🚀 go-pugleaf

enquire 4.3, as distributed with GCC 2.3.1, generates some garbage bits in
float.h (sparc, SunOS 4.1.2).  This causes spurious failures in regression
tests.  For example, I generated float.h twice and got the following spurious
difference report:

diff /reb0.1/sun4c-4.1.2/lib/gcc-lib/sparc-sun/2.3.1/include/float.h float.h
64c64
< #define LDBL_MIN (__extension__ ((union __convert_long_double) {0x100000, 0x0, 0x8000000, 0xf82b8000}).__convert_long_double_d)
---
> #define LDBL_MIN (__extension__ ((union __convert_long_double) {0x100000, 0x0, 0x8000000, 0xf8290000}).__convert_long_double_d)
70c70
< #define LDBL_MAX (__extension__ ((union __convert_long_double) {0x7fefffff, 0xffffffff, 0x8000000, 0xf82b8000}).__convert_long_double_d)
---
> #define LDBL_MAX (__extension__ ((union __convert_long_double) {0x7fefffff, 0xffffffff, 0x8000000, 0xf8290000}).__convert_long_double_d)

On a sparc, a long double is only two integers long, so the last two integers
of LDBL_MIN and LDBL_MAX are garbage.


Here is a patch.

===================================================================
RCS file: RCS/enquire.c,v
retrieving revision 4.3
retrieving revision 4.3.1.1
diff -c -r4.3 -r4.3.1.1
*** enquire.c	1992/07/31 06:29:07	4.3
--- enquire.c	1992/11/02 21:49:19	4.3.1.1
***************
*** 906,911 ****
--- 906,912 ----
  char *fake_f_rep(type, val) char *type; Long_double val; {
  	static char buf[1024];
  	union { int i[4]; Long_double ld;} u;
+ 	u.i[0] = u.i[1] = u.i[2] = u.i[3] = 0;
  	u.ld = val;
  	sprintf(buf, "(__extension__ ((union __convert_long_double) {0x%x, 0x%x, 0x%x, 0x%x}).__convert_long_double_d)",
  		u.i[0], u.i[1], u.i[2], u.i[3]);