...
Run Format

Source file src/fmt/scan.go

     1	// Copyright 2010 The Go Authors. All rights reserved.
     2	// Use of this source code is governed by a BSD-style
     3	// license that can be found in the LICENSE file.
     4	
     5	package fmt
     6	
     7	import (
     8		"errors"
     9		"io"
    10		"math"
    11		"os"
    12		"reflect"
    13		"strconv"
    14		"sync"
    15		"unicode/utf8"
    16	)
    17	
    18	// runeUnreader is the interface to something that can unread runes.
    19	// If the object provided to Scan does not satisfy this interface,
    20	// a local buffer will be used to back up the input, but its contents
    21	// will be lost when Scan returns.
    22	type runeUnreader interface {
    23		UnreadRune() error
    24	}
    25	
    26	// ScanState represents the scanner state passed to custom scanners.
    27	// Scanners may do rune-at-a-time scanning or ask the ScanState
    28	// to discover the next space-delimited token.
    29	type ScanState interface {
    30		// ReadRune reads the next rune (Unicode code point) from the input.
    31		// If invoked during Scanln, Fscanln, or Sscanln, ReadRune() will
    32		// return EOF after returning the first '\n' or when reading beyond
    33		// the specified width.
    34		ReadRune() (r rune, size int, err error)
    35		// UnreadRune causes the next call to ReadRune to return the same rune.
    36		UnreadRune() error
    37		// SkipSpace skips space in the input. Newlines are treated as space
    38		// unless the scan operation is Scanln, Fscanln or Sscanln, in which case
    39		// a newline is treated as EOF.
    40		SkipSpace()
    41		// Token skips space in the input if skipSpace is true, then returns the
    42		// run of Unicode code points c satisfying f(c).  If f is nil,
    43		// !unicode.IsSpace(c) is used; that is, the token will hold non-space
    44		// characters.  Newlines are treated as space unless the scan operation
    45		// is Scanln, Fscanln or Sscanln, in which case a newline is treated as
    46		// EOF.  The returned slice points to shared data that may be overwritten
    47		// by the next call to Token, a call to a Scan function using the ScanState
    48		// as input, or when the calling Scan method returns.
    49		Token(skipSpace bool, f func(rune) bool) (token []byte, err error)
    50		// Width returns the value of the width option and whether it has been set.
    51		// The unit is Unicode code points.
    52		Width() (wid int, ok bool)
    53		// Because ReadRune is implemented by the interface, Read should never be
    54		// called by the scanning routines and a valid implementation of
    55		// ScanState may choose always to return an error from Read.
    56		Read(buf []byte) (n int, err error)
    57	}
    58	
    59	// Scanner is implemented by any value that has a Scan method, which scans
    60	// the input for the representation of a value and stores the result in the
    61	// receiver, which must be a pointer to be useful.  The Scan method is called
    62	// for any argument to Scan, Scanf, or Scanln that implements it.
    63	type Scanner interface {
    64		Scan(state ScanState, verb rune) error
    65	}
    66	
    67	// Scan scans text read from standard input, storing successive
    68	// space-separated values into successive arguments.  Newlines count
    69	// as space.  It returns the number of items successfully scanned.
    70	// If that is less than the number of arguments, err will report why.
    71	func Scan(a ...interface{}) (n int, err error) {
    72		return Fscan(os.Stdin, a...)
    73	}
    74	
    75	// Scanln is similar to Scan, but stops scanning at a newline and
    76	// after the final item there must be a newline or EOF.
    77	func Scanln(a ...interface{}) (n int, err error) {
    78		return Fscanln(os.Stdin, a...)
    79	}
    80	
    81	// Scanf scans text read from standard input, storing successive
    82	// space-separated values into successive arguments as determined by
    83	// the format.  It returns the number of items successfully scanned.
    84	func Scanf(format string, a ...interface{}) (n int, err error) {
    85		return Fscanf(os.Stdin, format, a...)
    86	}
    87	
    88	type stringReader string
    89	
    90	func (r *stringReader) Read(b []byte) (n int, err error) {
    91		n = copy(b, *r)
    92		*r = (*r)[n:]
    93		if n == 0 {
    94			err = io.EOF
    95		}
    96		return
    97	}
    98	
    99	// Sscan scans the argument string, storing successive space-separated
   100	// values into successive arguments.  Newlines count as space.  It
   101	// returns the number of items successfully scanned.  If that is less
   102	// than the number of arguments, err will report why.
   103	func Sscan(str string, a ...interface{}) (n int, err error) {
   104		return Fscan((*stringReader)(&str), a...)
   105	}
   106	
   107	// Sscanln is similar to Sscan, but stops scanning at a newline and
   108	// after the final item there must be a newline or EOF.
   109	func Sscanln(str string, a ...interface{}) (n int, err error) {
   110		return Fscanln((*stringReader)(&str), a...)
   111	}
   112	
   113	// Sscanf scans the argument string, storing successive space-separated
   114	// values into successive arguments as determined by the format.  It
   115	// returns the number of items successfully parsed.
   116	func Sscanf(str string, format string, a ...interface{}) (n int, err error) {
   117		return Fscanf((*stringReader)(&str), format, a...)
