Source file src/pkg/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		"unicode/utf8"
    15	)
    16	
    17	// runeUnreader is the interface to something that can unread runes.
    18	// If the object provided to Scan does not satisfy this interface,
    19	// a local buffer will be used to back up the input, but its contents
    20	// will be lost when Scan returns.
    21	type runeUnreader interface {
    22		UnreadRune() error
    23	}
    24	
    25	// ScanState represents the scanner state passed to custom scanners.
    26	// Scanners may do rune-at-a-time scanning or ask the ScanState
    27	// to discover the next space-delimited token.
    28	type ScanState interface {
    29		// ReadRune reads the next rune (Unicode code point) from the input.
    30		// If invoked during Scanln, Fscanln, or Sscanln, ReadRune() will
    31		// return EOF after returning the first '\n' or when reading beyond
    32		// the specified width.
    33		ReadRune() (r rune, size int, err error)
    34		// UnreadRune causes the next call to ReadRune to return the same rune.
    35		UnreadRune() error
    36		// SkipSpace skips space in the input. Newlines are treated as space
    37		// unless the scan operation is Scanln, Fscanln or Sscanln, in which case
    38		// a newline is treated as EOF.
    39		SkipSpace()
    40		// Token skips space in the input if skipSpace is true, then returns the
    41		// run of Unicode code points c satisfying f(c).  If f is nil,
    42		// !unicode.IsSpace(c) is used; that is, the token will hold non-space
    43		// characters.  Newlines are treated as space unless the scan operation
    44		// is Scanln, Fscanln or Sscanln, in which case a newline is treated as
    45		// EOF.  The returned slice points to shared data that may be overwritten
    46		// by the next call to Token, a call to a Scan function using the ScanState
    47		// as input, or when the calling Scan method returns.
    48		Token(skipSpace bool, f func(rune) bool) (token []byte, err error)
    49		// Width returns the value of the width option and whether it has been set.
    50		// The unit is Unicode code points.
    51		Width() (wid int, ok bool)
    52		// Because ReadRune is implemented by the interface, Read should never be
    53		// called by the scanning routines and a valid implementation of
    54		// ScanState may choose always to return an error from Read.
    55		Read(buf []byte) (n int, err error)
    56	}
    57	
    58	// Scanner is implemented by any value that has a Scan method, which scans
    59	// the input for the representation of a value and stores the result in the
    60	// receiver, which must be a pointer to be useful.  The Scan method is called
    61	// for any argument to Scan, Scanf, or Scanln that implements it.
    62	type Scanner interface {
    63		Scan(state ScanState, verb rune) error
    64	}
    65	
    66	// Scan scans text read from standard input, storing successive
    67	// space-separated values into successive arguments.  Newlines count
    68	// as space.  It returns the number of items successfully scanned.
    69	// If that is less than the number of arguments, err will report why.
    70	func Scan(a ...interface{}) (n int, err error) {
    71		return Fscan(os.Stdin, a...)
    72	}
    73	
    74	// Scanln is similar to Scan, but stops scanning at a newline and
    75	// after the final item there must be a newline or EOF.
    76	func Scanln(a ...interface{}) (n int, err error) {
    77		return Fscanln(os.Stdin, a...)
    78	}
    79	
    80	// Scanf scans text read from standard input, storing successive
    81	// space-separated values into successive arguments as determined by
    82	// the format.  It returns the number of items successfully scanned.
    83	func Scanf(format string, a ...interface{}) (n int, err error) {
    84		return Fscanf(os.Stdin, format, a...)
    85	}
    86	
    87	type stringReader string
    88	
    89	func (r *stringReader) Read(b []byte) (n int, err error) {
    90		n = copy(b, *r)
    91		*r = (*r)[n:]
    92		if n == 0 {
    93			err = io.EOF
    94		}
    95		return
    96	}
    97	
    98	// Sscan scans the argument string, storing successive space-separated
    99	// values into successive arguments.  Newlines count as space.  It
   100	// returns the number of items successfully scanned.  If that is less
   101	// than the number of arguments, err will report why.
