1 | """data_lang/htm8.py
|
2 |
|
3 | TODO
|
4 |
|
5 | API:
|
6 | - Get rid of AttrValueLexer - this should be in the TagLexer
|
7 | - this also means that unquoted values can be more similar
|
8 | - We can use a single lexer mode for everything inside <>
|
9 | - the SPACE is the only difference
|
10 | - Deprecate tag_lexer.GetTagName() in favor of lx.CanonicalTagName() or
|
11 | _LiteralTagName()
|
12 | - UTF-8 check, like JSON8
|
13 | - re2c
|
14 | - port lexer, which will fix static typing issues
|
15 | - the abstraction needs to support submatch?
|
16 | - for finding the end of a tag, etc.?
|
17 |
|
18 | - LexError and ParseError need details
|
19 | - harmonize with data_lang/j8.py, which uses error.Decode(msg, ...,
|
20 | cur_line_num)
|
21 |
|
22 | - Copy all errors into doc/ref/chap-errors.md
|
23 | - This helps understand the language
|
24 |
|
25 | - Update doc/htm8.md
|
26 | - list of Algorithms:
|
27 | - lex just the top level
|
28 | - lex both levels
|
29 | - and match tags - this is the level for value.Htm8Frag?
|
30 | - convert to XML!
|
31 | - lazy selection by tag, or attr (id= and class=)
|
32 | - lazy selection by CSS selector expression
|
33 | - convert to DOMTree
|
34 | - sed-like replacement of DOM Tree or element
|
35 | - untrusted HTML filter, e.g. like StackOverflow / Reddit
|
36 | - this is Safe HTM8
|
37 | - should have a zero alloc way to support this, with good errors?
|
38 | - I think most of them silently strip data
|
39 | """
|
40 |
|
41 | import re
|
42 |
|
43 | from typing import Dict, List, Tuple, Optional, IO, Iterator, Any
|
44 |
|
45 | from _devbuild.gen.htm8_asdl import (h8_id, h8_id_t, h8_tag_id, h8_tag_id_t,
|
46 | h8_tag_id_str, attr_name, attr_name_t,
|
47 | attr_name_str, attr_value_e, attr_value_t,
|
48 | h8_val_id)
|
49 | from doctools.util import log
|
50 |
|
51 |
|
52 | class LexError(Exception):
|
53 | """
|
54 | Examples of lex errors:
|
55 |
|
56 | - h8_id.Invalid, like <> or &&
|
57 | - Unclosed <!-- <? <![CDATA[ <script> <style>
|
58 | """
|
59 |
|
60 | def __init__(self, s, start_pos):
|
61 | # type: (str, int) -> None
|
62 | self.s = s
|
63 | self.start_pos = start_pos
|
64 |
|
65 | def __str__(self):
|
66 | # type: () -> str
|
67 | return '(LexError %r)' % (self.s[self.start_pos:self.start_pos + 20])
|
68 |
|
69 |
|
70 | def _FindLineNum(s, error_pos):
|
71 | # type: (str, int) -> int
|
72 | current_pos = 0
|
73 | line_num = 1
|
74 | while True:
|
75 | newline_pos = s.find('\n', current_pos)
|
76 | #log('current = %d, N %d, line %d', current_pos, newline_pos, line_num)
|
77 |
|
78 | if newline_pos == -1: # this is the last line
|
79 | return line_num
|
80 | if newline_pos >= error_pos:
|
81 | return line_num
|
82 | line_num += 1
|
83 | current_pos = newline_pos + 1
|
84 |
|
85 |
|
86 | class ParseError(Exception):
|
87 | """
|
88 | Examples of parse errors
|
89 |
|
90 | - unbalanced tag structure
|
91 | - ul_table.py errors
|
92 | """
|
93 |
|
94 | def __init__(self, msg, s=None, start_pos=-1):
|
95 | # type: (str, Optional[str], int) -> None
|
96 | self.msg = msg
|
97 | self.s = s
|
98 | self.start_pos = start_pos
|
99 |
|
100 | def __str__(self):
|
101 | # type: () -> str
|
102 | if self.s is not None:
|
103 | assert self.start_pos != -1, self.start_pos
|
104 | snippet = (self.s[self.start_pos:self.start_pos + 20])
|
105 |
|
106 | line_num = _FindLineNum(self.s, self.start_pos)
|
107 | else:
|
108 | snippet = ''
|
109 | line_num = -1
|
110 | msg = 'line %d: %r %r' % (line_num, self.msg, snippet)
|
111 | return msg
|
112 |
|
113 |
|
114 | class Output(object):
|
115 | """Takes an underlying input buffer and an output file. Maintains a
|
116 | position in the input buffer.
|
117 |
|
118 | Print FROM the input or print new text to the output.
|
119 | """
|
120 |
|
121 | def __init__(self, s, f, left_pos=0, right_pos=-1):
|
122 | # type: (str, IO[str], int, int) -> None
|
123 | self.s = s
|
124 | self.f = f
|
125 | self.pos = left_pos
|
126 | self.right_pos = len(s) if right_pos == -1 else right_pos
|
127 |
|
128 | def SkipTo(self, pos):
|
129 | # type: (int) -> None
|
130 | """Skip to a position."""
|
131 | self.pos = pos
|
132 |
|
133 | def PrintUntil(self, pos):
|
134 | # type: (int) -> None
|
135 | """Print until a position."""
|
136 | piece = self.s[self.pos:pos]
|
137 | self.f.write(piece)
|
138 | self.pos = pos
|
139 |
|
140 | def PrintTheRest(self):
|
141 | # type: () -> None
|
142 | """Print until the end of the string."""
