rtl.read_rtx(). format was: *.
	file position: 3417
../../egcs/gcc/rtl.c:853: Internal compiler error in function read_rtx
make[2]: *** [s-codes] Error 33
make[2]: Leaving directory `/usr/src/objdir/gcc'
make[1]: *** [bootstrap] Error 2
make[1]: Leaving directory `/usr/src/objdir/gcc'
make: *** [bootstrap] Error 2

System is i586, linux 2.1.12x kernel, glibc 2.0.96, binutils 2.9.1.0.13
and the original egcs is: 

gcc version egcs-2.92.11 19980926 (gcc2 ss-980609 experimental)

I've tried with fresh cvs trees as well, no difference.

-- 
     +--------------------------+----------------------------------------+
     | by-tor@by-tor.dyn.ml.org | UNIX _is_ user-friendly. It's just not |
     |       By-Tor@EfNet       | ignorant-friendly and idiot-friendly.  |
     +--------------------------+----------------------------------------+