> I ran into this while doing a package compile of pvm.  The system is a
> vanilla 1.4 install. The binaries are those distributed from the 1.4
> relase.  The package source is the one distributed with the 1.4 release.
> 
> This is the first time I've ever seen an internal compiler error quite
> like the one that follows.  (Mmmmmm Lispy).
> 
> -brian.
> 
> 
> futplex# make
> ===>  Building for pvm-3.4.7
> ./lib/aimk -here -f ./conf/`./lib/pvmtmparch`.def -f ./Makefile.aimk all
> making in . for NETBSD
> building in src
> cd src; ../lib/aimk CC=cc  F77=/usr/bin/f77 install
> making in NETBSD/ for NETBSD
> done building in src
> building tracer
> cd tracer; ../lib/aimk CC=cc  F77=/usr/bin/f77 install
> making in NETBSD/ for NETBSD
> cc -O  -DIMA_NETBSD -DSOCKADHASLEN -DRSHCOMMAND=\"/usr/pkg/bin/ssh\"
> -DHASERRORVARS -DNOREXEC -DNEEDMENDIAN -DLOCALBASE=\"/usr/pkg\" -c
> -I../../tracer -I../../include -I../../tracer ../../tracer/trcmess.c
> ../../tracer/trcmess.c: In function `trc_handle_trace_message':
> ../../tracer/trcmess.c:787: internal error--unrecognizable insn:
> (jump_insn 1448 1522 1449 (set (pc)
>         (if_then_else (leu (minus:SI (mem:SI (reg:SI 237))
>                     (const_int -9))
>                 (const_int 8))
>             (plus:SI (sign_extend:SI (mem:HI (plus:SI (pc)
>                             (mult:SI (minus:SI (mem:SI (reg:SI 235))
>                                     (const_int -9))
>                                 (const_int 2)))))
>                 (label_ref:SI 1449))
>             (pc))) -1 (insn_list 1519 (insn_list 1522 (nil)))
>     (expr_list:REG_DEAD (reg:SI 237)
>         (expr_list:REG_DEAD (reg:SI 235)
>             (nil))))
> /usr/src/gnu/usr.bin/egcs/common/../../../dist/gcc/toplev.c:1360: Internal
> compiler error in function fatal_insn
> *** Error code 1

I've seen this a number of times.   Not sure if it exists because of compiler
errors or what, but I end up getting around it by turning off optimisation (remove
the -O).

Otherwise, I don't know the answer.  I get this in two locations of the vanilla
1.4 source tree, and have just used that cheat to get around it.   Haven't 
pursued the reason myself otherwise, at least until I get a full system
build complete (takes an EXTREMEly long time on uVAX II's).

--skots
--
Scott G. Akmentins-Taylor     InterNet: staylor@mrynet.com
MRY Systems			        staylor@idt.net, staylor@mrynet.lv
Westlake Village, CA  USA		
    (Skots Gregorijs Akmentins-Teilors -- just call me "Skots")
		----- Labak miris neka sarkans -----