1// Copyright 2018 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
5package text
6
7import (
8 "bytes"
9 "fmt"
10 "io"
11 "strconv"
12 "unicode/utf8"
13
14 "google.golang.org/protobuf/internal/errors"
15)
16
17// Decoder is a token-based textproto decoder.
18type Decoder struct {
19 // lastCall is last method called, either readCall or peekCall.
20 // Initial value is readCall.
21 lastCall call
22
23 // lastToken contains the last read token.
24 lastToken Token
25
26 // lastErr contains the last read error.
27 lastErr error
28
29 // openStack is a stack containing the byte characters for MessageOpen and
30 // ListOpen kinds. The top of stack represents the message or the list that
31 // the current token is nested in. An empty stack means the current token is
32 // at the top level message. The characters '{' and '<' both represent the
33 // MessageOpen kind.
34 openStack []byte
35
36 // orig is used in reporting line and column.
37 orig []byte
38 // in contains the unconsumed input.
39 in []byte
40}
41
42// NewDecoder returns a Decoder to read the given []byte.
43func NewDecoder(b []byte) *Decoder {
44 return &Decoder{orig: b, in: b}
45}
46
47// ErrUnexpectedEOF means that EOF was encountered in the middle of the input.
48var ErrUnexpectedEOF = errors.New("%v", io.ErrUnexpectedEOF)
49
50// call specifies which Decoder method was invoked.
51type call uint8
52
53const (
54 readCall call = iota
55 peekCall
56)
57
58// Peek looks ahead and returns the next token and error without advancing a read.
59func (d *Decoder) Peek() (Token, error) {
60 defer func() { d.lastCall = peekCall }()
61 if d.lastCall == readCall {
62 d.lastToken, d.lastErr = d.Read()
63 }
64 return d.lastToken, d.lastErr
65}
66
67// Read returns the next token.
68// It will return an error if there is no valid token.
69func (d *Decoder) Read() (Token, error) {
70 defer func() { d.lastCall = readCall }()
71 if d.lastCall == peekCall {
72 return d.lastToken, d.lastErr
73 }
74
75 tok, err := d.parseNext(d.lastToken.kind)
76 if err != nil {
77 return Token{}, err
78 }
79
80 switch tok.kind {
81 case comma, semicolon:
82 tok, err = d.parseNext(tok.kind)
83 if err != nil {
84 return Token{}, err
85 }
86 }
87 d.lastToken = tok
88 return tok, nil
89}
90
91const (
92 mismatchedFmt = "mismatched close character %q"
93 unexpectedFmt = "unexpected character %q"
94)
95
96// parseNext parses the next Token based on given last kind.
97func (d *Decoder) parseNext(lastKind Kind) (Token, error) {
98 // Trim leading spaces.
99 d.consume(0)
100 isEOF := false
101 if len(d.in) == 0 {
102 isEOF = true
103 }
104
105 switch lastKind {
106 case EOF:
107 return d.consumeToken(EOF, 0, 0), nil
108
109 case bof:
110 // Start of top level message. Next token can be EOF or Name.
111 if isEOF {
112 return d.consumeToken(EOF, 0, 0), nil
113 }
114 return d.parseFieldName()
115
116 case Name:
117 // Next token can be MessageOpen, ListOpen or Scalar.
118 if isEOF {
119 return Token{}, ErrUnexpectedEOF
120 }
121 switch ch := d.in[0]; ch {
122 case '{', '<':
123 d.pushOpenStack(ch)
124 return d.consumeToken(MessageOpen, 1, 0), nil
125 case '[':
126 d.pushOpenStack(ch)
127 return d.consumeToken(ListOpen, 1, 0), nil
128 default:
129 return d.parseScalar()
130 }
131
132 case Scalar:
133 openKind, closeCh := d.currentOpenKind()
134 switch openKind {
135 case bof:
136 // Top level message.
137 // Next token can be EOF, comma, semicolon or Name.
138 if isEOF {
139 return d.consumeToken(EOF, 0, 0), nil
140 }
141 switch d.in[0] {
142 case ',':
143 return d.consumeToken(comma, 1, 0), nil
144 case ';':
145 return d.consumeToken(semicolon, 1, 0), nil
146 default:
147 return d.parseFieldName()
148 }
149
150 case MessageOpen:
151 // Next token can be MessageClose, comma, semicolon or Name.