   118	}
   119	
   120	// Fscan scans text read from r, storing successive space-separated
   121	// values into successive arguments.  Newlines count as space.  It
   122	// returns the number of items successfully scanned.  If that is less
   123	// than the number of arguments, err will report why.
   124	func Fscan(r io.Reader, a ...interface{}) (n int, err error) {
   125		s, old := newScanState(r, true, false)
   126		n, err = s.doScan(a)
   127		s.free(old)
   128		return
   129	}
   130	
   131	// Fscanln is similar to Fscan, but stops scanning at a newline and
   132	// after the final item there must be a newline or EOF.
   133	func Fscanln(r io.Reader, a ...interface{}) (n int, err error) {
   134		s, old := newScanState(r, false, true)
   135		n, err = s.doScan(a)
   136		s.free(old)
   137		return
   138	}
   139	
   140	// Fscanf scans text read from r, storing successive space-separated
   141	// values into successive arguments as determined by the format.  It
   142	// returns the number of items successfully parsed.
   143	func Fscanf(r io.Reader, format string, a ...interface{}) (n int, err error) {
   144		s, old := newScanState(r, false, false)
   145		n, err = s.doScanf(format, a)
   146		s.free(old)
   147		return
   148	}
   149	
   150	// scanError represents an error generated by the scanning software.
   151	// It's used as a unique signature to identify such errors when recovering.
   152	type scanError struct {
   153		err error
   154	}
   155	
   156	const eof = -1
   157	
   158	// ss is the internal implementation of ScanState.
   159	type ss struct {
   160		rr       io.RuneReader // where to read input
   161		buf      buffer        // token accumulator
   162		peekRune rune          // one-rune lookahead
   163		prevRune rune          // last rune returned by ReadRune
   164		count    int           // runes consumed so far.
   165		atEOF    bool          // already read EOF
   166		ssave
   167	}
   168	
   169	// ssave holds the parts of ss that need to be
   170	// saved and restored on recursive scans.
   171	type ssave struct {
   172		validSave bool // is or was a part of an actual ss.
   173		nlIsEnd   bool // whether newline terminates scan
   174		nlIsSpace bool // whether newline counts as white space
   175		argLimit  int  // max value of ss.count for this arg; argLimit <= limit
   176		limit     int  // max value of ss.count.
   177		maxWid    int  // width of this arg.
   178	}
   179	
   180	// The Read method is only in ScanState so that ScanState
   181	// satisfies io.Reader. It will never be called when used as
   182	// intended, so there is no need to make it actually work.
   183	func (s *ss) Read(buf []byte) (n int, err error) {
   184		return 0, errors.New("ScanState's Read should not be called. Use ReadRune")
   185	}
   186	
   187	func (s *ss) ReadRune() (r rune, size int, err error) {
   188		if s.peekRune >= 0 {
   189			s.count++
   190			r = s.peekRune
   191			size = utf8.RuneLen(r)
   192			s.prevRune = r
   193			s.peekRune = -1
   194			return
   195		}
   196		if s.atEOF || s.nlIsEnd && s.prevRune == '\n' || s.count >= s.argLimit {
   197			err = io.EOF
   198			return
   199		}
   200	
   201		r, size, err = s.rr.ReadRune()
   202		if err == nil {
   203			s.count++
   204			s.prevRune = r
   205		} else if err == io.EOF {
   206			s.atEOF = true
   207		}
   208		return
   209	}
   210	
   211	func (s *ss) Width() (wid int, ok bool) {
   212		if s.maxWid == hugeWid {
   213			return 0, false
   214		}
   215		return s.maxWid, true
   216	}
   217	
   218	// The public method returns an error; this private one panics.
   219	// If getRune reaches EOF, the return value is EOF (-1).
   220	func (s *ss) getRune() (r rune) {
   221		r, _, err := s.ReadRune()
   222		if err != nil {
   223			if err == io.EOF {
   224				return eof
   225			}
   226			s.error(err)
   227		}
   228		return
   229	}
   230	
   231	// mustReadRune turns io.EOF into a panic(io.ErrUnexpectedEOF).
   232	// It is called in cases such as string scanning where an EOF is a
   233	// syntax error.
   234	func (s *ss) mustReadRune() (r rune) {
   235		r = s.getRune()
   236		if r == eof {
   237			s.error(io.ErrUnexpectedEOF)
   238		}
   239		return
   240	}
   241	
   242	func (s *ss) UnreadRune() error {
   243		if u, ok := s.rr.(runeUnreader); ok {
   244			u.UnreadRune()
   245		} else {
   246			s.peekRune = s.prevRune
   247		}
   248		s.prevRune = -1
   249		s.count--
   250		return nil
   251	}
   252	
   253	func (s *ss) error(err error) {
   254		panic(scanError{err})
   255	}
   256	
   257	func (s *ss) errorString(err string) {
   258		panic(scanError{errors.New(err)})
   259	}
   260	
   261	func (s *ss) Token(skipSpace bool, f func(rune) bool) (tok []byte, err error) {
   262		defer func() {
   263			if e := recover(); e != nil {
   264				if se, ok := e.(scanError); ok {
   265					err = se.err
   266				} else {
   267					panic(e)
   268				}
   269			}
   270		}()
   271		if f == nil {
   272			f = notSpace
   273		}
   274		s.buf = s.buf[:0]
   275		tok = s.token(skipSpace, f)
   276		return
   277	}
   278	
   279	// space is a copy of the unicode.White_Space ranges,
   280	// to avoid depending on package unicode.