|
143 | self.PrintUntil(self.right_pos)
|
144 |
|
145 | def Print(self, s):
|
146 | # type: (str) -> None
|
147 | """Print text to the underlying buffer."""
|
148 | self.f.write(s)
|
149 |
|
150 |
|
151 | def MakeLexer(rules):
|
152 | return [(re.compile(pat, re.VERBOSE), i) for (pat, i) in rules]
|
153 |
|
154 |
|
155 | #
|
156 | # Eggex
|
157 | #
|
158 | # Tag = / ~['>']+ /
|
159 |
|
160 | # Is this valid? A single character?
|
161 | # Tag = / ~'>'* /
|
162 |
|
163 | # Maybe better: / [NOT '>']+/
|
164 | # capital letters not allowed there?
|
165 | #
|
166 | # But then this is confusing:
|
167 | # / [NOT ~digit]+/
|
168 | #
|
169 | # / [NOT digit] / is [^\d]
|
170 | # / ~digit / is \D
|
171 | #
|
172 | # Or maybe:
|
173 | #
|
174 | # / [~ digit]+ /
|
175 | # / [~ '>']+ /
|
176 | # / [NOT '>']+ /
|
177 |
|
178 | # End = / '</' Tag '>' /
|
179 | # StartEnd = / '<' Tag '/>' /
|
180 | # Start = / '<' Tag '>' /
|
181 | #
|
182 | # EntityRef = / '&' dot{* N} ';' /
|
183 |
|
184 | # Tag name, or attribute name
|
185 | # colon is used in XML
|
186 |
|
187 | # https://www.w3.org/TR/xml/#NT-Name
|
188 | # Hm there is a lot of unicode stuff. We are simplifying parsing
|
189 |
|
190 | _NAME = r'[a-zA-Z][a-zA-Z0-9:_\-]*' # must start with letter
|
191 |
|
192 | CHAR_LEX = [
|
193 | # Characters
|
194 | # https://www.w3.org/TR/xml/#sec-references
|
195 | (r'&\# [0-9]+ ;', h8_id.DecChar),
|
196 | (r'&\# x[0-9a-fA-F]+ ;', h8_id.HexChar),
|
197 | (r'& %s ;' % _NAME, h8_id.CharEntity),
|
198 | # Allow unquoted, and quoted
|
199 | (r'&', h8_id.BadAmpersand),
|
200 | ]
|
201 |
|
202 | HTM8_LEX = CHAR_LEX + [
|
203 | # TODO: CommentBegin, ProcessingBegin, CDataBegin could have an additional
|
204 | # action associated with them? The ending substring
|
205 | (r'<!--', h8_id.CommentBegin),
|
206 |
|
207 | # Processing instruction are used for the XML header:
|
208 | # <?xml version="1.0" encoding="UTF-8"?>
|
209 | # They are technically XML-only, but in HTML5, they are another kind of
|
210 | # comment:
|
211 | #
|
212 | # https://developer.mozilla.org/en-US/docs/Web/API/ProcessingInstruction
|
213 | #
|
214 | (r'<\?', h8_id.ProcessingBegin),
|
215 | # Not necessary in HTML5, but occurs in XML
|
216 | (r'<!\[CDATA\[', h8_id.CDataBegin), # <![CDATA[
|
217 |
|
218 | # Markup declarations
|
219 | # - In HTML5, there is only <!DOCTYPE html>
|
220 | # - XML has 4 more declarations: <!ELEMENT ...> ATTLIST ENTITY NOTATION
|
221 | # - these seem to be part of DTD
|
222 | # - it's useful to skip these, and be able to parse the rest of the document
|
223 | # - Note: < is allowed?
|
224 | (r'<! [^>\x00]+ >', h8_id.Decl),
|
225 |
|
226 | # Tags
|
227 | # Notes:
|
228 | # - We look for a valid tag name, but we don't validate attributes.
|
229 | # That's done in the tag lexer.
|
230 | # - We don't allow leading whitespace
|
231 | (r'</ (%s) >' % _NAME, h8_id.EndTag),
|
232 | # self-closing <br/> comes before StartTag
|
233 | # could/should these be collapsed into one rule?
|
234 | (r'< (%s) [^>\x00]* />' % _NAME, h8_id.StartEndTag), # end </a>
|
235 | (r'< (%s) [^>\x00]* >' % _NAME, h8_id.StartTag), # start <a>
|
236 |
|
237 | # HTML5 allows unescaped > in raw data, but < is not allowed.
|
238 | # https://stackoverflow.com/questions/10462348/right-angle-bracket-in-html
|
239 | #
|
240 | # - My early blog has THREE errors when disallowing >
|
241 | # - So do some .wwz files
|
242 | (r'[^&<>\x00]+', h8_id.RawData),
|
243 | (r'>', h8_id.BadGreaterThan),
|
244 | # NUL is the end, an accomodation for re2c. Like we do in frontend/match.
|
245 | (r'\x00', h8_id.EndOfStream),
|
246 | # This includes < - it is not BadLessThan because it's NOT recoverable
|
247 | (r'.', h8_id.Invalid),
|
248 | ]
|
249 |
|
250 | # Old notes:
|
251 | #
|
252 | # Non-greedy matches are regular and can be matched in linear time
|
253 | # with RE2.
