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.
18 // readRuner is the interface to something that can read runes. If
19 // the object provided to Scan does not satisfy this interface, the
20 // object will be wrapped by a readRune object.
21 type readRuner interface {
22 ReadRune() (rune int, size int, err os.Error)
25 // unreadRuner is the interface to something that can unread runes.
26 // If the object provided to Scan does not satisfy this interface,
27 // a local buffer will be used to back up the input, but its contents
28 // will be lost when Scan returns.
29 type unreadRuner interface {
33 // ScanState represents the scanner state passed to custom scanners.
34 // Scanners may do rune-at-a-time scanning or ask the ScanState
35 // to discover the next space-delimited token.
36 type ScanState interface {
37 // GetRune reads the next rune (Unicode code point) from the input.
38 GetRune() (rune int, err os.Error)
39 // UngetRune causes the next call to GetRune to return the rune.
41 // Width returns the value of the width option and whether it has been set.
42 // The unit is Unicode code points.
43 Width() (wid int, ok bool)
44 // Token returns the next space-delimited token from the input. If
45 // a width has been specified, the returned token will be no longer
47 Token() (token string, err os.Error)
50 // Scanner is implemented by any value that has a Scan method, which scans
51 // the input for the representation of a value and stores the result in the
52 // receiver, which must be a pointer to be useful. The Scan method is called
53 // for any argument to Scan or Scanln that implements it.
54 type Scanner interface {
55 Scan(state ScanState, verb int) os.Error
58 // Scan scans text read from standard input, storing successive
59 // space-separated values into successive arguments. Newlines count
60 // as space. It returns the number of items successfully scanned.
61 // If that is less than the number of arguments, err will report why.
62 func Scan(a ...interface{}) (n int, err os.Error) {
63 return Fscan(os.Stdin, a...)
66 // Scanln is similar to Scan, but stops scanning at a newline and
67 // after the final item there must be a newline or EOF.
68 func Scanln(a ...interface{}) (n int, err os.Error) {
69 return Fscanln(os.Stdin, a...)
72 // Scanf scans text read from standard input, storing successive
73 // space-separated values into successive arguments as determined by
74 // the format. It returns the number of items successfully scanned.
75 func Scanf(format string, a ...interface{}) (n int, err os.Error) {
76 return Fscanf(os.Stdin, format, a...)
79 // Sscan scans the argument string, storing successive space-separated
80 // values into successive arguments. Newlines count as space. It
81 // returns the number of items successfully scanned. If that is less
82 // than the number of arguments, err will report why.
83 func Sscan(str string, a ...interface{}) (n int, err os.Error) {
84 return Fscan(strings.NewReader(str), a...)
87 // Sscanln is similar to Sscan, but stops scanning at a newline and
88 // after the final item there must be a newline or EOF.
89 func Sscanln(str string, a ...interface{}) (n int, err os.Error) {
90 return Fscanln(strings.NewReader(str), a...)
93 // Sscanf scans the argument string, storing successive space-separated
94 // values into successive arguments as determined by the format. It
95 // returns the number of items successfully parsed.
96 func Sscanf(str string, format string, a ...interface{}) (n int, err os.Error) {
97 return Fscanf(strings.NewReader(str), format, a...)
100 // Fscan scans text read from r, storing successive space-separated
101 // values into successive arguments. Newlines count as space. It
102 // returns the number of items successfully scanned. If that is less
103 // than the number of arguments, err will report why.
104 func Fscan(r io.Reader, a ...interface{}) (n int, err os.Error) {
105 s := newScanState(r, true)
111 // Fscanln is similar to Fscan, but stops scanning at a newline and
112 // after the final item there must be a newline or EOF.
113 func Fscanln(r io.Reader, a ...interface{}) (n int, err os.Error) {
114 s := newScanState(r, false)
120 // Fscanf scans text read from r, storing successive space-separated
121 // values into successive arguments as determined by the format. It
122 // returns the number of items successfully parsed.
123 func Fscanf(r io.Reader, format string, a ...interface{}) (n int, err os.Error) {
124 s := newScanState(r, false)
125 n, err = s.doScanf(format, a)
130 // scanError represents an error generated by the scanning software.
131 // It's used as a unique signature to identify such errors when recovering.
132 type scanError struct {
138 // ss is the internal implementation of ScanState.
