fmt, log: stop using unicode

$ go list -f '{{.ImportPath}} {{.Deps}}' fmt log
fmt [errors io math os reflect runtime strconv sync sync/atomic syscall time unicode/utf8 unsafe]
log [errors fmt io math os reflect runtime strconv sync sync/atomic syscall time unicode/utf8 unsafe]

R=bradfitz, rogpeppe, r, r, rsc
CC=golang-dev
https://golang.org/cl/5753055

src/pkg/fmt/export_test.go

+// Copyright 2012 The Go Authors.  All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package fmt
+
+var IsSpace = isSpace

src/pkg/fmt/fmt_test.go

@@ -13,6 +13,7 @@ import (
 	"strings"
 	"testing"
 	"time"
+	"unicode"
 )
 
 type (
@@ -828,3 +829,13 @@ func TestBadVerbRecursion(t *testing.T) {
 		t.Error("fail with value")
 	}
 }
+
+func TestIsSpace(t *testing.T) {
+	// This tests the internal isSpace function.
+	// IsSpace = isSpace is defined in export_test.go.
+	for i := rune(0); i <= unicode.MaxRune; i++ {
+		if IsSpace(i) != unicode.IsSpace(i) {
+			t.Errorf("isSpace(%U) = %v, want %v", IsSpace(i), unicode.IsSpace(i))
+		}
+	}
+}

src/pkg/fmt/format.go

@@ -5,9 +5,7 @@
 package fmt
 
 import (
-	"bytes"
 	"strconv"
-	"unicode"
 	"unicode/utf8"
 )
 
@@ -36,10 +34,10 @@ func init() {
 }
 
 // A fmt is the raw formatter used by Printf etc.
-// It prints into a bytes.Buffer that must be set up externally.
+// It prints into a buffer that must be set up separately.
 type fmt struct {
 	intbuf [nByte]byte
-	buf    *bytes.Buffer
+	buf    *buffer
 	// width, precision
 	wid  int
 	prec int
@@ -69,7 +67,7 @@ func (f *fmt) clearflags() {
 	f.zero = false
 }
 
-func (f *fmt) init(buf *bytes.Buffer) {
+func (f *fmt) init(buf *buffer) {
 	f.buf = buf
 	f.clearflags()
 }
@@ -247,7 +245,7 @@ func (f *fmt) integer(a int64, base uint64, signedness bool, digits string) {
 	}
 
 	// If we want a quoted char for %#U, move the data up to make room.
-	if f.unicode && f.uniQuote && a >= 0 && a <= unicode.MaxRune && unicode.IsPrint(rune(a)) {
+	if f.unicode && f.uniQuote && a >= 0 && a <= utf8.MaxRune && strconv.IsPrint(rune(a)) {
 		runeWidth := utf8.RuneLen(rune(a))
 		width := 1 + 1 + runeWidth + 1 // space, quote, rune, quote
 		copy(buf[i-width:], buf[i:])   // guaranteed to have enough room.
@@ -290,16 +288,15 @@ func (f *fmt) fmt_s(s string) {
 // fmt_sx formats a string as a hexadecimal encoding of its bytes.
 func (f *fmt) fmt_sx(s, digits string) {
 	// TODO: Avoid buffer by pre-padding.
-	var b bytes.Buffer
+	var b []byte
 	for i := 0; i < len(s); i++ {
 		if i > 0 && f.space {
-			b.WriteByte(' ')
+			b = append(b, ' ')
 		}
 		v := s[i]
-		b.WriteByte(digits[v>>4])
-		b.WriteByte(digits[v&0xF])
+		b = append(b, digits[v>>4], digits[v&0xF])
 	}
-	f.pad(b.Bytes())
+	f.pad(b)
 }
 
 // fmt_q formats a string as a double-quoted, escaped Go string constant.

src/pkg/fmt/print.go

@@ -5,13 +5,11 @@
 package fmt
 
 import (
-	"bytes"
 	"errors"
 	"io"
 	"os"
 	"reflect"
 	"sync"
-	"unicode"
 	"unicode/utf8"
 )
 