152 if isEOF {
153 return Token{}, ErrUnexpectedEOF
154 }
155 switch ch := d.in[0]; ch {
156 case closeCh:
157 d.popOpenStack()
158 return d.consumeToken(MessageClose, 1, 0), nil
159 case otherCloseChar[closeCh]:
160 return Token{}, d.newSyntaxError(mismatchedFmt, ch)
161 case ',':
162 return d.consumeToken(comma, 1, 0), nil
163 case ';':
164 return d.consumeToken(semicolon, 1, 0), nil
165 default:
166 return d.parseFieldName()
167 }
168
169 case ListOpen:
170 // Next token can be ListClose or comma.
171 if isEOF {
172 return Token{}, ErrUnexpectedEOF
173 }
174 switch ch := d.in[0]; ch {
175 case ']':
176 d.popOpenStack()
177 return d.consumeToken(ListClose, 1, 0), nil
178 case ',':
179 return d.consumeToken(comma, 1, 0), nil
180 default:
181 return Token{}, d.newSyntaxError(unexpectedFmt, ch)
182 }
183 }
184
185 case MessageOpen:
186 // Next token can be MessageClose or Name.
187 if isEOF {
188 return Token{}, ErrUnexpectedEOF
189 }
190 _, closeCh := d.currentOpenKind()
191 switch ch := d.in[0]; ch {
192 case closeCh:
193 d.popOpenStack()
194 return d.consumeToken(MessageClose, 1, 0), nil
195 case otherCloseChar[closeCh]:
196 return Token{}, d.newSyntaxError(mismatchedFmt, ch)
197 default:
198 return d.parseFieldName()
199 }
200
201 case MessageClose:
202 openKind, closeCh := d.currentOpenKind()
203 switch openKind {
204 case bof:
205 // Top level message.
206 // Next token can be EOF, comma, semicolon or Name.
207 if isEOF {
208 return d.consumeToken(EOF, 0, 0), nil
209 }
210 switch ch := d.in[0]; ch {
211 case ',':
212 return d.consumeToken(comma, 1, 0), nil
213 case ';':
214 return d.consumeToken(semicolon, 1, 0), nil
215 default:
216 return d.parseFieldName()
217 }
218
219 case MessageOpen:
220 // Next token can be MessageClose, comma, semicolon or Name.
221 if isEOF {
222 return Token{}, ErrUnexpectedEOF
223 }
224 switch ch := d.in[0]; ch {
225 case closeCh:
226 d.popOpenStack()
227 return d.consumeToken(MessageClose, 1, 0), nil
228 case otherCloseChar[closeCh]:
229 return Token{}, d.newSyntaxError(mismatchedFmt, ch)
230 case ',':
231 return d.consumeToken(comma, 1, 0), nil
232 case ';':
233 return d.consumeToken(semicolon, 1, 0), nil
234 default:
235 return d.parseFieldName()
236 }
237
238 case ListOpen:
239 // Next token can be ListClose or comma
240 if isEOF {
241 return Token{}, ErrUnexpectedEOF
242 }
243 switch ch := d.in[0]; ch {
244 case closeCh:
245 d.popOpenStack()
246 return d.consumeToken(ListClose, 1, 0), nil
247 case ',':
248 return d.consumeToken(comma, 1, 0), nil
249 default:
250 return Token{}, d.newSyntaxError(unexpectedFmt, ch)
251 }
252 }
253
254 case ListOpen:
255 // Next token can be ListClose, MessageStart or Scalar.
256 if isEOF {
257 return Token{}, ErrUnexpectedEOF
258 }
259 switch ch := d.in[0]; ch {
260 case ']':
261 d.popOpenStack()
262 return d.consumeToken(ListClose, 1, 0), nil
263 case '{', '<':
264 d.pushOpenStack(ch)
265 return d.consumeToken(MessageOpen, 1, 0), nil
266 default:
267 return d.parseScalar()
268 }
269
270 case ListClose:
271 openKind, closeCh := d.currentOpenKind()
272 switch openKind {
273 case bof:
274 // Top level message.
275 // Next token can be EOF, comma, semicolon or Name.
276 if isEOF {
277 return d.consumeToken(EOF, 0, 0), nil
278 }
279 switch ch := d.in[0]; ch {
280 case ',':
281 return d.consumeToken(comma, 1, 0), nil
282 case ';':
283 return d.consumeToken(semicolon, 1, 0), nil
284 default:
285 return d.parseFieldName()
286 }
287
288 case MessageOpen:
289 // Next token can be MessageClose, comma, semicolon or Name.
