Re: Strange bootstrap failure in

[David Ronis <ronis at onsager dot chem dot mcgill dot ca>]

Hi Tim,

I didn't want to hear that: -ffast-math is one of our workhorse options here.
Maybe we could prevail on the developers to clean this up, especially given that 
the problem is in the released 2.95.2.   [BTW, although I haven't checked things
recently (I'll try now) the build failures I've been having have all been
reproducible on a i586 box, with similar software, CFLAGS, etc.].

David

	From: Tim Prince <tprince@computer.org>
	Subject: Re: Strange bootstrap failure in
	To: ronis@onsager.chem.mcgill.ca, gcc-bugs@gcc.gnu.org
	MIME-version: 1.0
	Content-transfer-encoding: 7bit
	X-MSMail-Priority: Normal
	X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2919.5600
	X-Priority: 3
	
	I found this morning that I had to back off CXXFLAGS to -O or -Os in 
order
	to build yesterday's snapshot (a long standing bug in cp/tinfo).  g77 
has
	been
	building at -O and -O2 but not at
	-Os.  There are also long standing bugs with -ffast-math for i686.  If I
	were
	independently wealthy and didn't have to spend 50 hours a week looking 
for
	a job, maybe I would learn some of this stuff.
	
	----- Original Message -----
	From: "David Ronis" <ronis@montroll.chem.mcgill.ca>
	To: <gcc-bugs@gcc.gnu.org>
	Sent: Tuesday, March 21, 2000 9:36 AM
	Subject: Strange bootstrap failure in
	
	
	>
	> I tried an optimized bootstrap build of gcc version 2.96 20000321
	> (experimental) on an i686-linux(2.2.14)-gnu(2.1.3) box with
	> BOOT_CFLAGS
	>
	>       -O3 -march=pentiumpro -funroll-loops -fomit-frame-pointer
	>  -fstrict-aliasing -fforce-mem -fforce-addr -ffast-math -malign-double
	>
	> The last few lines of the log read (with lines broken by me):
	>
	> stage1/xgcc -Bstage1/ -B/usr/local/i686-pc-linux-gnu/bin/ -c -DIN_GCC
	>     -W -Wall -Wtraditional -O2 -O3 -march=pentiumpro -funroll-loops
	>     -fomit-frame-pointer -fstrict-aliasing -fforce-mem -fforce-addr
	>     -ffast-math -malign-double -DHAVE_CONFIG_H -I. -I../../egcs/gcc
	>     -I../../egcs/gcc/config -I../../egcs/gcc/../include
	>     ../../egcs/gcc/gccspec.c
	>
	> stage1/xgcc -Bstage1/ -B/usr/local/i686-pc-linux-gnu/bin/ -DIN_GCC -W
	>   -Wall -Wtraditional -O2 -O3 -march=pentiumpro -funroll-loops
	>   -fomit-frame-pointer -fstrict-aliasing -fforce-mem -fforce-addr
	>   -ffast-math -malign-double -DHAVE_CONFIG_H -o xgcc gcc.o gccspec.o
	>   intl.o prefix.o version.o obstack.o ../libiberty/libiberty.a
	>
	> echo "int xxy_us_dummy;" >tmp-dum.c
	>
	> ./xgcc -B/usr/local/i686-pc-linux-gnu/bin/ -B.
	>        -I/usr/local/i686-pc-linux-gnu/include -S tmp-dum.c
	>
	> make[2]: *** [s-under] Segmentation fault
	> make[2]: Leaving directory `/home/ronis/Project/notar/objdir/gcc'
	> make[1]: *** [bootstrap] Error 2
	> make[1]: Leaving directory `/home/ronis/Project/notar/objdir/gcc'
	> make: *** [bootstrap] Error 2
	>
	> The problem is that ./xgcc SIGSEGV's (when run with no arguments or
	> with -v) and thus I've not been able to run with -save-temps.  This
	> problem was also present in the preceding snapshot (and was reported).
	>
	> David
	>
	> P.S., there are several nasty bugs that crop up with high levels of
	> optimization (notably in g++) and are already present in 2.95.2 and
	> lead to massive g++ testfailures.  [I've reported the ones I've come
	> across, albeit without any ideas how to fix them].
	>