@@ -71,11 +69,45 @@ type GoStringer interface {
 	GoString() string
 }
 
+// Use simple []byte instead of bytes.Buffer to avoid large dependency.
+type buffer []byte
+
+func (b *buffer) Write(p []byte) (n int, err error) {
+	*b = append(*b, p...)
+	return len(p), nil
+}
+
+func (b *buffer) WriteString(s string) (n int, err error) {
+	*b = append(*b, s...)
+	return len(s), nil
+}
+
+func (b *buffer) WriteByte(c byte) error {
+	*b = append(*b, c)
+	return nil
+}
+
+func (bp *buffer) WriteRune(r rune) error {
+	if r < utf8.RuneSelf {
+		*bp = append(*bp, byte(r))
+		return nil
+	}
+
+	b := *bp
+	n := len(b)
+	for n+utf8.UTFMax > cap(b) {
+		b = append(b, 0)
+	}
+	w := utf8.EncodeRune(b[n:n+utf8.UTFMax], r)
+	*bp = b[:n+w]
+	return nil
+}
+
 type pp struct {
 	n         int
 	panicking bool
 	erroring  bool // printing an error condition
-	buf       bytes.Buffer
+	buf       buffer
 	// field holds the current item, as an interface{}.
 	field interface{}
 	// value holds the current item, as a reflect.Value, and will be
@@ -133,10 +165,10 @@ func newPrinter() *pp {
 // Save used pp structs in ppFree; avoids an allocation per invocation.
 func (p *pp) free() {
 	// Don't hold on to pp structs with large buffers.
-	if cap(p.buf.Bytes()) > 1024 {
+	if cap(p.buf) > 1024 {
 		return
 	}
-	p.buf.Reset()
+	p.buf = p.buf[:0]
 	p.field = nil
 	p.value = reflect.Value{}
 	ppFree.put(p)
@@ -179,7 +211,7 @@ func (p *pp) Write(b []byte) (ret int, err error) {
 func Fprintf(w io.Writer, format string, a ...interface{}) (n int, err error) {
 	p := newPrinter()
 	p.doPrintf(format, a)
-	n64, err := p.buf.WriteTo(w)
+	n64, err := w.Write(p.buf)
 	p.free()
 	return int(n64), err
 }
@@ -194,7 +226,7 @@ func Printf(format string, a ...interface{}) (n int, err error) {
 func Sprintf(format string, a ...interface{}) string {
 	p := newPrinter()
 	p.doPrintf(format, a)
-	s := p.buf.String()
+	s := string(p.buf)
 	p.free()
 	return s
 }
@@ -213,7 +245,7 @@ func Errorf(format string, a ...interface{}) error {
 func Fprint(w io.Writer, a ...interface{}) (n int, err error) {
 	p := newPrinter()
 	p.doPrint(a, false, false)
-	n64, err := p.buf.WriteTo(w)
+	n64, err := w.Write(p.buf)
 	p.free()
 	return int(n64), err
 }
@@ -230,7 +262,7 @@ func Print(a ...interface{}) (n int, err error) {
 func Sprint(a ...interface{}) string {
 	p := newPrinter()
 	p.doPrint(a, false, false)
-	s := p.buf.String()
+	s := string(p.buf)
 	p.free()
 	return s
 }
@@ -245,7 +277,7 @@ func Sprint(a ...interface{}) string {
 func Fprintln(w io.Writer, a ...interface{}) (n int, err error) {
 	p := newPrinter()
 	p.doPrint(a, true, true)
-	n64, err := p.buf.WriteTo(w)
+	n64, err := w.Write(p.buf)
 	p.free()
 	return int(n64), err
 }
@@ -262,7 +294,7 @@ func Println(a ...interface{}) (n int, err error) {
 func Sprintln(a ...interface{}) string {
 	p := newPrinter()
 	p.doPrint(a, true, true)
-	s := p.buf.String()
+	s := string(p.buf)
 	p.free()
 	return s
 }
@@ -352,7 +384,7 @@ func (p *pp) fmtInt64(v int64, verb rune) {
 	case 'o':
 		p.fmt.integer(v, 8, signed, ldigits)
 	case 'q':
-		if 0 <= v && v <= unicode.MaxRune {
+		if 0 <= v && v <= utf8.MaxRune {
 			p.fmt.fmt_qc(v)
 		} else {
 			p.badVerb(verb)
@@ -416,7 +448,7 @@ func (p *pp) fmtUint64(v uint64, verb rune, goSyntax bool) {
 	case 'o':
 		p.fmt.integer(int64(v), 8, unsigned, ldigits)
 	case 'q':
-		if 0 <= v && v <= unicode.MaxRune {
+		if 0 <= v && v <= utf8.MaxRune {
 			p.fmt.fmt_qc(int64(v))
 		} else {
 			p.badVerb(verb)

