Count utf8 runes, not bytes when determining string width. Note

that pad() still counts bytes, but it's currently only used for
1 byte runes.

Fixes #612.

R=r
CC=golang-dev
https://golang.org/cl/217064

src/pkg/fmt/fmt_test.go

@@ -86,6 +86,7 @@ var fmttests = []fmtTest{
 
 	// width
 	fmtTest{"%5s", "abc", "  abc"},
+	fmtTest{"%2s", "\u263a", " \u263a"},
 	fmtTest{"%-5s", "abc", "abc  "},
 	fmtTest{"%05s", "abc", "00abc"},
 

src/pkg/fmt/format.go

@@ -7,6 +7,7 @@ package fmt
 import (
 	"bytes"
 	"strconv"
+	"utf8"
 )
 
 const (
@@ -127,7 +128,7 @@ func (f *fmt) padString(s string) {
 	var padding []byte
 	var left, right int
 	if f.widPresent && f.wid != 0 {
-		padding, left, right = f.computePadding(len(s))
+		padding, left, right = f.computePadding(utf8.RuneCountInString(s))
 	}
 	if left > 0 {
 		f.writePadding(left, padding)

src/pkg/fmt/print.go

@@ -43,7 +43,9 @@
 	For numeric values, the width and precision flags control
 	formatting; width sets the width of the field, precision the
 	number of places after the decimal, if appropriate.  The
-	format %6.2f prints 123.45.
+	format %6.2f prints 123.45. The width of a field is the number
+	of Unicode code points in the string. This differs from C's printf where
+	the field width is the number of bytes.
 
 	Other flags:
 		+	always print a sign for numeric values