   102	func Sscan(str string, a ...interface{}) (n int, err error) {
   103		return Fscan((*stringReader)(&str), a...)
   104	}
   105	
   106	// Sscanln is similar to Sscan, but stops scanning at a newline and
   107	// after the final item there must be a newline or EOF.
   108	func Sscanln(str string, a ...interface{}) (n int, err error) {
   109		return Fscanln((*stringReader)(&str), a...)
   110	}
   111	
   112	// Sscanf scans the argument string, storing successive space-separated
   113	// values into successive arguments as determined by the format.  It
   114	// returns the number of items successfully parsed.
   115	func Sscanf(str string, format string, a ...interface{}) (n int, err error) {
   116		return Fscanf((*stringReader)(&str), format, a...)
   117	}
   118	
   119	// Fscan scans text read from r, storing successive space-separated
   120	// values into successive arguments.  Newlines count as space.  It
   121	// returns the number of items successfully scanned.  If that is less
   122	// than the number of arguments, err will report why.
   123	func Fscan(r io.Reader, a ...interface{}) (n int, err error) {
   124		s, old := newScanState(r, true, false)
   125		n, err = s.doScan(a)
   126		s.free(old)
   127		return
   128	}
   129	
   130	// Fscanln is similar to Fscan, but stops scanning at a newline and
   131	// after the final item there must be a newline or EOF.
   132	func Fscanln(r io.Reader, a ...interface{}) (n int, err error) {
   133		s, old := newScanState(r, false, true)
   134		n, err = s.doScan(a)
   135		s.free(old)
   136		return
   137	}
   138	
   139	// Fscanf scans text read from r, storing successive space-separated
   140	// values into successive arguments as determined by the format.  It
   141	// returns the number of items successfully parsed.
   142	func Fscanf(r io.Reader, format string, a ...interface{}) (n int, err error) {
   143		s, old := newScanState(r, false, false)
   144		n, err = s.doScanf(format, a)
   145		s.free(old)
   146		return
   147	}
   148	
   149	// scanError represents an error generated by the scanning software.
   150	// It's used as a unique signature to identify such errors when recovering.
   151	type scanError struct {
   152		err error
   153	}
   154	
   155	const eof = -1
   156	
   157	// ss is the internal implementation of ScanState.
   158	type ss struct {
   159		rr       io.RuneReader // where to read input
   160		buf      buffer        // token accumulator
   161		peekRune rune          // one-rune lookahead
   162		prevRune rune          // last rune returned by ReadRune
   163		count    int           // runes consumed so far.
   164		atEOF    bool          // already read EOF
   165		ssave
   166	}
   167	
   168	// ssave holds the parts of ss that need to be
   169	// saved and restored on recursive scans.
   170	type ssave struct {
   171		validSave  bool // is or was a part of an actual ss.
   172		nlIsEnd    bool // whether newline terminates scan
   173		nlIsSpace  bool // whether newline counts as white space
   174		fieldLimit int  // max value of ss.count for this field; fieldLimit <= limit
   175		limit      int  // max value of ss.count.
   176		maxWid     int  // width of this field.
   177	}
   178	
   179	// The Read method is only in ScanState so that ScanState
   180	// satisfies io.Reader. It will never be called when used as
   181	// intended, so there is no need to make it actually work.
   182	func (s *ss) Read(buf []byte) (n int, err error) {
   183		return 0, errors.New("ScanState's Read should not be called. Use ReadRune")
   184	}
   185	
   186	func (s *ss) ReadRune() (r rune, size int, err error) {
   187		if s.peekRune >= 0 {
   188			s.count++
   189			r = s.peekRune
   190			size = utf8.RuneLen(r)
   191			s.prevRune = r
   192			s.peekRune = -1
   193			return
   194		}
   195		if s.atEOF || s.nlIsEnd && s.prevRune == '\n' || s.count >= s.fieldLimit {
   196			err = io.EOF
   197			return
   198		}
   199	
   200		r, size, err = s.rr.ReadRune()
   201		if err == nil {
   202			s.count++
   203			s.prevRune = r
   204		} else if err == io.EOF {
   205			s.atEOF = true
   206		}
   207		return
   208	}
   209	
   210	func (s *ss) Width() (wid int, ok bool) {
   211		if s.maxWid == hugeWid {
   212			return 0, false
   213		}
   214		return s.maxWid, true
   215	}
   216	
   217	// The public method returns an error; this private one panics.