|
254 | #
|
255 | # https://news.ycombinator.com/item?id=27099798
|
256 | #
|
257 |
|
258 | # This person tried to do it with a regex:
|
259 | #
|
260 | # https://skeptric.com/html-comment-regexp/index.html
|
261 |
|
262 | # . is any char except newline
|
263 | # https://re2c.org/manual/manual_c.html
|
264 |
|
265 | # Discarded options
|
266 | #(r'<!-- .*? -->', h8_id.Comment),
|
267 |
|
268 | # Hack from Claude: \s\S instead of re.DOTALL. I don't like this
|
269 | #(r'<!-- [\s\S]*? -->', h8_id.Comment),
|
270 | #(r'<!-- (?:.|[\n])*? -->', h8_id.Comment),
|
271 |
|
272 | HTM8_LEX_COMPILED = MakeLexer(HTM8_LEX)
|
273 |
|
274 |
|
275 | class Lexer(object):
|
276 |
|
277 | def __init__(self, s, left_pos=0, right_pos=-1, no_special_tags=False):
|
278 | # type: (str, int, int, bool) -> None
|
279 | self.s = s
|
280 | self.pos = left_pos
|
281 | self.right_pos = len(s) if right_pos == -1 else right_pos
|
282 | self.no_special_tags = no_special_tags
|
283 |
|
284 | # string -> compiled regex pattern object
|
285 | self.cache = {} # type: Dict[str, Any]
|
286 |
|
287 | # either </script> or </style> - we search until we see that
|
288 | self.search_state = None # type: Optional[str]
|
289 |
|
290 | # Position of tag name, if applicable
|
291 | # - Set after you get a StartTag, EndTag, or StartEndTag
|
292 | # - Unset on other tags
|
293 | self.tag_pos_left = -1
|
294 | self.tag_pos_right = -1
|
295 |
|
296 | def _Read(self):
|
297 | # type: () -> Tuple[h8_id_t, int]
|
298 | if self.pos == self.right_pos:
|
299 | return h8_id.EndOfStream, self.pos
|
300 |
|
301 | assert self.pos < self.right_pos, self.pos
|
302 |
|
303 | if self.search_state is not None and not self.no_special_tags:
|
304 | # TODO: case-insensitive search for </SCRIPT> <SCRipt> ?
|
305 | #
|
306 | # Another strategy: enter a mode where we find ONLY the end tag
|
307 | # regex, and any data that's not <, and then check the canonical
|
308 | # tag name for 'script' or 'style'.
|
309 | pos = self.s.find(self.search_state, self.pos)
|
310 | if pos == -1:
|
311 | # unterminated <script> or <style>
|
312 | raise LexError(self.s, self.pos)
|
313 | self.search_state = None
|
314 | # beginning
|
315 | return h8_id.HtmlCData, pos
|
316 |
|
317 | # Find the first match.
|
318 | # Note: frontend/match.py uses _LongestMatch(), which is different!
|
319 | # TODO: reconcile them. This lexer should be expressible in re2c.
|
320 |
|
321 | for pat, tok_id in HTM8_LEX_COMPILED:
|
322 | m = pat.match(self.s, self.pos)
|
323 | if m:
|
324 | if tok_id in (h8_id.StartTag, h8_id.EndTag, h8_id.StartEndTag):
|
325 | self.tag_pos_left = m.start(1)
|
326 | self.tag_pos_right = m.end(1)
|
327 | else:
|
328 | # Reset state
|
329 | self.tag_pos_left = -1
|
330 | self.tag_pos_right = -1
|
331 |
|
332 | if tok_id == h8_id.CommentBegin:
|
333 | pos = self.s.find('-->', self.pos)
|
334 | if pos == -1:
|
335 | # unterminated <!--
|
336 | raise LexError(self.s, self.pos)
|
337 | return h8_id.Comment, pos + 3 # -->
|
338 |
|
339 | if tok_id == h8_id.ProcessingBegin:
|
340 | pos = self.s.find('?>', self.pos)
|
341 | if pos == -1:
|
342 | # unterminated <?
|
343 | raise LexError(self.s, self.pos)
|
344 | return h8_id.Processing, pos + 2 # ?>
|
345 |
|
346 | if tok_id == h8_id.CDataBegin:
|
347 | pos = self.s.find(']]>', self.pos)
|
348 | if pos == -1:
|
349 | # unterminated <![CDATA[
|
350 | raise LexError(self.s, self.pos)
|
351 | return h8_id.CData, pos + 3 # ]]>
|
352 |
|
353 | if tok_id == h8_id.StartTag:
|
354 | # TODO: reduce allocations
|
355 | if (self.TagNameEquals('script') or
|
356 | self.TagNameEquals('style')):
|
357 | # <SCRipt a=b> -> </SCRipt>
|
358 | self.search_state = '</' + self._LiteralTagName() + '>'
|
359 |
|
360 | return tok_id, m.end()
|
361 | else:
|
362 | raise AssertionError('h8_id.Invalid rule should have matched')
|
363 |
|
364 | def TagNamePos(self):
|
365 | """The right position of the tag pos"""
|
366 | assert self.tag_pos_right != -1, self.tag_pos_right
|
367 | return self.tag_pos_right
|
368 |
|
369 | def TagNameEquals(self, expected):
|
370 | # type: (str) -> bool
|
371 | assert self.tag_pos_left != -1, self.tag_pos_left
|
372 | assert self.tag_pos_right != -1, self.tag_pos_right
|
373 |
|
374 | # TODO: In C++, this does not need an allocation. Can we test
|
375 | # directly?