290 if isEOF {
291 return Token{}, ErrUnexpectedEOF
292 }
293 switch ch := d.in[0]; ch {
294 case closeCh:
295 d.popOpenStack()
296 return d.consumeToken(MessageClose, 1, 0), nil
297 case otherCloseChar[closeCh]:
298 return Token{}, d.newSyntaxError(mismatchedFmt, ch)
299 case ',':
300 return d.consumeToken(comma, 1, 0), nil
301 case ';':
302 return d.consumeToken(semicolon, 1, 0), nil
303 default:
304 return d.parseFieldName()
305 }
306
307 default:
308 // It is not possible to have this case. Let it panic below.
309 }
310
311 case comma, semicolon:
312 openKind, closeCh := d.currentOpenKind()
313 switch openKind {
314 case bof:
315 // Top level message. Next token can be EOF or Name.
316 if isEOF {
317 return d.consumeToken(EOF, 0, 0), nil
318 }
319 return d.parseFieldName()
320
321 case MessageOpen:
322 // Next token can be MessageClose or Name.
323 if isEOF {
324 return Token{}, ErrUnexpectedEOF
325 }
326 switch ch := d.in[0]; ch {
327 case closeCh:
328 d.popOpenStack()
329 return d.consumeToken(MessageClose, 1, 0), nil
330 case otherCloseChar[closeCh]:
331 return Token{}, d.newSyntaxError(mismatchedFmt, ch)
332 default:
333 return d.parseFieldName()
334 }
335
336 case ListOpen:
337 if lastKind == semicolon {
338 // It is not be possible to have this case as logic here
339 // should not have produced a semicolon Token when inside a
340 // list. Let it panic below.
341 break
342 }
343 // Next token can be MessageOpen or Scalar.
344 if isEOF {
345 return Token{}, ErrUnexpectedEOF
346 }
347 switch ch := d.in[0]; ch {
348 case '{', '<':
349 d.pushOpenStack(ch)
350 return d.consumeToken(MessageOpen, 1, 0), nil
351 default:
352 return d.parseScalar()
353 }
354 }
355 }
356
357 line, column := d.Position(len(d.orig) - len(d.in))
358 panic(fmt.Sprintf("Decoder.parseNext: bug at handling line %d:%d with lastKind=%v", line, column, lastKind))
359}
360
361var otherCloseChar = map[byte]byte{
362 '}': '>',
363 '>': '}',
364}
365
366// currentOpenKind indicates whether current position is inside a message, list
367// or top-level message by returning MessageOpen, ListOpen or bof respectively.
368// If the returned kind is either a MessageOpen or ListOpen, it also returns the
369// corresponding closing character.
370func (d *Decoder) currentOpenKind() (Kind, byte) {
371 if len(d.openStack) == 0 {
372 return bof, 0
373 }
374 openCh := d.openStack[len(d.openStack)-1]
375 switch openCh {
376 case '{':
377 return MessageOpen, '}'
378 case '<':
379 return MessageOpen, '>'
380 case '[':
381 return ListOpen, ']'
382 }
383 panic(fmt.Sprintf("Decoder: openStack contains invalid byte %c", openCh))
384}
385
386func (d *Decoder) pushOpenStack(ch byte) {
387 d.openStack = append(d.openStack, ch)
388}
389
390func (d *Decoder) popOpenStack() {
391 d.openStack = d.openStack[:len(d.openStack)-1]
392}
393
394// parseFieldName parses field name and separator.
395func (d *Decoder) parseFieldName() (tok Token, err error) {
396 defer func() {
397 if err == nil && d.tryConsumeChar(':') {
398 tok.attrs |= hasSeparator
399 }
400 }()
401
402 // Extension or Any type URL.
403 if d.in[0] == '[' {
404 return d.parseTypeName()
405 }
406
407 // Identifier.
408 if size := parseIdent(d.in, false); size > 0 {
409 return d.consumeToken(Name, size, uint8(IdentName)), nil
410 }
411
412 // Field number. Identify if input is a valid number that is not negative
413 // and is decimal integer within 32-bit range.
