gawk: fix for x86-64

[Jean-Marc Saffroy]

Hi,

I saw that some tests in the gawk (3.1.7) test suite fail on x86-64:

$ make check -C test/ &> log
$ grep differ: log
./double1.ok _double1 differ: char 1, line 1
./double2.ok _double2 differ: char 247, line 4
./intformat.ok _intformat differ: char 1, line 1

The patch below fixes these problems. On x86-64, LONG_MAX is actually 
64-bit wide, and comparing a double seems to lose accuracy in some cases 
(eg. when the double contains something close to LONG_MAX but with a 
53-bit mantissa).


Cheers,
Jean-Marc

-- 
address@hidden

diff -r d87a84edf71e node.c
--- a/node.c    Wed Apr 28 18:47:29 2010 +0200
+++ b/node.c    Thu Apr 29 00:16:24 2010 +0200
@@ -212,7 +212,7 @@
 
        /* not an integral value, or out of range */
        if ((val = double_to_int(s->numbr)) != s->numbr
-           || val < LONG_MIN || val > LONG_MAX) {
+           || val < INT_MIN || val > INT_MAX) {
                /*
                 * Once upon a time, if GFMT_WORKAROUND wasn't defined,
                 * we just blindly did this:
@@ -253,13 +253,13 @@
                goto no_malloc;
        } else {
                /* integral value */
-               /* force conversion to long only once */
-               register long num = (long) val;
+               /* force conversion to int only once */
+               int num = (long) val;
                if (num < NVAL && num >= 0) {
                        sp = (char *) values[num];
                        s->stlen = 1;
                } else {
-                       (void) sprintf(sp, "%ld", num);
+                       (void) sprintf(sp, "%d", num);
                        s->stlen = strlen(sp);
                }
                s->stfmt = -1;