   218	// If getRune reaches EOF, the return value is EOF (-1).
   219	func (s *ss) getRune() (r rune) {
   220		r, _, err := s.ReadRune()
   221		if err != nil {
   222			if err == io.EOF {
   223				return eof
   224			}
   225			s.error(err)
   226		}
   227		return
   228	}
   229	
   230	// mustReadRune turns io.EOF into a panic(io.ErrUnexpectedEOF).
   231	// It is called in cases such as string scanning where an EOF is a
   232	// syntax error.
   233	func (s *ss) mustReadRune() (r rune) {
   234		r = s.getRune()
   235		if r == eof {
   236			s.error(io.ErrUnexpectedEOF)
   237		}
   238		return
   239	}
   240	
   241	func (s *ss) UnreadRune() error {
   242		if u, ok := s.rr.(runeUnreader); ok {
   243			u.UnreadRune()
   244		} else {
   245			s.peekRune = s.prevRune
   246		}
   247		s.prevRune = -1
   248		s.count--
   249		return nil
   250	}
   251	
   252	func (s *ss) error(err error) {
   253		panic(scanError{err})
   254	}
   255	
   256	func (s *ss) errorString(err string) {
   257		panic(scanError{errors.New(err)})
   258	}
   259	
   260	func (s *ss) Token(skipSpace bool, f func(rune) bool) (tok []byte, err error) {
   261		defer func() {
   262			if e := recover(); e != nil {
   263				if se, ok := e.(scanError); ok {
   264					err = se.err
   265				} else {
   266					panic(e)
   267				}
   268			}
   269		}()
   270		if f == nil {
   271			f = notSpace
   272		}
   273		s.buf = s.buf[:0]
   274		tok = s.token(skipSpace, f)
   275		return
   276	}
   277	
   278	// space is a copy of the unicode.White_Space ranges,
   279	// to avoid depending on package unicode.
   280	var space = [][2]uint16{
   281		{0x0009, 0x000d},
   282		{0x0020, 0x0020},
   283		{0x0085, 0x0085},
   284		{0x00a0, 0x00a0},
   285		{0x1680, 0x1680},
   286		{0x180e, 0x180e},
   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			return
   364		}
   365		var n int
   366		for n = 1; !utf8.FullRune(r.buf[0:n]); n++ {
   367			r.buf[n], err = r.readByte()
   368			if err != nil {
   369				if err == io.EOF {
   370					err = nil
   371					break
   372				}
   373				return
   374			}
   375		}
   376		rr, size = utf8.DecodeRune(r.buf[0:n])
   377		if size < n { // an error
   378			r.unread(r.buf[size:n])
   379		}
   380		return
   381	}
   382	
   383	var ssFree = newCache(func() interface{} { return new(ss) })
   384	
   385	// newScanState allocates a new ss struct or grab a cached one.
   386	func newScanState(r io.Reader, nlIsSpace, nlIsEnd bool) (s *ss, old ssave) {
   387		// If the reader is a *ss, then we've got a recursive
   388		// call to Scan, so re-use the scan state.
   389		s, ok := r.(*ss)
   390		if ok {
   391			old = s.ssave
   392			s.limit = s.fieldLimit
   393			s.nlIsEnd = nlIsEnd || s.nlIsEnd
   394			s.nlIsSpace = nlIsSpace
   395			return
   396		}
   397	
   398		s = ssFree.get().(*ss)
   399		if rr, ok := r.(io.RuneReader); ok {
   400			s.rr = rr
   401		} else {
   402			s.rr = &readRune{reader: r}
   403		}
   404		s.nlIsSpace = nlIsSpace
   405		s.nlIsEnd = nlIsEnd
   406		s.prevRune = -1
   407		s.peekRune = -1
   408		s.atEOF = false
   409		s.limit = hugeWid
   410		s.fieldLimit = hugeWid
   411		s.maxWid = hugeWid
   412		s.validSave = true
   413		s.count = 0
   414		return
   415	}
   416	
   417	// free saves used ss structs in ssFree; avoid an allocation per invocation.