src/pkg/fmt/scan.go

@@ -5,15 +5,12 @@
 package fmt
 
 import (
-	"bytes"
 	"errors"
 	"io"
 	"math"
 	"os"
 	"reflect"
 	"strconv"
-	"strings"
-	"unicode"
 	"unicode/utf8"
 )
 
@@ -87,25 +84,36 @@ func Scanf(format string, a ...interface{}) (n int, err error) {
 	return Fscanf(os.Stdin, format, a...)
 }
 
+type stringReader string
+
+func (r *stringReader) Read(b []byte) (n int, err error) {
+	n = copy(b, *r)
+	*r = (*r)[n:]
+	if n == 0 {
+		err = io.EOF
+	}
+	return
+}
+
 // Sscan scans the argument string, storing successive space-separated
 // values into successive arguments.  Newlines count as space.  It
 // returns the number of items successfully scanned.  If that is less
 // than the number of arguments, err will report why.
 func Sscan(str string, a ...interface{}) (n int, err error) {
-	return Fscan(strings.NewReader(str), a...)
+	return Fscan((*stringReader)(&str), a...)
 }
 
 // Sscanln is similar to Sscan, but stops scanning at a newline and
 // after the final item there must be a newline or EOF.
 func Sscanln(str string, a ...interface{}) (n int, err error) {
-	return Fscanln(strings.NewReader(str), a...)
+	return Fscanln((*stringReader)(&str), a...)
 }
 
 // Sscanf scans the argument string, storing successive space-separated
 // values into successive arguments as determined by the format.  It
 // returns the number of items successfully parsed.
 func Sscanf(str string, format string, a ...interface{}) (n int, err error) {
-	return Fscanf(strings.NewReader(str), format, a...)
+	return Fscanf((*stringReader)(&str), format, a...)
 }
 
 // Fscan scans text read from r, storing successive space-separated
@@ -149,7 +157,7 @@ const eof = -1
 // ss is the internal implementation of ScanState.
 type ss struct {
 	rr       io.RuneReader // where to read input
-	buf      bytes.Buffer  // token accumulator
+	buf      buffer        // token accumulator
 	peekRune rune          // one-rune lookahead
 	prevRune rune          // last rune returned by ReadRune
 	count    int           // runes consumed so far.
@@ -262,14 +270,46 @@ func (s *ss) Token(skipSpace bool, f func(rune) bool) (tok []byte, err error) {
 	if f == nil {
 		f = notSpace
 	}
-	s.buf.Reset()
+	s.buf = s.buf[:0]
 	tok = s.token(skipSpace, f)
 	return
 }
 
+// space is a copy of the unicode.White_Space ranges,
+// to avoid depending on package unicode.
+var space = [][2]uint16{
+	{0x0009, 0x000d},
+	{0x0020, 0x0020},
+	{0x0085, 0x0085},
+	{0x00a0, 0x00a0},
+	{0x1680, 0x1680},
+	{0x180e, 0x180e},
+	{0x2000, 0x200a},
+	{0x2028, 0x2029},
+	{0x202f, 0x202f},
+	{0x205f, 0x205f},
+	{0x3000, 0x3000},
+}
+
+func isSpace(r rune) bool {
+	if r >= 1<<16 {
+		return false
+	}
+	rx := uint16(r)
+	for _, rng := range space {
+		if rx < rng[0] {
+			return false
+		}
+		if rx <= rng[1] {
+			return true
+		}
+	}
+	return false
+}
+
 // notSpace is the default scanning function used in Token.
 func notSpace(r rune) bool {
-	return !unicode.IsSpace(r)
+	return !isSpace(r)
 }
 