   281	var space = [][2]uint16{
   282		{0x0009, 0x000d},
   283		{0x0020, 0x0020},
   284		{0x0085, 0x0085},
   285		{0x00a0, 0x00a0},
   286		{0x1680, 0x1680},
   287		{0x2000, 0x200a},
   288		{0x2028, 0x2029},
   289		{0x202f, 0x202f},
   290		{0x205f, 0x205f},
   291		{0x3000, 0x3000},
   292	}
   293	
   294	func isSpace(r rune) bool {
   295		if r >= 1<<16 {
   296			return false
   297		}
   298		rx := uint16(r)
   299		for _, rng := range space {
   300			if rx < rng[0] {
   301				return false
   302			}
   303			if rx <= rng[1] {
   304				return true
   305			}
   306		}
   307		return false
   308	}
   309	
   310	// notSpace is the default scanning function used in Token.
   311	func notSpace(r rune) bool {
   312		return !isSpace(r)
   313	}
   314	
   315	// SkipSpace provides Scan methods the ability to skip space and newline
   316	// characters in keeping with the current scanning mode set by format strings
   317	// and Scan/Scanln.
   318	func (s *ss) SkipSpace() {
   319		s.skipSpace(false)
   320	}
   321	
   322	// readRune is a structure to enable reading UTF-8 encoded code points
   323	// from an io.Reader.  It is used if the Reader given to the scanner does
   324	// not already implement io.RuneReader.
   325	type readRune struct {
   326		reader  io.Reader
   327		buf     [utf8.UTFMax]byte // used only inside ReadRune
   328		pending int               // number of bytes in pendBuf; only >0 for bad UTF-8
   329		pendBuf [utf8.UTFMax]byte // bytes left over
   330	}
   331	
   332	// readByte returns the next byte from the input, which may be
   333	// left over from a previous read if the UTF-8 was ill-formed.
   334	func (r *readRune) readByte() (b byte, err error) {
   335		if r.pending > 0 {
   336			b = r.pendBuf[0]
   337			copy(r.pendBuf[0:], r.pendBuf[1:])
   338			r.pending--
   339			return
   340		}
   341		n, err := io.ReadFull(r.reader, r.pendBuf[0:1])
   342		if n != 1 {
   343			return 0, err
   344		}
   345		return r.pendBuf[0], err
   346	}
   347	
   348	// unread saves the bytes for the next read.
   349	func (r *readRune) unread(buf []byte) {
   350		copy(r.pendBuf[r.pending:], buf)
   351		r.pending += len(buf)
   352	}
   353	
   354	// ReadRune returns the next UTF-8 encoded code point from the
   355	// io.Reader inside r.
   356	func (r *readRune) ReadRune() (rr rune, size int, err error) {
   357		r.buf[0], err = r.readByte()
   358		if err != nil {
   359			return 0, 0, err
   360		}
   361		if r.buf[0] < utf8.RuneSelf { // fast check for common ASCII case
   362			rr = rune(r.buf[0])
   363			size = 1 // Known to be 1.
   364			return
   365		}
   366		var n int
   367		for n = 1; !utf8.FullRune(r.buf[0:n]); n++ {
   368			r.buf[n], err = r.readByte()
   369			if err != nil {
   370				if err == io.EOF {
   371					err = nil
   372					break
   373				}
   374				return
   375			}
   376		}
   377		rr, size = utf8.DecodeRune(r.buf[0:n])
   378		if size < n { // an error
   379			r.unread(r.buf[size:n])
   380		}
   381		return
   382	}
   383	
   384	var ssFree = sync.Pool{
   385		New: func() interface{} { return new(ss) },
   386	}
   387	
   388	// newScanState allocates a new ss struct or grab a cached one.
   389	func newScanState(r io.Reader, nlIsSpace, nlIsEnd bool) (s *ss, old ssave) {
   390		// If the reader is a *ss, then we've got a recursive
   391		// call to Scan, so re-use the scan state.
   392		s, ok := r.(*ss)
   393		if ok {
   394			old = s.ssave
   395			s.limit = s.argLimit
   396			s.nlIsEnd = nlIsEnd || s.nlIsEnd
   397			s.nlIsSpace = nlIsSpace
   398			return
   399		}
   400	
   401		s = ssFree.Get().(*ss)
   402		if rr, ok := r.(io.RuneReader); ok {
   403			s.rr = rr
   404		} else {
   405			s.rr = &readRune{reader: r}
   406		}
   407		s.nlIsSpace = nlIsSpace
   408		s.nlIsEnd = nlIsEnd
   409		s.prevRune = -1
   410		s.peekRune = -1
   411		s.atEOF = false
   412		s.limit = hugeWid
   413		s.argLimit = hugeWid
   414		s.maxWid = hugeWid
   415		s.validSave = true
   416		s.count = 0
   417		return
   418	}
   419	
   420	// free saves used ss structs in ssFree; avoid an allocation per invocation.