   418	func (s *ss) free(old ssave) {
   419		// If it was used recursively, just restore the old state.
   420		if old.validSave {
   421			s.ssave = old
   422			return
   423		}
   424		// Don't hold on to ss structs with large buffers.
   425		if cap(s.buf) > 1024 {
   426			return
   427		}
   428		s.buf = s.buf[:0]
   429		s.rr = nil
   430		ssFree.put(s)
   431	}
   432	
   433	// skipSpace skips spaces and maybe newlines.
   434	func (s *ss) skipSpace(stopAtNewline bool) {
   435		for {
   436			r := s.getRune()
   437			if r == eof {
   438				return
   439			}
   440			if r == '\n' {
   441				if stopAtNewline {
   442					break
   443				}
   444				if s.nlIsSpace {
   445					continue
   446				}
   447				s.errorString("unexpected newline")
   448				return
   449			}
   450			if !isSpace(r) {
   451				s.UnreadRune()
   452				break
   453			}
   454		}
   455	}
   456	
   457	// token returns the next space-delimited string from the input.  It
   458	// skips white space.  For Scanln, it stops at newlines.  For Scan,
   459	// newlines are treated as spaces.
   460	func (s *ss) token(skipSpace bool, f func(rune) bool) []byte {
   461		if skipSpace {
   462			s.skipSpace(false)
   463		}
   464		// read until white space or newline
   465		for {
   466			r := s.getRune()
   467			if r == eof {
   468				break
   469			}
   470			if !f(r) {
   471				s.UnreadRune()
   472				break
   473			}
   474			s.buf.WriteRune(r)
   475		}
   476		return s.buf
   477	}
   478	
   479	// typeError indicates that the type of the operand did not match the format
   480	func (s *ss) typeError(field interface{}, expected string) {
   481		s.errorString("expected field of type pointer to " + expected + "; found " + reflect.TypeOf(field).String())
   482	}
   483	
   484	var complexError = errors.New("syntax error scanning complex number")
   485	var boolError = errors.New("syntax error scanning boolean")
   486	
   487	func indexRune(s string, r rune) int {
   488		for i, c := range s {
   489			if c == r {
   490				return i
   491			}
   492		}
   493		return -1
   494	}
   495	
   496	// consume reads the next rune in the input and reports whether it is in the ok string.
   497	// If accept is true, it puts the character into the input token.
   498	func (s *ss) consume(ok string, accept bool) bool {
   499		r := s.getRune()
   500		if r == eof {
   501			return false
   502		}
   503		if indexRune(ok, r) >= 0 {
   504			if accept {
   505				s.buf.WriteRune(r)
   506			}
   507			return true
   508		}
   509		if r != eof && accept {
   510			s.UnreadRune()
   511		}
   512		return false
   513	}
   514	
   515	// peek reports whether the next character is in the ok string, without consuming it.
   516	func (s *ss) peek(ok string) bool {
   517		r := s.getRune()
   518		if r != eof {
   519			s.UnreadRune()
   520		}
   521		return indexRune(ok, r) >= 0
   522	}
   523	
   524	func (s *ss) notEOF() {
   525		// Guarantee there is data to be read.
   526		if r := s.getRune(); r == eof {
   527			panic(io.EOF)
   528		}
   529		s.UnreadRune()
   530	}
   531	
   532	// accept checks the next rune in the input.  If it's a byte (sic) in the string, it puts it in the
   533	// buffer and returns true. Otherwise it return false.
   534	func (s *ss) accept(ok string) bool {
   535		return s.consume(ok, true)
   536	}
   537	
   538	// okVerb verifies that the verb is present in the list, setting s.err appropriately if not.