140 rr readRuner // where to read input
141 buf bytes.Buffer // token accumulator
142 nlIsSpace bool // whether newline counts as white space
143 peekRune int // one-rune lookahead
144 prevRune int // last rune returned by GetRune
145 atEOF bool // already read EOF
146 maxWid int // max width of field, in runes
147 widPresent bool // width was specified
148 wid int // width consumed so far; used in accept()
151 func (s *ss) GetRune() (rune int, err os.Error) {
158 rune, _, err = s.rr.ReadRune()
165 func (s *ss) Width() (wid int, ok bool) {
166 return s.maxWid, s.widPresent
169 // The public method returns an error; this private one panics.
170 // If getRune reaches EOF, the return value is EOF (-1).
171 func (s *ss) getRune() (rune int) {
181 rune, _, err := s.rr.ReadRune()
184 } else if err != nil {
194 // mustGetRune turns os.EOF into a panic(io.ErrUnexpectedEOF).
195 // It is called in cases such as string scanning where an EOF is a
197 func (s *ss) mustGetRune() (rune int) {
199 s.error(io.ErrUnexpectedEOF)
206 rune, _, err := s.rr.ReadRune()
209 err = io.ErrUnexpectedEOF
217 func (s *ss) UngetRune() {
218 if u, ok := s.rr.(unreadRuner); ok {
221 s.peekRune = s.prevRune
225 func (s *ss) error(err os.Error) {
226 panic(scanError{err})
229 func (s *ss) errorString(err string) {
230 panic(scanError{os.ErrorString(err)})
233 func (s *ss) Token() (tok string, err os.Error) {
235 if e := recover(); e != nil {
236 if se, ok := e.(scanError); ok {
247 // readRune is a structure to enable reading UTF-8 encoded code points
248 // from an io.Reader. It is used if the Reader given to the scanner does
249 // not already implement ReadRuner.
250 type readRune struct {
252 buf [utf8.UTFMax]byte // used only inside ReadRune
253 pending int // number of bytes in pendBuf; only >0 for bad UTF-8
254 pendBuf [utf8.UTFMax]byte // bytes left over
257 // readByte returns the next byte from the input, which may be
258 // left over from a previous read if the UTF-8 was ill-formed.
259 func (r *readRune) readByte() (b byte, err os.Error) {
262 copy(r.pendBuf[0:], r.pendBuf[1:])
266 _, err = r.reader.Read(r.pendBuf[0:1])
267 return r.pendBuf[0], err
270 // unread saves the bytes for the next read.
271 func (r *readRune) unread(buf []byte) {
272 copy(r.pendBuf[r.pending:], buf)
273 r.pending += len(buf)
276 // ReadRune returns the next UTF-8 encoded code point from the
277 // io.Reader inside r.
278 func (r *readRune) ReadRune() (rune int, size int, err os.Error) {
279 r.buf[0], err = r.readByte()
283 if r.buf[0] < utf8.RuneSelf { // fast check for common ASCII case
288 for n = 1; !utf8.FullRune(r.buf[0:n]); n++ {
289 r.buf[n], err = r.readByte()
298 rune, size = utf8.DecodeRune(r.buf[0:n])
299 if size < n { // an error
300 r.unread(r.buf[size:n])
306 // A leaky bucket of reusable ss structures.
307 var ssFree = make(chan *ss, 100)
309 // Allocate a new ss struct. Probably can grab the previous one from ssFree.
310 func newScanState(r io.Reader, nlIsSpace bool) *ss {
315 if rr, ok := r.(readRuner); ok {
318 s.rr = &readRune{reader: r}
320 s.nlIsSpace = nlIsSpace
328 // Save used ss structs in ssFree; avoid an allocation per invocation.
329 func (s *ss) free() {
330 // Don't hold on to ss structs with large buffers.
331 if cap(s.buf.Bytes()) > 1024 {
339 // skipSpace skips spaces and maybe newlines.
340 func (s *ss) skipSpace(stopAtNewline bool) {
353 s.errorString("unexpected newline")
356 if !unicode.IsSpace(rune) {
363 // token returns the next space-delimited string from the input. It
364 // skips white space. For Scanln, it stops at newlines. For Scan,
365 // newlines are treated as spaces.
366 func (s *ss) token() string {
368 // read until white space or newline
369 for nrunes := 0; !s.widPresent || nrunes < s.maxWid; nrunes++ {
374 if unicode.IsSpace(rune) {
378 s.buf.WriteRune(rune)
380 return s.buf.String()
383 // typeError indicates that the type of the operand did not match the format
384 func (s *ss) typeError(field interface{}, expected string) {
385 s.errorString("expected field of type pointer to " + expected + "; found " + reflect.Typeof(field).String())
388 var complexError = os.ErrorString("syntax error scanning complex number")
389 var boolError = os.ErrorString("syntax error scanning boolean")
391 // accepts checks the next rune in the input. If it's a byte (sic) in the string, it puts it in the
392 // buffer and returns true. Otherwise it return false.
