awk (or at least our port) has a bug where substr() mixes up rune position and byte
position in its calculations.  this is illustrated by the following

	; echo 1/🎄bcdefgh|6.out '{for(i=1; i < 4; i++) print i " " substr($1, i, 1)}'
	1 1
	2 /
	3 �

repairing the algorithm to correctly use byte positions was initially tried, but the code
was fussy, and slow.

this changes the algorithm for substr so that we calculate the byte position of the
start and end of the substr as we count off the wide characters (this is ape, so not runes)

the nice side effect is that this makes substr() about 30% faster for shortish strings.

	0.064u 0.014s 0.098r 	 6.out {s=substr($0,1,3)}
	0.092u 0.017s 0.123r 	 awk {s=substr($0,1,3)}

Notes:
posting note for visibility before this patch is applied.
- quanstro


Reference: /n/atom/patch/applied/awksubstr
Date: Sun Oct  4 02:12:47 CES 2015
Signed-off-by: quanstro@quanstro.net
Reviewed-by: quanstro

--- /sys/src/cmd/awk/run.c	Sun Oct  4 02:12:37 2015
+++ /sys/src/cmd/awk/run.c	Sun Oct  4 02:12:39 2015
@@ -35,7 +35,7 @@
 #include "awk.h"
 #include "y.tab.h"
 
-#define tempfree(x)	if (istemp(x)) tfree(x); else
+#define tempfree(x)	do {if (istemp(x)) tfree(x); }while(0)
 
 /*
 #undef tempfree
@@ -731,8 +731,8 @@
 
 Cell *substr(Node **a, int nnn)		/* substr(a[0], a[1], a[2]) */
 {
-	int k, m, n;
-	char *s, *p;
+	int m, n, c, i;
+	char *s, *b, *e;
 	int temp;
 	Cell *x, *y, *z = 0;
 
@@ -741,31 +741,39 @@
 	if (a[2] != 0)
 		z = execute(a[2]);
 	s = getsval(x);
-	k = countposn(s, strlen(s)) + 1;
-	if (k <= 1) {
-		tempfree(x);
-		tempfree(y);
-		if (a[2] != 0) {
-			tempfree(z);
-		}
-		x = gettemp();
-		setsval(x, "");
-		return(x);
-	}
-	m = getival(y, 1, k);	/* 1 <= m <= k */
+	m = getival(y, 1, Imax);
 	tempfree(y);
 	if (a[2] != 0) {
-		n = getival(z, 0, k-m);
-				/* n <= 0 <= k-m */
+		n = getival(z, 0, Imax);
 		tempfree(z);
 	} else
-		n = k - 1;
+		n = Imax;
+
+	c = 0;
+	b = NULL;
+
+	for (e = s; *e != 0; e += i) {
+		c++;
+		if (c == m)
+			b = e;
+		if (c == m+n)
+			break;
+		i = mblen(e, 100);	/* ok because mblen is careful */
+		if (i < 0)
+			i = 1;
+	}
+
+	/* invalid index? */
+	if(b == NULL)
+		b = e;
+
 	   dprintf( ("substr: m=%d, n=%d, s=%s\n", m, n, s) );
+
 	y = gettemp();
-	temp = s[n+m-1];	/* with thanks to John Linderman */
-	s[n+m-1] = '\0';
-	setsval(y, s + m - 1);
-	s[n+m-1] = temp;
+	temp = *e;
+	*e = 0;		/* with thanks to John Linderman */
+	setsval(y, b);
+	*e = temp;
 	tempfree(x);
 	return(y);
 }
@@ -804,7 +812,7 @@
 	char *p, *t;
 	const char *os;
 	Cell *x;
-	int flag = 0, n, ch;
+	int flag, n, ch;
 	int fmtwd; /* format width */
 	int fmtsz = recsize;
 	char *buf = *pbuf;
@@ -1628,8 +1636,7 @@
 
 Cell *nullproc(Node **a, int n)
 {
-	n = n;
-	a = a;
+	USED(a, n);
 	return 0;
 }
 
@@ -1727,7 +1734,7 @@
 	Cell *x;
 	int i, stat;
 
-	n = n;
+	USED(n);
 	x = execute(a[0]);
 	getsval(x);
 	stat = -1;