   421	func (s *ss) free(old ssave) {
   422		// If it was used recursively, just restore the old state.
   423		if old.validSave {
   424			s.ssave = old
   425			return
   426		}
   427		// Don't hold on to ss structs with large buffers.
   428		if cap(s.buf) > 1024 {
   429			return
   430		}
   431		s.buf = s.buf[:0]
   432		s.rr = nil
   433		ssFree.Put(s)
   434	}
   435	
   436	// skipSpace skips spaces and maybe newlines.
   437	func (s *ss) skipSpace(stopAtNewline bool) {
   438		for {
   439			r := s.getRune()
   440			if r == eof {
   441				return
   442			}
   443			if r == '\r' && s.peek("\n") {
   444				continue
   445			}
   446			if r == '\n' {
   447				if stopAtNewline {
   448					break
   449				}
   450				if s.nlIsSpace {
   451					continue
   452				}
   453				s.errorString("unexpected newline")
   454				return
   455			}
   456			if !isSpace(r) {
   457				s.UnreadRune()
   458				break
   459			}
   460		}
   461	}
   462	
   463	// token returns the next space-delimited string from the input.  It
   464	// skips white space.  For Scanln, it stops at newlines.  For Scan,
   465	// newlines are treated as spaces.
   466	func (s *ss) token(skipSpace bool, f func(rune) bool) []byte {
   467		if skipSpace {
   468			s.skipSpace(false)
   469		}
   470		// read until white space or newline
   471		for {
   472			r := s.getRune()
   473			if r == eof {
   474				break
   475			}
   476			if !f(r) {
   477				s.UnreadRune()
   478				break
   479			}
   480			s.buf.WriteRune(r)
   481		}
   482		return s.buf
   483	}
   484	
   485	var complexError = errors.New("syntax error scanning complex number")
   486	var boolError = errors.New("syntax error scanning boolean")
   487	
   488	func indexRune(s string, r rune) int {
   489		for i, c := range s {
   490			if c == r {
   491				return i
   492			}
   493		}
   494		return -1
   495	}
   496	
   497	// consume reads the next rune in the input and reports whether it is in the ok string.
   498	// If accept is true, it puts the character into the input token.
   499	func (s *ss) consume(ok string, accept bool) bool {
   500		r := s.getRune()
   501		if r == eof {
   502			return false
   503		}
   504		if indexRune(ok, r) >= 0 {
   505			if accept {
   506				s.buf.WriteRune(r)
   507			}
   508			return true
   509		}
   510		if r != eof && accept {
   511			s.UnreadRune()
   512		}
   513		return false
   514	}
   515	
   516	// peek reports whether the next character is in the ok string, without consuming it.
   517	func (s *ss) peek(ok string) bool {
   518		r := s.getRune()
   519		if r != eof {
   520			s.UnreadRune()
   521		}
   522		return indexRune(ok, r) >= 0
   523	}
   524	
   525	func (s *ss) notEOF() {
   526		// Guarantee there is data to be read.
   527		if r := s.getRune(); r == eof {
   528			panic(io.EOF)
   529		}
   530		s.UnreadRune()
   531	}
   532	
   533	// accept checks the next rune in the input.  If it's a byte (sic) in the string, it puts it in the
   534	// buffer and returns true. Otherwise it return false.
   535	func (s *ss) accept(ok string) bool {
   536		return s.consume(ok, true)
   537	}
   538	
   539	// okVerb verifies that the verb is present in the list, setting s.err appropriately if not.
   540	func (s *ss) okVerb(verb rune, okVerbs, typ string) bool {
   541		for _, v := range okVerbs {
   542			if v == verb {
   543				return true
   544			}
   545		}
   546		s.errorString("bad verb %" + string(verb) + " for " + typ)
   547		return false
   548	}
   549	
   550	// scanBool returns the value of the boolean represented by the next token.
   551	func (s *ss) scanBool(verb rune) bool {
   552		s.skipSpace(false)
   553		s.notEOF()
   554		if !s.okVerb(verb, "tv", "boolean") {
   555			return false
   556		}
   557		// Syntax-checking a boolean is annoying.  We're not fastidious about case.
   558		switch s.getRune() {
   559		case '0':
   560			return false
   561		case '1':
   562			return true
   563		case 't', 'T':
   564			if s.accept("rR") && (!s.accept("uU") || !s.accept("eE")) {
   565				s.error(boolError)
   566			}
   567			return true
   568		case 'f', 'F':
   569			if s.accept("aA") && (!s.accept("lL") || !s.accept("sS") || !s.accept("eE")) {
   570				s.error(boolError)
   571			}
   572			return false
   573		}
   574		return false
   575	}
   576	
   577	// Numerical elements
   578	const (
   579		binaryDigits      = "01"
   580		octalDigits       = "01234567"
   581		decimalDigits     = "0123456789"
   582		hexadecimalDigits = "0123456789aAbBcCdDeEfF"
   583		sign              = "+-"
   584		period            = "."
   585		exponent          = "eEp"
   586	)
   587	
   588	// getBase returns the numeric base represented by the verb and its digit string.