|
376 | return expected == self.CanonicalTagName()
|
377 |
|
378 | def _LiteralTagName(self):
|
379 | # type: () -> str
|
380 | assert self.tag_pos_left != -1, self.tag_pos_left
|
381 | assert self.tag_pos_right != -1, self.tag_pos_right
|
382 |
|
383 | return self.s[self.tag_pos_left:self.tag_pos_right]
|
384 |
|
385 | def CanonicalTagName(self):
|
386 | # type: () -> str
|
387 | tag_name = self._LiteralTagName()
|
388 | # Most tags are already lower case, so avoid allocation with this conditional
|
389 | # TODO: this could go in the mycpp runtime?
|
390 | if tag_name.islower():
|
391 | return tag_name
|
392 | else:
|
393 | return tag_name.lower()
|
394 |
|
395 | def Read(self):
|
396 | # type: () -> Tuple[h8_id_t, int]
|
397 | tok_id, end_pos = self._Read()
|
398 | self.pos = end_pos # advance
|
399 | return tok_id, end_pos
|
400 |
|
401 | def LookAhead(self, regex):
|
402 | # type: (str) -> bool
|
403 | """
|
404 | Currently used for ul_table.py. But taking a dynamic regex string is
|
405 | not the right interface.
|
406 | """
|
407 | # Cache the regex compilation. This could also be LookAheadFor(THEAD)
|
408 | # or something.
|
409 | pat = self.cache.get(regex)
|
410 | if pat is None:
|
411 | pat = re.compile(regex)
|
412 | self.cache[regex] = pat
|
413 |
|
414 | m = pat.match(self.s, self.pos)
|
415 | return m is not None
|
416 |
|
417 |
|
418 | A_NAME_LEX = [
|
419 | # Leading whitespace is required, to separate attributes.
|
420 | #
|
421 | # If the = is not present, then we set the lexer in a state for
|
422 | # attr_value_e.Missing.
|
423 | (r'\s+ (%s) \s* (=)? \s*' % _NAME, attr_name.Ok),
|
424 | # unexpected EOF
|
425 |
|
426 | # The closing > or /> is treated as end of stream, and it's not an error.
|
427 | (r'\s* /? >', attr_name.Done),
|
428 |
|
429 | # NUL should not be possible, because the top-level
|
430 |
|
431 | # This includes < - it is not BadLessThan because it's NOT recoverable
|
432 | (r'.', attr_name.Invalid),
|
433 | ]
|
434 |
|
435 | A_NAME_LEX_COMPILED = MakeLexer(A_NAME_LEX)
|
436 |
|
437 | # Here we just loop on regular tokens
|
438 | #
|
439 | # Examples:
|
440 | # <a href = unquoted&foo >
|
441 | # <a href = unquoted&foo > # BadAmpersand is allowed I guess
|
442 | # <a href ="unquoted&foo" > # double quoted
|
443 | # <a href ='unquoted&foo' > # single quoted
|
444 | # <a href = what"foo" > # HTML5 allows this, but we could disallow it if
|
445 | # it's not common. It opens up the j"" and $"" extensions
|
446 | # <a href = what'foo' > # ditto
|
447 | #
|
448 | # Problem: <a href=foo/> - this is hard to recognize
|
449 | # Because is the unquoted value "foo/" or "foo" ?
|
450 |
|
451 | # Be very lenient - just no whitespace or special HTML chars
|
452 | # I don't think this is more lenient than HTML5, though we should check.
|
453 | #
|
454 | # Bug fix: Also disallow /
|
455 |
|
456 | # TODO: get rid of OLD copy
|
457 | _UNQUOTED_VALUE_OLD = r'''[^ \t\r\n<>&/"'\x00]*'''
|
458 | _UNQUOTED_VALUE = r'''[^ \t\r\n<>&/"'\x00]+'''
|
459 |
|
460 | # Restrictive definition, similar to _NAME
|
461 | # I was trying to capture #ble.sh and so forth
|
462 | # I also had unquoted //github.com, etc.
|
463 |
|
464 | # _UNQUOTED_VALUE = r'''[a-zA-Z0-9:_\-]+'''
|
465 | #
|
466 | # For now, I guess we live with <a href=?foo/>
|
467 |
|
468 | # What comes after = ?
|
469 | A_VALUE_LEX = [
|
470 | (r'"', h8_val_id.DoubleQuote),
|
471 | (r"'", h8_val_id.SingleQuote),
|
472 | (_UNQUOTED_VALUE, h8_val_id.UnquotedVal),
|
473 |
|
474 | #(r'[ \r\n\t]', h8_id.Whitespace), # terminates unquoted values
|
475 | #(r'[^ \r\n\t&>\x00]', h8_id.RawData),
|
476 | #(r'[>\x00]', h8_id.EndOfStream),
|
477 | # e.g. < is an error
|
478 | (r'.', h8_val_id.NoMatch),
|
479 | ]
|
480 |
|
481 | A_VALUE_LEX_COMPILED = MakeLexer(A_VALUE_LEX)
|
482 |
|
483 | # What's inside "" or '' ?
|
484 | QUOTED_VALUE_LEX = CHAR_LEX + [
|
485 | (r'"', h8_id.DoubleQuote),
|
486 | (r"'", h8_id.SingleQuote),
|
487 | (r'<', h8_id.BadLessThan), # BadAmpersand is in CharLex
|
488 | (r'''[^"'<>&\x00]+''', h8_id.RawData),
|
489 | # This includes > - it is not BadGreaterThan because it's NOT recoverable
|
490 | (r'.', h8_id.Invalid),
|
491 | ]
|
492 |
|
493 | QUOTED_VALUE_LEX_COMPILED = MakeLexer(QUOTED_VALUE_LEX)
|
494 |
|
495 |
|
496 | class AttrLexer(object):
|
497 | """
|
498 | Typical usage:
|
499 |
|
500 | while True:
|
501 | n, start_pos, end_pos = attr_lx.ReadName()
|
502 | if n == attr_name.Ok:
|
503 | if attr_lx.AttrNameEquals('div'):
|
504 | print('div')
|
505 |
|
506 | # TODO: also pass Optional[List[]] out_tokens?