414 if num := parseNumber(d.in); num.size > 0 {
415 str := num.string(d.in)
416 if !num.neg && num.kind == numDec {
417 if _, err := strconv.ParseInt(str, 10, 32); err == nil {
418 return d.consumeToken(Name, num.size, uint8(FieldNumber)), nil
419 }
420 }
421 return Token{}, d.newSyntaxError("invalid field number: %s", str)
422 }
423
424 return Token{}, d.newSyntaxError("invalid field name: %s", errId(d.in))
425}
426
427// parseTypeName parses Any type URL or extension field name. The name is
428// enclosed in [ and ] characters. The C++ parser does not handle many legal URL
429// strings. This implementation is more liberal and allows for the pattern
430// ^[-_a-zA-Z0-9]+([./][-_a-zA-Z0-9]+)*`). Whitespaces and comments are allowed
431// in between [ ], '.', '/' and the sub names.
432func (d *Decoder) parseTypeName() (Token, error) {
433 startPos := len(d.orig) - len(d.in)
434 // Use alias s to advance first in order to use d.in for error handling.
435 // Caller already checks for [ as first character.
436 s := consume(d.in[1:], 0)
437 if len(s) == 0 {
438 return Token{}, ErrUnexpectedEOF
439 }
440
441 var name []byte
442 for len(s) > 0 && isTypeNameChar(s[0]) {
443 name = append(name, s[0])
444 s = s[1:]
445 }
446 s = consume(s, 0)
447
448 var closed bool
449 for len(s) > 0 && !closed {
450 switch {
451 case s[0] == ']':
452 s = s[1:]
453 closed = true
454
455 case s[0] == '/', s[0] == '.':
456 if len(name) > 0 && (name[len(name)-1] == '/' || name[len(name)-1] == '.') {
457 return Token{}, d.newSyntaxError("invalid type URL/extension field name: %s",
458 d.orig[startPos:len(d.orig)-len(s)+1])
459 }
460 name = append(name, s[0])
461 s = s[1:]
462 s = consume(s, 0)
463 for len(s) > 0 && isTypeNameChar(s[0]) {
464 name = append(name, s[0])
465 s = s[1:]
466 }
467 s = consume(s, 0)
468
469 default:
470 return Token{}, d.newSyntaxError(
471 "invalid type URL/extension field name: %s", d.orig[startPos:len(d.orig)-len(s)+1])
472 }
473 }
474
475 if !closed {
476 return Token{}, ErrUnexpectedEOF
477 }
478
479 // First character cannot be '.'. Last character cannot be '.' or '/'.
480 size := len(name)
481 if size == 0 || name[0] == '.' || name[size-1] == '.' || name[size-1] == '/' {
482 return Token{}, d.newSyntaxError("invalid type URL/extension field name: %s",
483 d.orig[startPos:len(d.orig)-len(s)])
484 }
485
486 d.in = s
487 endPos := len(d.orig) - len(d.in)
488 d.consume(0)
489
490 return Token{
491 kind: Name,
492 attrs: uint8(TypeName),
493 pos: startPos,
494 raw: d.orig[startPos:endPos],
495 str: string(name),
496 }, nil
497}
498
499func isTypeNameChar(b byte) bool {
500 return (b == '-' || b == '_' ||
501 ('0' <= b && b <= '9') ||
502 ('a' <= b && b <= 'z') ||
503 ('A' <= b && b <= 'Z'))
504}
505
506func isWhiteSpace(b byte) bool {
507 switch b {
508 case ' ', '\n', '\r', '\t':
509 return true
510 default:
511 return false
512 }
513}
514
515// parseIdent parses an unquoted proto identifier and returns size.
516// If allowNeg is true, it allows '-' to be the first character in the
517// identifier. This is used when parsing literal values like -infinity, etc.
518// Regular expression matches an identifier: `^[_a-zA-Z][_a-zA-Z0-9]*`
519func parseIdent(input []byte, allowNeg bool) int {
520 var size int
521
522 s := input
523 if len(s) == 0 {
524 return 0
525 }
526
527 if allowNeg && s[0] == '-' {
528 s = s[1:]
529 size++
530 if len(s) == 0 {
531 return 0
532 }
533 }
534
535 switch {
536 case s[0] == '_',
537 'a' <= s[0] && s[0] <= 'z',
538 'A' <= s[0] && s[0] <= 'Z':
539 s = s[1:]
540 size++
541 default:
542 return 0
543 }
544
545 for len(s) > 0 && (s[0] == '_' ||
546 'a' <= s[0] && s[0] <= 'z' ||
547 'A' <= s[0] && s[0] <= 'Z' ||
548 '0' <= s[0] && s[0] <= '9') {
549 s = s[1:]
550 size++
551 }
552
553 if len(s) > 0 && !isDelim(s[0]) {
554 return 0
555 }
556
557 return size
558}
559
560// parseScalar parses for a string, literal or number value.