 // skipSpace provides Scan() methods the ability to skip space and newline characters 
@@ -378,10 +418,10 @@ func (s *ss) free(old ssave) {
 		return
 	}
 	// Don't hold on to ss structs with large buffers.
-	if cap(s.buf.Bytes()) > 1024 {
+	if cap(s.buf) > 1024 {
 		return
 	}
-	s.buf.Reset()
+	s.buf = s.buf[:0]
 	s.rr = nil
 	ssFree.put(s)
 }
@@ -403,7 +443,7 @@ func (s *ss) skipSpace(stopAtNewline bool) {
 			s.errorString("unexpected newline")
 			return
 		}
-		if !unicode.IsSpace(r) {
+		if !isSpace(r) {
 			s.UnreadRune()
 			break
 		}
@@ -429,7 +469,7 @@ func (s *ss) token(skipSpace bool, f func(rune) bool) []byte {
 		}
 		s.buf.WriteRune(r)
 	}
-	return s.buf.Bytes()
+	return s.buf
 }
 
 // typeError indicates that the type of the operand did not match the format
@@ -440,6 +480,15 @@ func (s *ss) typeError(field interface{}, expected string) {
 var complexError = errors.New("syntax error scanning complex number")
 var boolError = errors.New("syntax error scanning boolean")
 
+func indexRune(s string, r rune) int {
+	for i, c := range s {
+		if c == r {
+			return i
+		}
+	}
+	return -1
+}
+
 // consume reads the next rune in the input and reports whether it is in the ok string.
 // If accept is true, it puts the character into the input token.
 func (s *ss) consume(ok string, accept bool) bool {
@@ -447,7 +496,7 @@ func (s *ss) consume(ok string, accept bool) bool {
 	if r == eof {
 		return false
 	}
-	if strings.IndexRune(ok, r) >= 0 {
+	if indexRune(ok, r) >= 0 {
 		if accept {
 			s.buf.WriteRune(r)
 		}
@@ -465,7 +514,7 @@ func (s *ss) peek(ok string) bool {
 	if r != eof {
 		s.UnreadRune()
 	}
-	return strings.IndexRune(ok, r) >= 0
+	return indexRune(ok, r) >= 0
 }
 
 func (s *ss) notEOF() {
@@ -560,7 +609,7 @@ func (s *ss) scanNumber(digits string, haveDigits bool) string {
 	}
 	for s.accept(digits) {
 	}
-	return s.buf.String()
+	return string(s.buf)
 }
 
 // scanRune returns the next rune value in the input.
@@ -660,16 +709,16 @@ func (s *ss) scanUint(verb rune, bitSize int) uint64 {
 // if the width is specified. It's not rigorous about syntax because it doesn't check that
 // we have at least some digits, but Atof will do that.
 func (s *ss) floatToken() string {
-	s.buf.Reset()
+	s.buf = s.buf[:0]
 	// NaN?
 	if s.accept("nN") && s.accept("aA") && s.accept("nN") {
-		return s.buf.String()
+		return string(s.buf)
 	}
 	// leading sign?
 	s.accept(sign)
 	// Inf?
 	if s.accept("iI") && s.accept("nN") && s.accept("fF") {
-		return s.buf.String()
+		return string(s.buf)
 	}
 	// digits?
 	for s.accept(decimalDigits) {
@@ -688,7 +737,7 @@ func (s *ss) floatToken() string {
 		for s.accept(decimalDigits) {
 		}
 	}
-	return s.buf.String()
+	return string(s.buf)
 }
 