393 func (s *ss) accept(ok string) bool {
394 if s.wid >= s.maxWid {
401 for i := 0; i < len(ok); i++ {
402 if int(ok[i]) == rune {
403 s.buf.WriteRune(rune)
414 // okVerb verifies that the verb is present in the list, setting s.err appropriately if not.
415 func (s *ss) okVerb(verb int, okVerbs, typ string) bool {
416 for _, v := range okVerbs {
421 s.errorString("bad verb %" + string(verb) + " for " + typ)
425 // scanBool returns the value of the boolean represented by the next token.
426 func (s *ss) scanBool(verb int) bool {
427 if !s.okVerb(verb, "tv", "boolean") {
430 // Syntax-checking a boolean is annoying. We're not fastidious about case.
431 switch s.mustGetRune() {
437 if s.accept("rR") && (!s.accept("uU") || !s.accept("eE")) {
442 if s.accept("aL") && (!s.accept("lL") || !s.accept("sS") || !s.accept("eE")) {
450 // Numerical elements
453 octalDigits = "01234567"
454 decimalDigits = "0123456789"
455 hexadecimalDigits = "0123456789aAbBcCdDeEfF"
461 // getBase returns the numeric base represented by the verb and its digit string.
462 func (s *ss) getBase(verb int) (base int, digits string) {
463 s.okVerb(verb, "bdoxXv", "integer") // sets s.err
465 digits = decimalDigits
469 digits = binaryDigits
475 digits = hexadecimalDigits
480 // scanNumber returns the numerical string with specified digits starting here.
481 func (s *ss) scanNumber(digits string) string {
482 if !s.accept(digits) {
483 s.errorString("expected integer")
485 for s.accept(digits) {
487 return s.buf.String()
490 // scanRune returns the next rune value in the input.
491 func (s *ss) scanRune(bitSize int) int64 {
492 rune := int64(s.mustGetRune())
494 x := (rune << (64 - n)) >> (64 - n)
496 s.errorString("overflow on character value " + string(rune))
501 // scanInt returns the value of the integer represented by the next
502 // token, checking for overflow. Any error is stored in s.err.
503 func (s *ss) scanInt(verb int, bitSize int) int64 {
505 return s.scanRune(bitSize)
507 base, digits := s.getBase(verb)
509 s.accept(sign) // If there's a sign, it will be left in the token buffer.
510 tok := s.scanNumber(digits)
511 i, err := strconv.Btoi64(tok, base)
516 x := (i << (64 - n)) >> (64 - n)
518 s.errorString("integer overflow on token " + tok)
523 // scanUint returns the value of the unsigned integer represented
524 // by the next token, checking for overflow. Any error is stored in s.err.
525 func (s *ss) scanUint(verb int, bitSize int) uint64 {
527 return uint64(s.scanRune(bitSize))
529 base, digits := s.getBase(verb)
531 tok := s.scanNumber(digits)
532 i, err := strconv.Btoui64(tok, base)
537 x := (i << (64 - n)) >> (64 - n)
539 s.errorString("unsigned integer overflow on token " + tok)
544 // floatToken returns the floating-point number starting here, no longer than swid
545 // if the width is specified. It's not rigorous about syntax because it doesn't check that
546 // we have at least some digits, but Atof will do that.
547 func (s *ss) floatToken() string {
552 for s.accept(decimalDigits) {
555 if s.accept(period) {
557 for s.accept(decimalDigits) {
561 if s.accept(exponent) {
565 for s.accept(decimalDigits) {
568 return s.buf.String()
571 // complexTokens returns the real and imaginary parts of the complex number starting here.
572 // The number might be parenthesized and has the format (N+Ni) where N is a floating-point
573 // number and there are no spaces within.
574 func (s *ss) complexTokens() (real, imag string) {
575 // TODO: accept N and Ni independently?
576 parens := s.accept("(")
577 real = s.floatToken()
579 // Must now have a sign.
581 s.error(complexError)
583 // Sign is now in buffer
584 imagSign := s.buf.String()
585 imag = s.floatToken()
587 s.error(complexError)
589 if parens && !s.accept(")") {
590 s.error(complexError)
592 return real, imagSign + imag
595 // convertFloat converts the string to a float64value.
596 func (s *ss) convertFloat(str string, n int) float64 {
597 f, err := strconv.AtofN(str, n)
604 // convertComplex converts the next token to a complex128 value.