   539	func (s *ss) okVerb(verb rune, okVerbs, typ string) bool {
   540		for _, v := range okVerbs {
   541			if v == verb {
   542				return true
   543			}
   544		}
   545		s.errorString("bad verb %" + string(verb) + " for " + typ)
   546		return false
   547	}
   548	
   549	// scanBool returns the value of the boolean represented by the next token.
   550	func (s *ss) scanBool(verb rune) bool {
   551		s.skipSpace(false)
   552		s.notEOF()
   553		if !s.okVerb(verb, "tv", "boolean") {
   554			return false
   555		}
   556		// Syntax-checking a boolean is annoying.  We're not fastidious about case.
   557		switch s.getRune() {
   558		case '0':
   559			return false
   560		case '1':
   561			return true
   562		case 't', 'T':
   563			if s.accept("rR") && (!s.accept("uU") || !s.accept("eE")) {
   564				s.error(boolError)
   565			}
   566			return true
   567		case 'f', 'F':
   568			if s.accept("aA") && (!s.accept("lL") || !s.accept("sS") || !s.accept("eE")) {
   569				s.error(boolError)
   570			}
   571			return false
   572		}
   573		return false
   574	}
   575	
   576	// Numerical elements
   577	const (
   578		binaryDigits      = "01"
   579		octalDigits       = "01234567"
   580		decimalDigits     = "0123456789"
   581		hexadecimalDigits = "0123456789aAbBcCdDeEfF"
   582		sign              = "+-"
   583		period            = "."
   584		exponent          = "eEp"
   585	)
   586	
   587	// getBase returns the numeric base represented by the verb and its digit string.
   588	func (s *ss) getBase(verb rune) (base int, digits string) {
   589		s.okVerb(verb, "bdoUxXv", "integer") // sets s.err
   590		base = 10
   591		digits = decimalDigits
   592		switch verb {
   593		case 'b':
   594			base = 2
   595			digits = binaryDigits
   596		case 'o':
   597			base = 8
   598			digits = octalDigits
   599		case 'x', 'X', 'U':
   600			base = 16
   601			digits = hexadecimalDigits
   602		}
   603		return
   604	}
   605	
   606	// scanNumber returns the numerical string with specified digits starting here.
   607	func (s *ss) scanNumber(digits string, haveDigits bool) string {
   608		if !haveDigits {
   609			s.notEOF()
   610			if !s.accept(digits) {
   611				s.errorString("expected integer")
   612			}
   613		}
   614		for s.accept(digits) {
   615		}
   616		return string(s.buf)
   617	}
   618	
   619	// scanRune returns the next rune value in the input.
   620	func (s *ss) scanRune(bitSize int) int64 {
   621		s.notEOF()
   622		r := int64(s.getRune())
   623		n := uint(bitSize)
   624		x := (r << (64 - n)) >> (64 - n)
   625		if x != r {
   626			s.errorString("overflow on character value " + string(r))
   627		}
   628		return r
   629	}
   630	
   631	// scanBasePrefix reports whether the integer begins with a 0 or 0x,
   632	// and returns the base, digit string, and whether a zero was found.
   633	// It is called only if the verb is %v.
   634	func (s *ss) scanBasePrefix() (base int, digits string, found bool) {
   635		if !s.peek("0") {
   636			return 10, decimalDigits, false
   637		}
   638		s.accept("0")
   639		found = true // We've put a digit into the token buffer.
   640		// Special cases for '0' && '0x'
   641		base, digits = 8, octalDigits
   642		if s.peek("xX") {
   643			s.consume("xX", false)
   644			base, digits = 16, hexadecimalDigits
   645		}
   646		return
   647	}
   648	
   649	// scanInt returns the value of the integer represented by the next
   650	// token, checking for overflow.  Any error is stored in s.err.