|
507 | v, start_pos, end_pos = attr_lx.ReadValue()
|
508 | """
|
509 |
|
510 | def __init__(self, s):
|
511 | # type: (str) -> None
|
512 | self.s = s
|
513 | self.tag_name_pos = -1 # Invalid
|
514 | self.tag_end_pos = -1
|
515 | self.pos = -1
|
516 |
|
517 | self.name_start = -1
|
518 | self.name_end = -1
|
519 | self.next_value_is_missing = False
|
520 |
|
521 | self.init_t = -1
|
522 | self.init_e = -1
|
523 |
|
524 | def Init(self, tag_name_pos, end_pos):
|
525 | # type: (int, int) -> None
|
526 | """Initialize so we can read names and values.
|
527 |
|
528 | Example:
|
529 | 'x <a y>' # tag_name_pos=4, end_pos=6
|
530 | 'x <a>' # tag_name_pos=4, end_pos=4
|
531 |
|
532 | The Reset() method is used to reuse instances of the AttrLexer object.
|
533 | """
|
534 | assert tag_name_pos >= 0, tag_name_pos
|
535 | assert end_pos >= 0, end_pos
|
536 |
|
537 | #log('TAG NAME POS %d', tag_name_pos)
|
538 |
|
539 | self.tag_name_pos = tag_name_pos
|
540 | self.end_pos = end_pos
|
541 |
|
542 | self.pos = tag_name_pos
|
543 |
|
544 | # For Reset()
|
545 | self.init_t = tag_name_pos
|
546 | self.init_e = end_pos
|
547 |
|
548 | def Reset(self):
|
549 | self.tag_name_pos = self.init_t
|
550 | self.end_pos = self.init_e
|
551 | self.pos = self.init_t
|
552 |
|
553 | def ReadName(self):
|
554 | # type: () -> Tuple[attr_name_t, int, int]
|
555 | """Reads the attribute name
|
556 |
|
557 | EOF case:
|
558 | <a>
|
559 | <a >
|
560 |
|
561 | Error case:
|
562 | <a !>
|
563 | <a foo=bar !>
|
564 | """
|
565 | for pat, a in A_NAME_LEX_COMPILED:
|
566 | m = pat.match(self.s, self.pos)
|
567 | #log('ReadName() matching %r at %d', self.s, self.pos)
|
568 | if m:
|
569 | #log('ReadName() tag_name_pos %d pos, %d %s', self.tag_name_pos, self.pos, m.groups())
|
570 | if a == attr_name.Invalid:
|
571 | #log('m.groups %s', m.groups())
|
572 | return attr_name.Invalid, -1, -1
|
573 |
|
574 | self.pos = m.end(0) # Advance if it's not invalid
|
575 |
|
576 | if a == attr_name.Ok:
|
577 | #log('%r', m.groups())
|
578 | self.name_start = m.start(1)
|
579 | self.name_end = m.end(1)
|
580 | # Is the equals sign missing? Set state.
|
581 | if m.group(2) is None:
|
582 | self.next_value_is_missing = True
|
583 | # HACK: REWIND, since we don't want to consume whitespace
|
584 | self.pos = self.name_end
|
585 | return attr_name.Ok, self.name_start, self.name_end
|
586 | else:
|
587 | # Reset state - e.g. you must call AttrNameEquals
|
588 | self.name_start = -1
|
589 | self.name_end = -1
|
590 | self.next_value_is_missing = False
|
591 |
|
592 | if a == attr_name.Done:
|
593 | return attr_name.Done, -1, -1
|
594 | else:
|
595 | context = self.s[self.pos:]
|
596 | #log('s %r %d', self.s, self.pos)
|
597 | raise AssertionError('h8_id.Invalid rule should have matched %r' %
|
598 | context)
|
599 |
|
600 | def _CanonicalAttrName(self):
|
601 | # type: () -> str
|
602 | """Return the lower case attribute name.
|
603 |
|
604 | Must call after ReadName()
|
605 | """
|
606 | assert self.name_start >= 0, self.name_start
|
607 | assert self.name_end >= 0, self.name_end
|
608 |
|
609 | attr_name = self.s[self.name_start:self.name_end]
|
610 | if attr_name.islower():
|
611 | return attr_name
|
612 | else:
|
613 | return attr_name.lower()
|
614 |
|
615 | def AttrNameEquals(self, expected):
|
616 | # type: (str) -> bool
|
617 | """
|
618 | Must call after ReadName()
|
619 |
|
620 | TODO: This can be optimized to be "in place", with zero allocs.