605 // The atof argument is a type-specific reader for the underlying type.
606 // If we're reading complex64, atof will parse float32s and convert them
607 // to float64's to avoid reproducing this code for each complex type.
608 func (s *ss) scanComplex(verb int, n int) complex128 {
609 if !s.okVerb(verb, floatVerbs, "complex") {
613 sreal, simag := s.complexTokens()
614 real := s.convertFloat(sreal, n/2)
615 imag := s.convertFloat(simag, n/2)
616 return cmplx(real, imag)
619 // convertString returns the string represented by the next input characters.
620 // The format of the input is determined by the verb.
621 func (s *ss) convertString(verb int) (str string) {
622 if !s.okVerb(verb, "svqx", "string") {
628 str = s.quotedString()
632 str = s.token() // %s and %v just return the next word
634 // Empty strings other than with %q are not OK.
635 if len(str) == 0 && verb != 'q' && s.maxWid > 0 {
636 s.errorString("Scan: no data for string")
641 // quotedString returns the double- or back-quoted string represented by the next input characters.
642 func (s *ss) quotedString() string {
643 quote := s.mustGetRune()
646 // Back-quoted: Anything goes until EOF or back quote.
648 rune := s.mustGetRune()
652 s.buf.WriteRune(rune)
654 return s.buf.String()
656 // Double-quoted: Include the quotes and let strconv.Unquote do the backslash escapes.
657 s.buf.WriteRune(quote)
659 rune := s.mustGetRune()
660 s.buf.WriteRune(rune)
662 // In a legal backslash escape, no matter how long, only the character
663 // immediately after the escape can itself be a backslash or quote.
664 // Thus we only need to protect the first character after the backslash.
665 rune := s.mustGetRune()
666 s.buf.WriteRune(rune)
667 } else if rune == '"' {
671 result, err := strconv.Unquote(s.buf.String())
677 s.errorString("expected quoted string")
682 // hexDigit returns the value of the hexadecimal digit
683 func (s *ss) hexDigit(digit int) int {
685 case '0', '1', '2', '3', '4', '5', '6', '7', '8', '9':
687 case 'a', 'b', 'c', 'd', 'e', 'f':
688 return 10 + digit - 'a'
689 case 'A', 'B', 'C', 'D', 'E', 'F':
690 return 10 + digit - 'A'
692 s.errorString("Scan: illegal hex digit")
696 // hexByte returns the next hex-encoded (two-character) byte from the input.
697 // There must be either two hexadecimal digits or a space character in the input.
698 func (s *ss) hexByte() (b byte, ok bool) {
703 if unicode.IsSpace(rune1) {
707 rune2 := s.mustGetRune()
708 return byte(s.hexDigit(rune1)<<4 | s.hexDigit(rune2)), true
711 // hexString returns the space-delimited hexpair-encoded string.
712 func (s *ss) hexString() string {
720 if s.buf.Len() == 0 {
721 s.errorString("Scan: no hex data for %x string")
724 return s.buf.String()
727 const floatVerbs = "eEfFgGv"
729 // scanOne scans a single value, deriving the scanner from the type of the argument.
730 func (s *ss) scanOne(verb int, field interface{}) {
733 // If the parameter has its own Scan method, use that.
734 if v, ok := field.(Scanner); ok {
735 err = v.Scan(s, verb)
742 s.maxWid = 1 << 30 // Huge
745 switch v := field.(type) {
747 *v = s.scanBool(verb)
749 *v = complex(s.scanComplex(verb, int(complexBits)))
751 *v = complex64(s.scanComplex(verb, 64))
753 *v = s.scanComplex(verb, 128)
755 *v = int(s.scanInt(verb, intBits))
757 *v = int8(s.scanInt(verb, 8))
759 *v = int16(s.scanInt(verb, 16))
761 *v = int32(s.scanInt(verb, 32))
763 *v = s.scanInt(verb, 64)
765 *v = uint(s.scanUint(verb, intBits))
767 *v = uint8(s.scanUint(verb, 8))
769 *v = uint16(s.scanUint(verb, 16))
771 *v = uint32(s.scanUint(verb, 32))
773 *v = s.scanUint(verb, 64)
775 *v = uintptr(s.scanUint(verb, uintptrBits))
776 // Floats are tricky because you want to scan in the precision of the result, not
777 // scan in high precision and convert, in order to preserve the correct error condition.
