🚀 go-pugleaf

About Peter Miller's bug report, I wrote ``GCC is acting correctly in
the cases that you cite.'' I was right about one case, but wrong about
the other -- there is a bug in GCC 2.2.2 (sparc, SunOS 4.1.2).  Here's
a simple test case that illustrates it:

Program:
	double normalize(x) double x; { if (x == 0) x = 0; return x; }
	main() { printf("%g\n", normalize(-0.0)); }

Transcript:
	$ gcc -O -v t.c
	Reading specs from /reb0.1/sun4c-4.1.2/lib/gcc-lib/sparc-sun/2.2.2/specs
	gcc version 2.2.2
	 /reb0.1/sun4c-4.1.2/lib/gcc-lib/sparc-sun/2.2.2/cpp -lang-c -v -undef -D__GNUC__=2 -Dsparc -Dsun -Dunix -D__sparc__ -D__sun__ -D__unix__ -D__sparc -D__sun -D__unix -D__OPTIMIZE__ t.c /usr/tmp/cca24499.i
	GNU CPP version 2.2.2 (sparc)
	 /reb0.1/sun4c-4.1.2/lib/gcc-lib/sparc-sun/2.2.2/cc1 /usr/tmp/cca24499.i -quiet -dumpbase t.c -O -version -o /usr/tmp/cca24499.s
	GNU C version 2.2.2 (sparc) compiled by GNU C version 2.2.2.
	 as -o t.o /usr/tmp/cca24499.s
	 /reb0.1/sun4c-4.1.2/lib/gcc-lib/sparc-sun/2.2.2/ld -e start -dc -dp /lib/crt0.o -L/reb0.1/sun4c-4.1.2/lib/gcc-lib/sparc-sun/2.2.2 -L/reb0.1/sun4c-4.1.2/lib t.o -lgcc -lc -lgcc
	$ ./a.out
	-0

The output should be `0', not `-0'.  The problem is that GCC 2.2.2 is
optimizing away `if (x == 0) x = 0;', but this optimization is incorrect
for the case where x is -0.0 in IEEE floating point.