|
621 | """
|
622 | return expected == self._CanonicalAttrName()
|
623 |
|
624 | def _QuotedRead(self):
|
625 | # type: () -> Tuple[h8_id_t, int]
|
626 |
|
627 | for pat, tok_id in QUOTED_VALUE_LEX_COMPILED:
|
628 | m = pat.match(self.s, self.pos)
|
629 | if m:
|
630 | end_pos = m.end(0) # Advance
|
631 | #log('_QuotedRead %r', self.s[self.pos:end_pos])
|
632 | return tok_id, end_pos
|
633 | else:
|
634 | context = self.s[self.pos:self.pos + 10]
|
635 | raise AssertionError('h8_id.Invalid rule should have matched %r' %
|
636 | context)
|
637 |
|
638 | def ReadValue(self, tokens_out=None):
|
639 | # type: (Optional[List[Tuple[h8_id, int]]]) -> Tuple[attr_value_t, int, int]
|
640 | """Read the attribute value.
|
641 |
|
642 | In general, it is escaped or "raw"
|
643 |
|
644 | Can only be called after a SUCCESSFUL ReadName().
|
645 | Assuming ReadName() returned a value, this should NOT fail.
|
646 | """
|
647 | # ReadName() invariant
|
648 | assert self.name_start >= 0, self.name_start
|
649 | assert self.name_end >= 0, self.name_end
|
650 |
|
651 | self.name_start = -1
|
652 | self.name_end = -1
|
653 |
|
654 | if self.next_value_is_missing:
|
655 | # Do not advance self.pos
|
656 | #log('-> MISSING pos %d : %r', self.pos, self.s[self.pos:])
|
657 | return attr_value_e.Missing, -1, -1
|
658 |
|
659 | # Now read " ', unquoted or empty= is valid too.
|
660 | for pat, a in A_VALUE_LEX_COMPILED:
|
661 | m = pat.match(self.s, self.pos)
|
662 | if m:
|
663 | first_end_pos = m.end(0)
|
664 | #log('m %s', m.groups())
|
665 |
|
666 | # Note: Unquoted value can't contain & etc. now, so there
|
667 | # is no unquoting, and no respecting tokens_raw.
|
668 | if a == h8_val_id.UnquotedVal:
|
669 | self.pos = first_end_pos # Advance
|
670 | return attr_value_e.Unquoted, m.start(0), first_end_pos
|
671 |
|
672 | # TODO: respect tokens_out
|
673 | if a == h8_val_id.DoubleQuote:
|
674 | self.pos = first_end_pos
|
675 | while True:
|
676 | tok_id, q_end_pos = self._QuotedRead()
|
677 | #log('self.pos %d q_end_pos %d', self.pos, q_end_pos)
|
678 | if tok_id == h8_id.Invalid:
|
679 | raise LexError(self.s, self.pos)
|
680 | if tok_id == h8_id.DoubleQuote:
|
681 | right_pos = self.pos
|
682 | self.pos = q_end_pos # Advance past "
|
683 | return attr_value_e.DoubleQuoted, first_end_pos, right_pos
|
684 | self.pos = q_end_pos # Advance _QuotedRead
|
685 |
|
686 | # TODO: respect tokens_out
|
687 | if a == h8_val_id.SingleQuote:
|
688 | self.pos = first_end_pos
|
689 | while True:
|
690 | tok_id, q_end_pos = self._QuotedRead()
|
691 | if tok_id == h8_id.Invalid:
|
692 | raise LexError(self.s, self.pos)
|
693 | if tok_id == h8_id.SingleQuote:
|
694 | right_pos = self.pos
|
695 | self.pos = q_end_pos # Advance past "
|
696 | return attr_value_e.SingleQuoted, first_end_pos, right_pos
|
697 | self.pos = q_end_pos # Advance _QuotedRead
|
698 |
|
699 | if a == h8_val_id.NoMatch:
|
700 | # <a foo = >
|
701 | return attr_value_e.Empty, -1, -1
|
702 | else:
|
703 | raise AssertionError('h8_val_id.NoMatch rule should have matched')
|
704 |
|
705 |
|
706 | def GetAttrRaw(attr_lx, name):
|
707 | # type: (AttrLexer, str) -> Optional[str]
|
708 | while True:
|
709 | n, name_start, name_end = attr_lx.ReadName()
|
710 | #log('==> ReadName %s %d %d', attr_name_str(n), name_start, name_end)
|
711 | if n == attr_name.Ok:
|
712 | if attr_lx.AttrNameEquals(name):
|
713 | v, val_start, val_end = attr_lx.ReadValue()
|
714 | return attr_lx.s[val_start:val_end]
|
715 | else:
|
716 | # Problem with stateful API: You are forced to either ReadValue()
|
717 | # or SkipVlaue()
|
718 | attr_lx.ReadValue()
|
719 | elif n == attr_name.Done:
|
720 | break
|
721 | elif n == attr_name.Invalid:
|
722 | raise LexError(attr_lx.s, attr_lx.pos)
|
723 | else:
|
724 | raise AssertionError()
|
725 |
|
726 | return None
|
727 |
|
728 |
|
729 | def AllAttrsRaw(attr_lx):
|
730 | # type: (AttrLexer) -> List[Tuple[str,str]]
|
731 | result = []
|
732 | while True:
|
733 | n, name_start, name_end = attr_lx.ReadName()
|
734 | if 0:
|
735 | log(' AllAttrsRaw ==> ReadName %s %d %d', attr_name_str(n),
|
736 | name_start, name_end)
|
737 | if n == attr_name.Ok:
|
738 | name = attr_lx.s[name_start:name_end]
|
739 | #log(' Name %r', name)
|
740 |
|
741 | v, val_start, val_end = attr_lx.ReadValue()
|
742 | val = attr_lx.s[val_start:val_end]
|
743 | #log(' ReadValue %r', val)
|
744 | result.append((name, val))
|
745 | elif n == attr_name.Done:
|
746 | break
|
747 | elif n == attr_name.Invalid:
|
748 | raise LexError(attr_lx.s, attr_lx.pos)
|
749 | else:
|
750 | raise AssertionError()
|
751 |
|
752 | return result
|
753 |
|
754 |
|
755 | #
|
756 | # OLD API - REMOVE THIS
|
757 | #
|
758 |
|
759 | # Tag names:
|
760 | # Match <a or </a
|
761 | # Match <h2, but not <2h
|
762 | #
|
763 | # HTML 5 doesn't restrict tag names at all
|
764 | # https://html.spec.whatwg.org/#toc-syntax
|
765 | #
|
766 | # XML allows : - .