561func (d *Decoder) parseScalar() (Token, error) {
562 if d.in[0] == '"' || d.in[0] == '\'' {
563 return d.parseStringValue()
564 }
565
566 if tok, ok := d.parseLiteralValue(); ok {
567 return tok, nil
568 }
569
570 if tok, ok := d.parseNumberValue(); ok {
571 return tok, nil
572 }
573
574 return Token{}, d.newSyntaxError("invalid scalar value: %s", errId(d.in))
575}
576
577// parseLiteralValue parses a literal value. A literal value is used for
578// bools, special floats and enums. This function simply identifies that the
579// field value is a literal.
580func (d *Decoder) parseLiteralValue() (Token, bool) {
581 size := parseIdent(d.in, true)
582 if size == 0 {
583 return Token{}, false
584 }
585 return d.consumeToken(Scalar, size, literalValue), true
586}
587
588// consumeToken constructs a Token for given Kind from d.in and consumes given
589// size-length from it.
590func (d *Decoder) consumeToken(kind Kind, size int, attrs uint8) Token {
591 // Important to compute raw and pos before consuming.
592 tok := Token{
593 kind: kind,
594 attrs: attrs,
595 pos: len(d.orig) - len(d.in),
596 raw: d.in[:size],
597 }
598 d.consume(size)
599 return tok
600}
601
602// newSyntaxError returns a syntax error with line and column information for
603// current position.
604func (d *Decoder) newSyntaxError(f string, x ...any) error {
605 e := errors.New(f, x...)
606 line, column := d.Position(len(d.orig) - len(d.in))
607 return errors.New("syntax error (line %d:%d): %v", line, column, e)
608}
609
610// Position returns line and column number of given index of the original input.
611// It will panic if index is out of range.
612func (d *Decoder) Position(idx int) (line int, column int) {
613 b := d.orig[:idx]
614 line = bytes.Count(b, []byte("\n")) + 1
615 if i := bytes.LastIndexByte(b, '\n'); i >= 0 {
616 b = b[i+1:]
617 }
618 column = utf8.RuneCount(b) + 1 // ignore multi-rune characters
619 return line, column
620}
621
622func (d *Decoder) tryConsumeChar(c byte) bool {
623 if len(d.in) > 0 && d.in[0] == c {
624 d.consume(1)
625 return true
626 }
627 return false
628}
629
630// consume consumes n bytes of input and any subsequent whitespace or comments.
631func (d *Decoder) consume(n int) {
632 d.in = consume(d.in, n)
633 return
634}
635
636// consume consumes n bytes of input and any subsequent whitespace or comments.
637func consume(b []byte, n int) []byte {
638 b = b[n:]
639 for len(b) > 0 {
640 switch b[0] {
641 case ' ', '\n', '\r', '\t':
642 b = b[1:]
643 case '#':
644 if i := bytes.IndexByte(b, '\n'); i >= 0 {
645 b = b[i+len("\n"):]
646 } else {
647 b = nil
648 }
649 default:
650 return b
651 }
652 }
653 return b
654}
655
656// errId extracts a byte sequence that looks like an invalid ID
657// (for the purposes of error reporting).
658func errId(seq []byte) []byte {
659 const maxLen = 32
660 for i := 0; i < len(seq); {
661 if i > maxLen {
662 return append(seq[:i:i], "…"...)
663 }
664 r, size := utf8.DecodeRune(seq[i:])
665 if r > utf8.RuneSelf || (r != '/' && isDelim(byte(r))) {
666 if i == 0 {
667 // Either the first byte is invalid UTF-8 or a
668 // delimiter, or the first rune is non-ASCII.
669 // Return it as-is.
670 i = size
671 }
672 return seq[:i:i]
673 }
674 i += size
675 }
676 // No delimiter found.
677 return seq
678}
679
680// isDelim returns true if given byte is a delimiter character.
681func isDelim(c byte) bool {
682 return !(c == '-' || c == '+' || c == '.' || c == '_' ||
683 ('a' <= c && c <= 'z') ||
684 ('A' <= c && c <= 'Z') ||
685 ('0' <= c && c <= '9'))
686}