 // complexTokens returns the real and imaginary parts of the complex number starting here.
@@ -698,13 +747,13 @@ func (s *ss) complexTokens() (real, imag string) {
 	// TODO: accept N and Ni independently?
 	parens := s.accept("(")
 	real = s.floatToken()
-	s.buf.Reset()
+	s.buf = s.buf[:0]
 	// Must now have a sign.
 	if !s.accept("+-") {
 		s.error(complexError)
 	}
 	// Sign is now in buffer
-	imagSign := s.buf.String()
+	imagSign := string(s.buf)
 	imag = s.floatToken()
 	if !s.accept("i") {
 		s.error(complexError)
@@ -717,7 +766,7 @@ func (s *ss) complexTokens() (real, imag string) {
 
 // convertFloat converts the string to a float64value.
 func (s *ss) convertFloat(str string, n int) float64 {
-	if p := strings.Index(str, "p"); p >= 0 {
+	if p := indexRune(str, 'p'); p >= 0 {
 		// Atof doesn't handle power-of-2 exponents,
 		// but they're easy to evaluate.
 		f, err := strconv.ParseFloat(str[:p], n)
@@ -794,7 +843,7 @@ func (s *ss) quotedString() string {
 			}
 			s.buf.WriteRune(r)
 		}
-		return s.buf.String()
+		return string(s.buf)
 	case '"':
 		// Double-quoted: Include the quotes and let strconv.Unquote do the backslash escapes.
 		s.buf.WriteRune(quote)
@@ -811,7 +860,7 @@ func (s *ss) quotedString() string {
 				break
 			}
 		}
-		result, err := strconv.Unquote(s.buf.String())
+		result, err := strconv.Unquote(string(s.buf))
 		if err != nil {
 			s.error(err)
 		}
@@ -844,7 +893,7 @@ func (s *ss) hexByte() (b byte, ok bool) {
 	if rune1 == eof {
 		return
 	}
-	if unicode.IsSpace(rune1) {
+	if isSpace(rune1) {
 		s.UnreadRune()
 		return
 	}
@@ -862,11 +911,11 @@ func (s *ss) hexString() string {
 		}
 		s.buf.WriteByte(b)
 	}
-	if s.buf.Len() == 0 {
+	if len(s.buf) == 0 {
 		s.errorString("Scan: no hex data for %x string")
 		return ""
 	}
-	return s.buf.String()
+	return string(s.buf)
 }
 
 const floatVerbs = "beEfFgGv"
@@ -875,7 +924,7 @@ const hugeWid = 1 << 30
 
 // scanOne scans a single value, deriving the scanner from the type of the argument.
 func (s *ss) scanOne(verb rune, field interface{}) {
-	s.buf.Reset()
+	s.buf = s.buf[:0]
 	var err error
 	// If the parameter has its own Scan method, use that.
 	if v, ok := field.(Scanner); ok {
@@ -1004,7 +1053,7 @@ func (s *ss) doScan(a []interface{}) (numProcessed int, err error) {
 			if r == '\n' || r == eof {
 				break
 			}
-			if !unicode.IsSpace(r) {
+			if !isSpace(r) {
 				s.errorString("Scan: expected newline")
 				break
 			}
@@ -1032,7 +1081,7 @@ func (s *ss) advance(format string) (i int) {
 			i += w // skip the first %
 		}
 		sawSpace := false
-		for unicode.IsSpace(fmtc) && i < len(format) {
+		for isSpace(fmtc) && i < len(format) {
 			sawSpace = true
 			i += w
 			fmtc, w = utf8.DecodeRuneInString(format[i:])
@@ -1044,7 +1093,7 @@ func (s *ss) advance(format string) (i int) {
 			if inputc == eof {
 				return
 			}
-			if !unicode.IsSpace(inputc) {
+			if !isSpace(inputc) {
 				// Space in format but not in input: error
 				s.errorString("expected space in input to match format")
 			}

src/pkg/log/log.go

@@ -13,7 +13,6 @@
 package log
 
 import (
-	"bytes"
 	"fmt"
 	"io"
 	"os"
@@ -41,11 +40,11 @@ const (
 // the Writer's Write method.  A Logger can be used simultaneously from
 // multiple goroutines; it guarantees to serialize access to the Writer.
 type Logger struct {
-	mu     sync.Mutex   // ensures atomic writes; protects the following fields
-	prefix string       // prefix to write at beginning of each line
-	flag   int          // properties
-	out    io.Writer    // destination for output
-	buf    bytes.Buffer // for accumulating text to write
+	mu     sync.Mutex // ensures atomic writes; protects the following fields
+	prefix string     // prefix to write at beginning of each line
+	flag   int        // properties
+	out    io.Writer  // destination for output
+	buf    []byte     // for accumulating text to write
 }
 