|
767 | # https://www.w3.org/TR/xml/#NT-NameChar
|
768 |
|
769 | # Namespaces for MathML, SVG
|
770 | # XLink, XML, XMLNS
|
771 | #
|
772 | # https://infra.spec.whatwg.org/#namespaces
|
773 | #
|
774 | # Allow - for td-attrs
|
775 |
|
776 | # TODO: we don't need to capture the tag name here? That's done at the top
|
777 | # level
|
778 | _TAG_RE = re.compile(r'/? \s* (%s)' % _NAME, re.VERBOSE)
|
779 |
|
780 | _TAG_LAST_RE = re.compile(r'\s* /? >', re.VERBOSE)
|
781 |
|
782 | # To match href="foo"
|
783 | # Note: in HTML5 and XML, single quoted attributes are also valid
|
784 |
|
785 | # <button disabled> is standard usage
|
786 |
|
787 | # NOTE: This used to allow whitespace around =
|
788 | # <a foo = "bar"> makes sense in XML
|
789 | # But then you also have
|
790 | # <a foo= bar> - which is TWO attributes, in HTML5
|
791 | # So the space is problematic
|
792 |
|
793 | _ATTR_RE = re.compile(
|
794 | r'''
|
795 | \s+ # Leading whitespace is required
|
796 | (%s) # Attribute name
|
797 | (?: # Optional attribute value
|
798 | \s* = \s* # Spaces allowed around =
|
799 | (?:
|
800 | " ([^>"\x00]*) " # double quoted value
|
801 | | ' ([^>'\x00]*) ' # single quoted value
|
802 | | (%s) # Attribute value
|
803 | )
|
804 | )?
|
805 | ''' % (_NAME, _UNQUOTED_VALUE_OLD), re.VERBOSE)
|
806 |
|
807 |
|
808 | class TagLexer(object):
|
809 | """
|
810 | Given a tag like <a href="..."> or <link type="..." />, the TagLexer
|
811 | provides a few operations:
|
812 |
|
813 | - What is the tag?
|
814 | - Iterate through the attributes, giving (name, value_start_pos, value_end_pos)
|
815 | """
|
816 |
|
817 | def __init__(self, s):
|
818 | # type: (str) -> None
|
819 | self.s = s
|
820 | self.start_pos = -1 # Invalid
|
821 | self.end_pos = -1
|
822 |
|
823 | def Reset(self, start_pos, end_pos):
|
824 | # type: (int, int) -> None
|
825 | """Reuse instances of this object."""
|
826 | assert start_pos >= 0, start_pos
|
827 | assert end_pos >= 0, end_pos
|
828 |
|
829 | self.start_pos = start_pos
|
830 | self.end_pos = end_pos
|
831 |
|
832 | def WholeTagString(self):
|
833 | # type: () -> str
|
834 | """Return the entire tag string, e.g. <a href='foo'>"""
|
835 | return self.s[self.start_pos:self.end_pos]
|
836 |
|
837 | def GetTagName(self):
|
838 | # type: () -> str
|
839 | # First event
|
840 | tok_id, start, end = next(self.Tokens())
|
841 | return self.s[start:end]
|
842 |
|
843 | def GetSpanForAttrValue(self, attr_name):
|
844 | # type: (str) -> Tuple[int, int]
|
845 | """
|
846 | Used by oils_doc.py, for href shortcuts
|
847 | """
|
848 | # Algorithm: search for QuotedValue or UnquotedValue after AttrName
|
849 | # TODO: Could also cache these
|
850 |
|
851 | events = self.Tokens()
|
852 | val = (-1, -1)
|
853 | try:
|
854 | while True:
|
855 | tok_id, start, end = next(events)
|
856 | if tok_id == h8_tag_id.AttrName:
|
857 | name = self.s[start:end]
|
858 | if name == attr_name:
|
859 | # The value should come next
|
860 | tok_id, start, end = next(events)
|
861 | assert tok_id in (
|
862 | h8_tag_id.QuotedValue, h8_tag_id.UnquotedValue,
|
863 | h8_tag_id.MissingValue), h8_tag_id_str(tok_id)
|
864 | val = start, end
|
865 | break
|
866 |
|
867 | except StopIteration:
|
868 | pass
|
869 | return val
|
870 |
|
871 | def GetAttrRaw(self, attr_name):
|
872 | # type: (str) -> Optional[str]
|
873 | """
|
874 | Return the value, which may be UNESCAPED.