   589	func (s *ss) getBase(verb rune) (base int, digits string) {
   590		s.okVerb(verb, "bdoUxXv", "integer") // sets s.err
   591		base = 10
   592		digits = decimalDigits
   593		switch verb {
   594		case 'b':
   595			base = 2
   596			digits = binaryDigits
   597		case 'o':
   598			base = 8
   599			digits = octalDigits
   600		case 'x', 'X', 'U':
   601			base = 16
   602			digits = hexadecimalDigits
   603		}
   604		return
   605	}
   606	
   607	// scanNumber returns the numerical string with specified digits starting here.
   608	func (s *ss) scanNumber(digits string, haveDigits bool) string {
   609		if !haveDigits {
   610			s.notEOF()
   611			if !s.accept(digits) {
   612				s.errorString("expected integer")
   613			}
   614		}
   615		for s.accept(digits) {
   616		}
   617		return string(s.buf)
   618	}
   619	
   620	// scanRune returns the next rune value in the input.
   621	func (s *ss) scanRune(bitSize int) int64 {
   622		s.notEOF()
   623		r := int64(s.getRune())
   624		n := uint(bitSize)
   625		x := (r << (64 - n)) >> (64 - n)
   626		if x != r {
   627			s.errorString("overflow on character value " + string(r))
   628		}
   629		return r
   630	}
   631	
   632	// scanBasePrefix reports whether the integer begins with a 0 or 0x,
   633	// and returns the base, digit string, and whether a zero was found.
   634	// It is called only if the verb is %v.
   635	func (s *ss) scanBasePrefix() (base int, digits string, found bool) {
   636		if !s.peek("0") {
   637			return 10, decimalDigits, false
   638		}
   639		s.accept("0")
   640		found = true // We've put a digit into the token buffer.
   641		// Special cases for '0' && '0x'
   642		base, digits = 8, octalDigits
   643		if s.peek("xX") {
   644			s.consume("xX", false)
   645			base, digits = 16, hexadecimalDigits
   646		}
   647		return
   648	}
   649	
   650	// scanInt returns the value of the integer represented by the next
   651	// token, checking for overflow.  Any error is stored in s.err.
   652	func (s *ss) scanInt(verb rune, bitSize int) int64 {
   653		if verb == 'c' {
   654			return s.scanRune(bitSize)
   655		}
   656		s.skipSpace(false)
   657		s.notEOF()
   658		base, digits := s.getBase(verb)
   659		haveDigits := false
   660		if verb == 'U' {
   661			if !s.consume("U", false) || !s.consume("+", false) {
   662				s.errorString("bad unicode format ")
   663			}
   664		} else {
   665			s.accept(sign) // If there's a sign, it will be left in the token buffer.
   666			if verb == 'v' {
   667				base, digits, haveDigits = s.scanBasePrefix()
   668			}
   669		}
   670		tok := s.scanNumber(digits, haveDigits)
   671		i, err := strconv.ParseInt(tok, base, 64)
   672		if err != nil {
   673			s.error(err)
   674		}
   675		n := uint(bitSize)
   676		x := (i << (64 - n)) >> (64 - n)
   677		if x != i {
   678			s.errorString("integer overflow on token " + tok)
   679		}
   680		return i
   681	}
   682	
   683	// scanUint returns the value of the unsigned integer represented
   684	// by the next token, checking for overflow.  Any error is stored in s.err.
   685	func (s *ss) scanUint(verb rune, bitSize int) uint64 {
   686		if verb == 'c' {
   687			return uint64(s.scanRune(bitSize))
   688		}
   689		s.skipSpace(false)
   690		s.notEOF()
   691		base, digits := s.getBase(verb)
   692		haveDigits := false
   693		if verb == 'U' {
   694			if !s.consume("U", false) || !s.consume("+", false) {
   695				s.errorString("bad unicode format ")
   696			}
   697		} else if verb == 'v' {
   698			base, digits, haveDigits = s.scanBasePrefix()
   699		}
   700		tok := s.scanNumber(digits, haveDigits)
   701		i, err := strconv.ParseUint(tok, base, 64)
   702		if err != nil {
   703			s.error(err)
   704		}
   705		n := uint(bitSize)
   706		x := (i << (64 - n)) >> (64 - n)
   707		if x != i {
   708			s.errorString("unsigned integer overflow on token " + tok)
   709		}
   710		return i
   711	}
   712	
   713	// floatToken returns the floating-point number starting here, no longer than swid
   714	// if the width is specified. It's not rigorous about syntax because it doesn't check that
   715	// we have at least some digits, but Atof will do that.
   716	func (s *ss) floatToken() string {
   717		s.buf = s.buf[:0]
   718		// NaN?
   719		if s.accept("nN") && s.accept("aA") && s.accept("nN") {
   720			return string(s.buf)
   721		}
   722		// leading sign?
   723		s.accept(sign)
   724		// Inf?
   725		if s.accept("iI") && s.accept("nN") && s.accept("fF") {
   726			return string(s.buf)
   727		}
   728		// digits?
   729		for s.accept(decimalDigits) {
   730		}
   731		// decimal point?
   732		if s.accept(period) {
   733			// fraction?
   734			for s.accept(decimalDigits) {
   735			}
   736		}
   737		// exponent?
   738		if s.accept(exponent) {
   739			// leading sign?