   651	func (s *ss) scanInt(verb rune, bitSize int) int64 {
   652		if verb == 'c' {
   653			return s.scanRune(bitSize)
   654		}
   655		s.skipSpace(false)
   656		s.notEOF()
   657		base, digits := s.getBase(verb)
   658		haveDigits := false
   659		if verb == 'U' {
   660			if !s.consume("U", false) || !s.consume("+", false) {
   661				s.errorString("bad unicode format ")
   662			}
   663		} else {
   664			s.accept(sign) // If there's a sign, it will be left in the token buffer.
   665			if verb == 'v' {
   666				base, digits, haveDigits = s.scanBasePrefix()
   667			}
   668		}
   669		tok := s.scanNumber(digits, haveDigits)
   670		i, err := strconv.ParseInt(tok, base, 64)
   671		if err != nil {
   672			s.error(err)
   673		}
   674		n := uint(bitSize)
   675		x := (i << (64 - n)) >> (64 - n)
   676		if x != i {
   677			s.errorString("integer overflow on token " + tok)
   678		}
   679		return i
   680	}
   681	
   682	// scanUint returns the value of the unsigned integer represented
   683	// by the next token, checking for overflow.  Any error is stored in s.err.
   684	func (s *ss) scanUint(verb rune, bitSize int) uint64 {
   685		if verb == 'c' {
   686			return uint64(s.scanRune(bitSize))
   687		}
   688		s.skipSpace(false)
   689		s.notEOF()
   690		base, digits := s.getBase(verb)
   691		haveDigits := false
   692		if verb == 'U' {
   693			if !s.consume("U", false) || !s.consume("+", false) {
   694				s.errorString("bad unicode format ")
   695			}
   696		} else if verb == 'v' {
   697			base, digits, haveDigits = s.scanBasePrefix()
   698		}
   699		tok := s.scanNumber(digits, haveDigits)
   700		i, err := strconv.ParseUint(tok, base, 64)
   701		if err != nil {
   702			s.error(err)
   703		}
   704		n := uint(bitSize)
   705		x := (i << (64 - n)) >> (64 - n)
   706		if x != i {
   707			s.errorString("unsigned integer overflow on token " + tok)
   708		}
   709		return i
   710	}
   711	
   712	// floatToken returns the floating-point number starting here, no longer than swid
   713	// if the width is specified. It's not rigorous about syntax because it doesn't check that
   714	// we have at least some digits, but Atof will do that.
   715	func (s *ss) floatToken() string {
   716		s.buf = s.buf[:0]
   717		// NaN?
   718		if s.accept("nN") && s.accept("aA") && s.accept("nN") {
   719			return string(s.buf)
   720		}
   721		// leading sign?
   722		s.accept(sign)
   723		// Inf?
   724		if s.accept("iI") && s.accept("nN") && s.accept("fF") {
   725			return string(s.buf)
   726		}
   727		// digits?
   728		for s.accept(decimalDigits) {
   729		}
   730		// decimal point?
   731		if s.accept(period) {
   732			// fraction?
   733			for s.accept(decimalDigits) {
   734			}
   735		}
   736		// exponent?
   737		if s.accept(exponent) {
   738			// leading sign?
   739			s.accept(sign)
   740			// digits?
   741			for s.accept(decimalDigits) {
   742			}
   743		}
   744		return string(s.buf)
   745	}
   746	
   747	// complexTokens returns the real and imaginary parts of the complex number starting here.
   748	// The number might be parenthesized and has the format (N+Ni) where N is a floating-point
   749	// number and there are no spaces within.
   750	func (s *ss) complexTokens() (real, imag string) {
   751		// TODO: accept N and Ni independently?
   752		parens := s.accept("(")
   753		real = s.floatToken()
   754		s.buf = s.buf[:0]
   755		// Must now have a sign.
   756		if !s.accept("+-") {
   757			s.error(complexError)
   758		}
   759		// Sign is now in buffer
   760		imagSign := string(s.buf)
   761		imag = s.floatToken()
   762		if !s.accept("i") {
   763			s.error(complexError)
   764		}
   765		if parens && !s.accept(")") {
   766			s.error(complexError)
   767		}
   768		return real, imagSign + imag
   769	}
   770	
   771	// convertFloat converts the string to a float64value.