 // New creates a new Logger.   The out variable sets the
@@ -60,10 +59,10 @@ var std = New(os.Stderr, "", LstdFlags)
 
 // Cheap integer to fixed-width decimal ASCII.  Give a negative width to avoid zero-padding.
 // Knows the buffer has capacity.
-func itoa(buf *bytes.Buffer, i int, wid int) {
+func itoa(buf *[]byte, i int, wid int) {
 	var u uint = uint(i)
 	if u == 0 && wid <= 1 {
-		buf.WriteByte('0')
+		*buf = append(*buf, '0')
 		return
 	}
 
@@ -75,38 +74,33 @@ func itoa(buf *bytes.Buffer, i int, wid int) {
 		wid--
 		b[bp] = byte(u%10) + '0'
 	}
-
-	// avoid slicing b to avoid an allocation.
-	for bp < len(b) {
-		buf.WriteByte(b[bp])
-		bp++
-	}
+	*buf = append(*buf, b[bp:]...)
 }
 
-func (l *Logger) formatHeader(buf *bytes.Buffer, t time.Time, file string, line int) {
-	buf.WriteString(l.prefix)
+func (l *Logger) formatHeader(buf *[]byte, t time.Time, file string, line int) {
+	*buf = append(*buf, l.prefix...)
 	if l.flag&(Ldate|Ltime|Lmicroseconds) != 0 {
 		if l.flag&Ldate != 0 {
 			year, month, day := t.Date()
 			itoa(buf, year, 4)
-			buf.WriteByte('/')
+			*buf = append(*buf, '/')
 			itoa(buf, int(month), 2)
-			buf.WriteByte('/')
+			*buf = append(*buf, '/')
 			itoa(buf, day, 2)
-			buf.WriteByte(' ')
+			*buf = append(*buf, ' ')
 		}
 		if l.flag&(Ltime|Lmicroseconds) != 0 {
 			hour, min, sec := t.Clock()
 			itoa(buf, hour, 2)
-			buf.WriteByte(':')
+			*buf = append(*buf, ':')
 			itoa(buf, min, 2)
-			buf.WriteByte(':')
+			*buf = append(*buf, ':')
 			itoa(buf, sec, 2)
 			if l.flag&Lmicroseconds != 0 {
-				buf.WriteByte('.')
+				*buf = append(*buf, '.')
 				itoa(buf, t.Nanosecond()/1e3, 6)
 			}
-			buf.WriteByte(' ')
+			*buf = append(*buf, ' ')
 		}
 	}
 	if l.flag&(Lshortfile|Llongfile) != 0 {
@@ -120,10 +114,10 @@ func (l *Logger) formatHeader(buf *bytes.Buffer, t time.Time, file string, line
 			}
 			file = short
 		}
-		buf.WriteString(file)
-		buf.WriteByte(':')
+		*buf = append(*buf, file...)
+		*buf = append(*buf, ':')
 		itoa(buf, line, -1)
-		buf.WriteString(": ")
+		*buf = append(*buf, ": "...)
 	}
 }
 
@@ -150,13 +144,13 @@ func (l *Logger) Output(calldepth int, s string) error {
 		}
 		l.mu.Lock()
 	}
-	l.buf.Reset()
+	l.buf = l.buf[:0]
 	l.formatHeader(&l.buf, now, file, line)
-	l.buf.WriteString(s)
+	l.buf = append(l.buf, s...)
 	if len(s) > 0 && s[len(s)-1] != '\n' {
-		l.buf.WriteByte('\n')
+		l.buf = append(l.buf, '\n')
 	}
-	_, err := l.out.Write(l.buf.Bytes())
+	_, err := l.out.Write(l.buf)
 	return err
 }