   740			s.accept(sign)
   741			// digits?
   742			for s.accept(decimalDigits) {
   743			}
   744		}
   745		return string(s.buf)
   746	}
   747	
   748	// complexTokens returns the real and imaginary parts of the complex number starting here.
   749	// The number might be parenthesized and has the format (N+Ni) where N is a floating-point
   750	// number and there are no spaces within.
   751	func (s *ss) complexTokens() (real, imag string) {
   752		// TODO: accept N and Ni independently?
   753		parens := s.accept("(")
   754		real = s.floatToken()
   755		s.buf = s.buf[:0]
   756		// Must now have a sign.
   757		if !s.accept("+-") {
   758			s.error(complexError)
   759		}
   760		// Sign is now in buffer
   761		imagSign := string(s.buf)
   762		imag = s.floatToken()
   763		if !s.accept("i") {
   764			s.error(complexError)
   765		}
   766		if parens && !s.accept(")") {
   767			s.error(complexError)
   768		}
   769		return real, imagSign + imag
   770	}
   771	
   772	// convertFloat converts the string to a float64value.
   773	func (s *ss) convertFloat(str string, n int) float64 {
   774		if p := indexRune(str, 'p'); p >= 0 {
   775			// Atof doesn't handle power-of-2 exponents,
   776			// but they're easy to evaluate.
   777			f, err := strconv.ParseFloat(str[:p], n)
   778			if err != nil {
   779				// Put full string into error.
   780				if e, ok := err.(*strconv.NumError); ok {
   781					e.Num = str
   782				}
   783				s.error(err)
   784			}
   785			m, err := strconv.Atoi(str[p+1:])
   786			if err != nil {
   787				// Put full string into error.
   788				if e, ok := err.(*strconv.NumError); ok {
   789					e.Num = str
   790				}
   791				s.error(err)
   792			}
   793			return math.Ldexp(f, m)
   794		}
   795		f, err := strconv.ParseFloat(str, n)
   796		if err != nil {
   797			s.error(err)
   798		}
   799		return f
   800	}
   801	
   802	// convertComplex converts the next token to a complex128 value.
   803	// The atof argument is a type-specific reader for the underlying type.
   804	// If we're reading complex64, atof will parse float32s and convert them
   805	// to float64's to avoid reproducing this code for each complex type.
   806	func (s *ss) scanComplex(verb rune, n int) complex128 {
   807		if !s.okVerb(verb, floatVerbs, "complex") {
   808			return 0
   809		}
   810		s.skipSpace(false)
   811		s.notEOF()
   812		sreal, simag := s.complexTokens()
   813		real := s.convertFloat(sreal, n/2)
   814		imag := s.convertFloat(simag, n/2)
   815		return complex(real, imag)
   816	}
   817	
   818	// convertString returns the string represented by the next input characters.
   819	// The format of the input is determined by the verb.
   820	func (s *ss) convertString(verb rune) (str string) {
   821		if !s.okVerb(verb, "svqx", "string") {
   822			return ""
   823		}
   824		s.skipSpace(false)
   825		s.notEOF()
   826		switch verb {
   827		case 'q':
   828			str = s.quotedString()
   829		case 'x':
   830			str = s.hexString()
   831		default:
   832			str = string(s.token(true, notSpace)) // %s and %v just return the next word
   833		}
   834		return
   835	}
   836	
   837	// quotedString returns the double- or back-quoted string represented by the next input characters.
   838	func (s *ss) quotedString() string {
   839		s.notEOF()
   840		quote := s.getRune()
   841		switch quote {
   842		case '`':
   843			// Back-quoted: Anything goes until EOF or back quote.
   844			for {
   845				r := s.mustReadRune()
   846				if r == quote {
   847					break
   848				}
   849				s.buf.WriteRune(r)
   850			}
   851			return string(s.buf)
   852		case '"':
   853			// Double-quoted: Include the quotes and let strconv.Unquote do the backslash escapes.
   854			s.buf.WriteRune(quote)
   855			for {
   856				r := s.mustReadRune()
   857				s.buf.WriteRune(r)
   858				if r == '\\' {
   859					// In a legal backslash escape, no matter how long, only the character
   860					// immediately after the escape can itself be a backslash or quote.
   861					// Thus we only need to protect the first character after the backslash.
   862					s.buf.WriteRune(s.mustReadRune())
   863				} else if r == '"' {
   864					break
   865				}
   866			}
   867			result, err := strconv.Unquote(string(s.buf))
   868			if err != nil {
   869				s.error(err)
   870			}
   871			return result
   872		default:
   873			s.errorString("expected quoted string")
   874		}
   875		return ""
   876	}
   877	
   878	// hexDigit returns the value of the hexadecimal digit
   879	func (s *ss) hexDigit(d rune) int {
   880		digit := int(d)
   881		switch digit {
   882		case '0', '1', '2', '3', '4', '5', '6', '7', '8', '9':
   883			return digit - '0'
   884		case 'a', 'b', 'c', 'd', 'e', 'f':
   885			return 10 + digit - 'a'
   886		case 'A', 'B', 'C', 'D', 'E', 'F':
   887			return 10 + digit - 'A'
   888		}
   889		s.errorString("illegal hex digit")
   890		return 0
   891	}
   892	
   893	// hexByte returns the next hex-encoded (two-character) byte from the input.