779 if s.okVerb(verb, floatVerbs, "float") {
781 *v = float(s.convertFloat(s.floatToken(), int(floatBits)))
784 if s.okVerb(verb, floatVerbs, "float32") {
786 *v = float32(s.convertFloat(s.floatToken(), 32))
789 if s.okVerb(verb, floatVerbs, "float64") {
791 *v = s.convertFloat(s.floatToken(), 64)
794 *v = s.convertString(verb)
796 // We scan to string and convert so we get a copy of the data.
797 // If we scanned to bytes, the slice would point at the buffer.
798 *v = []byte(s.convertString(verb))
800 val := reflect.NewValue(v)
801 ptr, ok := val.(*reflect.PtrValue)
803 s.errorString("Scan: type not a pointer: " + val.Type().String())
806 switch v := ptr.Elem().(type) {
807 case *reflect.BoolValue:
808 v.Set(s.scanBool(verb))
809 case *reflect.IntValue:
810 v.Set(s.scanInt(verb, v.Type().Bits()))
811 case *reflect.UintValue:
812 v.Set(s.scanUint(verb, v.Type().Bits()))
813 case *reflect.StringValue:
814 v.Set(s.convertString(verb))
815 case *reflect.SliceValue:
816 // For now, can only handle (renamed) []byte.
817 typ := v.Type().(*reflect.SliceType)
818 if typ.Elem().Kind() != reflect.Uint8 {
821 str := s.convertString(verb)
822 v.Set(reflect.MakeSlice(typ, len(str), len(str)))
823 for i := 0; i < len(str); i++ {
824 v.Elem(i).(*reflect.UintValue).Set(uint64(str[i]))
826 case *reflect.FloatValue:
828 v.Set(s.convertFloat(s.floatToken(), v.Type().Bits()))
829 case *reflect.ComplexValue:
830 v.Set(s.scanComplex(verb, v.Type().Bits()))
833 s.errorString("Scan: can't handle type: " + val.Type().String())
838 // errorHandler turns local panics into error returns. EOFs are benign.
839 func errorHandler(errp *os.Error) {
840 if e := recover(); e != nil {
841 if se, ok := e.(scanError); ok { // catch local error
842 if se.err != os.EOF {
851 // doScan does the real work for scanning without a format string.
852 // At the moment, it handles only pointers to basic types.
853 func (s *ss) doScan(a []interface{}) (numProcessed int, err os.Error) {
854 defer errorHandler(&err)
855 for _, field := range a {
856 s.scanOne('v', field)
859 // Check for newline if required.
863 if rune == '\n' || rune == EOF {
866 if !unicode.IsSpace(rune) {
867 s.errorString("Scan: expected newline")
875 // advance determines whether the next characters in the input match
876 // those of the format. It returns the number of bytes (sic) consumed
877 // in the format. Newlines included, all runs of space characters in
878 // either input or format behave as a single space. This routine also
879 // handles the %% case. If the return value is zero, either format
880 // starts with a % (with no following %) or the input is empty.
881 // If it is negative, the input did not match the string.
882 func (s *ss) advance(format string) (i int) {
883 for i < len(format) {
884 fmtc, w := utf8.DecodeRuneInString(format[i:])
886 // %% acts like a real percent
887 nextc, _ := utf8.DecodeRuneInString(format[i+w:]) // will not match % if string is empty
891 i += w // skip the first %
894 for unicode.IsSpace(fmtc) && i < len(format) {
897 fmtc, w = utf8.DecodeRuneInString(format[i:])
900 // There was space in the format, so there should be space (EOF)
902 inputc := s.getRune()
906 if !unicode.IsSpace(inputc) {
907 // Space in format but not in input: error
908 s.errorString("expected space in input to match format")
913 inputc := s.mustGetRune()
923 // doScanf does the real work when scanning with a format string.
924 // At the moment, it handles only pointers to basic types.
925 func (s *ss) doScanf(format string, a []interface{}) (numProcessed int, err os.Error) {
926 defer errorHandler(&err)
927 end := len(format) - 1
928 // We process one item per non-trivial format
929 for i := 0; i <= end; {
930 w := s.advance(format[i:])
935 // Either we failed to advance, we have a percent character, or we ran out of input.
936 if format[i] != '%' {
937 // Can't advance format. Why not?
939 s.errorString("input does not match format")
941 // Otherwise at EOF; "too many operands" error handled below
946 // do we have 20 (width)?
947 s.maxWid, s.widPresent, i = parsenum(format, i, end)
949 c, w := utf8.DecodeRuneInString(format[i:])
952 if numProcessed >= len(a) { // out of operands
953 s.errorString("too few operands for format %" + format[i-w:])
956 field := a[numProcessed]
961 if numProcessed < len(a) {
962 s.errorString("too many operands")