   772	func (s *ss) convertFloat(str string, n int) float64 {
   773		if p := indexRune(str, 'p'); p >= 0 {
   774			// Atof doesn't handle power-of-2 exponents,
   775			// but they're easy to evaluate.
   776			f, err := strconv.ParseFloat(str[:p], n)
   777			if err != nil {
   778				// Put full string into error.
   779				if e, ok := err.(*strconv.NumError); ok {
   780					e.Num = str
   781				}
   782				s.error(err)
   783			}
   784			n, err := strconv.Atoi(str[p+1:])
   785			if err != nil {
   786				// Put full string into error.
   787				if e, ok := err.(*strconv.NumError); ok {
   788					e.Num = str
   789				}
   790				s.error(err)
   791			}
   792			return math.Ldexp(f, n)
   793		}
   794		f, err := strconv.ParseFloat(str, n)
   795		if err != nil {
   796			s.error(err)
   797		}
   798		return f
   799	}
   800	
   801	// convertComplex converts the next token to a complex128 value.
   802	// The atof argument is a type-specific reader for the underlying type.
   803	// If we're reading complex64, atof will parse float32s and convert them
   804	// to float64's to avoid reproducing this code for each complex type.
   805	func (s *ss) scanComplex(verb rune, n int) complex128 {
   806		if !s.okVerb(verb, floatVerbs, "complex") {
   807			return 0
   808		}
   809		s.skipSpace(false)
   810		s.notEOF()
   811		sreal, simag := s.complexTokens()
   812		real := s.convertFloat(sreal, n/2)
   813		imag := s.convertFloat(simag, n/2)
   814		return complex(real, imag)
   815	}
   816	
   817	// convertString returns the string represented by the next input characters.
   818	// The format of the input is determined by the verb.
   819	func (s *ss) convertString(verb rune) (str string) {
   820		if !s.okVerb(verb, "svqx", "string") {
   821			return ""
   822		}
   823		s.skipSpace(false)
   824		s.notEOF()
   825		switch verb {
   826		case 'q':
   827			str = s.quotedString()
   828		case 'x':
   829			str = s.hexString()
   830		default:
   831			str = string(s.token(true, notSpace)) // %s and %v just return the next word
   832		}
   833		return
   834	}
   835	
   836	// quotedString returns the double- or back-quoted string represented by the next input characters.
   837	func (s *ss) quotedString() string {
   838		s.notEOF()
   839		quote := s.getRune()
   840		switch quote {
   841		case '`':
   842			// Back-quoted: Anything goes until EOF or back quote.
   843			for {
   844				r := s.mustReadRune()
   845				if r == quote {
   846					break
   847				}
   848				s.buf.WriteRune(r)
   849			}
   850			return string(s.buf)
   851		case '"':
   852			// Double-quoted: Include the quotes and let strconv.Unquote do the backslash escapes.
   853			s.buf.WriteRune(quote)
   854			for {
   855				r := s.mustReadRune()
   856				s.buf.WriteRune(r)
   857				if r == '\\' {
   858					// In a legal backslash escape, no matter how long, only the character
   859					// immediately after the escape can itself be a backslash or quote.
   860					// Thus we only need to protect the first character after the backslash.
   861					r := s.mustReadRune()
   862					s.buf.WriteRune(r)
   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("Scan: 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("Scan: 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, field 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 := field.(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 := field.(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("Scan: 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("Scan: can't handle 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("Scan: can't handle 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 _, field := range a {
  1050			s.scanOne('v', field)
  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("Scan: 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.fieldLimit = s.limit
  1148			if f := s.count + s.maxWid; f < s.fieldLimit {
  1149				s.fieldLimit = 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			field := a[numProcessed]
  1160	
  1161			s.scanOne(c, field)
  1162			numProcessed++
  1163			s.fieldLimit = s.limit
  1164		}
  1165		if numProcessed < len(a) {
  1166			s.errorString("too many operands")
  1167		}
  1168		return
  1169	}

View as plain text