   894	// There must be either two hexadecimal digits or a space character in the input.
   895	func (s *ss) hexByte() (b byte, ok bool) {
   896		rune1 := s.getRune()
   897		if rune1 == eof {
   898			return
   899		}
   900		if isSpace(rune1) {
   901			s.UnreadRune()
   902			return
   903		}
   904		rune2 := s.mustReadRune()
   905		return byte(s.hexDigit(rune1)<<4 | s.hexDigit(rune2)), true
   906	}
   907	
   908	// hexString returns the space-delimited hexpair-encoded string.
   909	func (s *ss) hexString() string {
   910		s.notEOF()
   911		for {
   912			b, ok := s.hexByte()
   913			if !ok {
   914				break
   915			}
   916			s.buf.WriteByte(b)
   917		}
   918		if len(s.buf) == 0 {
   919			s.errorString("no hex data for %x string")
   920			return ""
   921		}
   922		return string(s.buf)
   923	}
   924	
   925	const floatVerbs = "beEfFgGv"
   926	
   927	const hugeWid = 1 << 30
   928	
   929	// scanOne scans a single value, deriving the scanner from the type of the argument.
   930	func (s *ss) scanOne(verb rune, arg interface{}) {
   931		s.buf = s.buf[:0]
   932		var err error
   933		// If the parameter has its own Scan method, use that.
   934		if v, ok := arg.(Scanner); ok {
   935			err = v.Scan(s, verb)
   936			if err != nil {
   937				if err == io.EOF {
   938					err = io.ErrUnexpectedEOF
   939				}
   940				s.error(err)
   941			}
   942			return
   943		}
   944	
   945		switch v := arg.(type) {
   946		case *bool:
   947			*v = s.scanBool(verb)
   948		case *complex64:
   949			*v = complex64(s.scanComplex(verb, 64))
   950		case *complex128:
   951			*v = s.scanComplex(verb, 128)
   952		case *int:
   953			*v = int(s.scanInt(verb, intBits))
   954		case *int8:
   955			*v = int8(s.scanInt(verb, 8))
   956		case *int16:
   957			*v = int16(s.scanInt(verb, 16))
   958		case *int32:
   959			*v = int32(s.scanInt(verb, 32))
   960		case *int64:
   961			*v = s.scanInt(verb, 64)
   962		case *uint:
   963			*v = uint(s.scanUint(verb, intBits))
   964		case *uint8:
   965			*v = uint8(s.scanUint(verb, 8))
   966		case *uint16:
   967			*v = uint16(s.scanUint(verb, 16))
   968		case *uint32:
   969			*v = uint32(s.scanUint(verb, 32))
   970		case *uint64:
   971			*v = s.scanUint(verb, 64)
   972		case *uintptr:
   973			*v = uintptr(s.scanUint(verb, uintptrBits))
   974		// Floats are tricky because you want to scan in the precision of the result, not
   975		// scan in high precision and convert, in order to preserve the correct error condition.
   976		case *float32:
   977			if s.okVerb(verb, floatVerbs, "float32") {
   978				s.skipSpace(false)
   979				s.notEOF()
   980				*v = float32(s.convertFloat(s.floatToken(), 32))
   981			}
   982		case *float64:
   983			if s.okVerb(verb, floatVerbs, "float64") {
   984				s.skipSpace(false)
   985				s.notEOF()
   986				*v = s.convertFloat(s.floatToken(), 64)
   987			}
   988		case *string:
   989			*v = s.convertString(verb)
   990		case *[]byte:
   991			// We scan to string and convert so we get a copy of the data.
   992			// If we scanned to bytes, the slice would point at the buffer.
   993			*v = []byte(s.convertString(verb))
   994		default:
   995			val := reflect.ValueOf(v)
   996			ptr := val
   997			if ptr.Kind() != reflect.Ptr {
   998				s.errorString("type not a pointer: " + val.Type().String())
   999				return
  1000			}
  1001			switch v := ptr.Elem(); v.Kind() {
  1002			case reflect.Bool:
  1003				v.SetBool(s.scanBool(verb))
  1004			case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
  1005				v.SetInt(s.scanInt(verb, v.Type().Bits()))
  1006			case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
  1007				v.SetUint(s.scanUint(verb, v.Type().Bits()))
  1008			case reflect.String:
  1009				v.SetString(s.convertString(verb))
  1010			case reflect.Slice:
  1011				// For now, can only handle (renamed) []byte.
  1012				typ := v.Type()
  1013				if typ.Elem().Kind() != reflect.Uint8 {
  1014					s.errorString("can't scan type: " + val.Type().String())
  1015				}
  1016				str := s.convertString(verb)
  1017				v.Set(reflect.MakeSlice(typ, len(str), len(str)))
  1018				for i := 0; i < len(str); i++ {
  1019					v.Index(i).SetUint(uint64(str[i]))
  1020				}
  1021			case reflect.Float32, reflect.Float64:
  1022				s.skipSpace(false)
  1023				s.notEOF()
  1024				v.SetFloat(s.convertFloat(s.floatToken(), v.Type().Bits()))
  1025			case reflect.Complex64, reflect.Complex128:
  1026				v.SetComplex(s.scanComplex(verb, v.Type().Bits()))
  1027			default:
  1028				s.errorString("can't scan type: " + val.Type().String())
  1029			}
  1030		}
  1031	}
  1032	
  1033	// errorHandler turns local panics into error returns.
  1034	func errorHandler(errp *error) {
  1035		if e := recover(); e != nil {
  1036			if se, ok := e.(scanError); ok { // catch local error
  1037				*errp = se.err
  1038			} else if eof, ok := e.(error); ok && eof == io.EOF { // out of input
  1039				*errp = eof
  1040			} else {
  1041				panic(e)
  1042			}
  1043		}
  1044	}
  1045	
  1046	// doScan does the real work for scanning without a format string.
  1047	func (s *ss) doScan(a []interface{}) (numProcessed int, err error) {
  1048		defer errorHandler(&err)
  1049		for _, arg := range a {
  1050			s.scanOne('v', arg)
  1051			numProcessed++
  1052		}
  1053		// Check for newline if required.
  1054		if !s.nlIsSpace {
  1055			for {
  1056				r := s.getRune()
  1057				if r == '\n' || r == eof {
  1058					break
  1059				}
  1060				if !isSpace(r) {
  1061					s.errorString("expected newline")
  1062					break
  1063				}
  1064			}
  1065		}
  1066		return
  1067	}
  1068	
  1069	// advance determines whether the next characters in the input match
  1070	// those of the format.  It returns the number of bytes (sic) consumed
  1071	// in the format. Newlines included, all runs of space characters in
  1072	// either input or format behave as a single space. This routine also
  1073	// handles the %% case.  If the return value is zero, either format
  1074	// starts with a % (with no following %) or the input is empty.
  1075	// If it is negative, the input did not match the string.
  1076	func (s *ss) advance(format string) (i int) {
  1077		for i < len(format) {
  1078			fmtc, w := utf8.DecodeRuneInString(format[i:])
  1079			if fmtc == '%' {
  1080				// %% acts like a real percent
  1081				nextc, _ := utf8.DecodeRuneInString(format[i+w:]) // will not match % if string is empty
  1082				if nextc != '%' {
  1083					return
  1084				}
  1085				i += w // skip the first %
  1086			}
  1087			sawSpace := false
  1088			for isSpace(fmtc) && i < len(format) {
  1089				sawSpace = true
  1090				i += w
  1091				fmtc, w = utf8.DecodeRuneInString(format[i:])
  1092			}
  1093			if sawSpace {
  1094				// There was space in the format, so there should be space (EOF)
  1095				// in the input.
  1096				inputc := s.getRune()
  1097				if inputc == eof || inputc == '\n' {
  1098					// If we've reached a newline, stop now; don't read ahead.
  1099					return
  1100				}
  1101				if !isSpace(inputc) {
  1102					// Space in format but not in input: error
  1103					s.errorString("expected space in input to match format")
  1104				}
  1105				s.skipSpace(true)
  1106				continue
  1107			}
  1108			inputc := s.mustReadRune()
  1109			if fmtc != inputc {
  1110				s.UnreadRune()
  1111				return -1
  1112			}
  1113			i += w
  1114		}
  1115		return
  1116	}
  1117	
  1118	// doScanf does the real work when scanning with a format string.
  1119	//  At the moment, it handles only pointers to basic types.
  1120	func (s *ss) doScanf(format string, a []interface{}) (numProcessed int, err error) {
  1121		defer errorHandler(&err)
  1122		end := len(format) - 1
  1123		// We process one item per non-trivial format
  1124		for i := 0; i <= end; {
  1125			w := s.advance(format[i:])
  1126			if w > 0 {
  1127				i += w
  1128				continue
  1129			}
  1130			// Either we failed to advance, we have a percent character, or we ran out of input.
  1131			if format[i] != '%' {
  1132				// Can't advance format.  Why not?
  1133				if w < 0 {
  1134					s.errorString("input does not match format")
  1135				}
  1136				// Otherwise at EOF; "too many operands" error handled below
  1137				break
  1138			}
  1139			i++ // % is one byte
  1140	
  1141			// do we have 20 (width)?
  1142			var widPresent bool
  1143			s.maxWid, widPresent, i = parsenum(format, i, end)
  1144			if !widPresent {
  1145				s.maxWid = hugeWid
  1146			}
  1147			s.argLimit = s.limit
  1148			if f := s.count + s.maxWid; f < s.argLimit {
  1149				s.argLimit = f
  1150			}
  1151	
  1152			c, w := utf8.DecodeRuneInString(format[i:])
  1153			i += w
  1154	
  1155			if numProcessed >= len(a) { // out of operands
  1156				s.errorString("too few operands for format %" + format[i-w:])
  1157				break
  1158			}
  1159			arg := a[numProcessed]
  1160	
  1161			s.scanOne(c, arg)
  1162			numProcessed++
  1163			s.argLimit = s.limit
  1164		}
  1165		if numProcessed < len(a) {
  1166			s.errorString("too many operands")
  1167		}
  1168		return
  1169	}
  1170	

View as plain text