|
875 | """
|
876 | start, end = self.GetSpanForAttrValue(attr_name)
|
877 | if start == -1:
|
878 | return None
|
879 | return self.s[start:end]
|
880 |
|
881 | def AllAttrsRawSlice(self):
|
882 | # type: () -> List[Tuple[str, int, int]]
|
883 | """
|
884 | Get a list of pairs [('class', 3, 5), ('href', 9, 12)]
|
885 | """
|
886 | slices = []
|
887 | events = self.Tokens()
|
888 | try:
|
889 | while True:
|
890 | tok_id, start, end = next(events)
|
891 | if tok_id == h8_tag_id.AttrName:
|
892 | name = self.s[start:end]
|
893 |
|
894 | # The value should come next
|
895 | tok_id, start, end = next(events)
|
896 | assert tok_id in (
|
897 | h8_tag_id.QuotedValue, h8_tag_id.UnquotedValue,
|
898 | h8_tag_id.MissingValue), h8_tag_id_str(tok_id)
|
899 | # Note: quoted values may have &
|
900 | # We would need ANOTHER lexer to unescape them, but we
|
901 | # don't need that for ul-table
|
902 | slices.append((name, start, end))
|
903 | except StopIteration:
|
904 | pass
|
905 | return slices
|
906 |
|
907 | def AllAttrsRaw(self):
|
908 | # type: () -> List[Tuple[str, str]]
|
909 | """
|
910 | Get a list of pairs [('class', 'foo'), ('href', '?foo=1&bar=2')]
|
911 |
|
912 | The quoted values may be escaped. We would need another lexer to
|
913 | unescape them.
|
914 | """
|
915 | slices = self.AllAttrsRawSlice()
|
916 | pairs = []
|
917 | for name, start, end in slices:
|
918 | pairs.append((name, self.s[start:end]))
|
919 | return pairs
|
920 |
|
921 | def Tokens(self):
|
922 | # type: () -> Iterator[Tuple[h8_tag_id_t, int, int]]
|
923 | """
|
924 | Yields a sequence of tokens: Tag (AttrName AttrValue?)*
|
925 |
|
926 | Where each Token is (Type, start_pos, end_pos)
|
927 |
|
928 | Note that start and end are NOT redundant! We skip over some unwanted
|
929 | characters.
|
930 | """
|
931 | m = _TAG_RE.match(self.s, self.start_pos + 1)
|
932 | if not m:
|
933 | raise RuntimeError("Couldn't find HTML tag in %r" %
|
934 | self.WholeTagString())
|
935 | yield h8_tag_id.TagName, m.start(1), m.end(1)
|
936 |
|
937 | pos = m.end(0)
|
938 | #log('POS %d', pos)
|
939 |
|
940 | while True:
|
941 | # don't search past the end
|
942 | m = _ATTR_RE.match(self.s, pos, self.end_pos)
|
943 | if not m:
|
944 | #log('BREAK pos %d', pos)
|
945 | break
|
946 | #log('AttrName %r', m.group(1))
|
947 |
|
948 | yield h8_tag_id.AttrName, m.start(1), m.end(1)
|
949 |
|
950 | #log('m.groups() %r', m.groups())
|
951 | if m.group(2) is not None:
|
952 | # double quoted
|
953 | yield h8_tag_id.QuotedValue, m.start(2), m.end(2)
|
954 | elif m.group(3) is not None:
|
955 | # single quoted - TODO: could have different token types
|
956 | yield h8_tag_id.QuotedValue, m.start(3), m.end(3)
|
957 | elif m.group(4) is not None:
|
958 | yield h8_tag_id.UnquotedValue, m.start(4), m.end(4)
|
959 | else:
|
960 | # <button disabled>
|
961 | end = m.end(0)
|
962 | yield h8_tag_id.MissingValue, end, end
|
963 |
|
964 | # Skip past the "
|
965 | pos = m.end(0)
|
966 |
|
967 | #log('TOK %r', self.s)
|
968 |
|
969 | m = _TAG_LAST_RE.match(self.s, pos)
|
970 | #log('_TAG_LAST_RE match %r', self.s[pos:])
|
971 | if not m:
|
972 | # Extra data at end of tag. TODO: add messages for all these.
|
973 | raise LexError(self.s, pos)
|
974 |
|
975 |
|
976 | # This is similar but not identical to
|
977 | # " ([^>"\x00]*) " # double quoted value
|
978 | # | ' ([^>'\x00]*) ' # single quoted value
|
979 | #
|
980 | # Note: for unquoted values, & isn't allowed, and thus & and c and
|
981 | # ™ are not allowed. We could relax that?
|
982 | ATTR_VALUE_LEX = CHAR_LEX + [
|
983 | (r'[^>&\x00]+', h8_id.RawData),
|
984 | (r'.', h8_id.Invalid),
|
985 | ]
|
986 |
|
987 | ATTR_VALUE_LEX_COMPILED = MakeLexer(ATTR_VALUE_LEX)
|
988 |
|
989 |
|
990 | class AttrValueLexer(object):
|
991 | """
|
992 | <a href="foo=99&bar">
|
993 | <a href='foo=99&bar'>
|
994 | <a href=unquoted>
|
995 | """
|
996 |
|
997 | def __init__(self, s):
|
998 | # type: (str) -> None
|
999 | self.s = s
|
1000 | self.start_pos = -1 # Invalid
|
1001 | self.end_pos = -1
|
1002 |
|
1003 | def Reset(self, start_pos, end_pos):
|
1004 | # type: (int, int) -> None
|
1005 | """Reuse instances of this object."""
|
1006 | assert start_pos >= 0, start_pos
|
1007 | assert end_pos >= 0, end_pos
|
1008 |
|
1009 | self.start_pos = start_pos
|
1010 | self.end_pos = end_pos
|
1011 |
|
1012 | def NumTokens(self):
|
1013 | # type: () -> int
|
1014 | num_tokens = 0
|
1015 | pos = self.start_pos
|
1016 | for tok_id, end_pos in self.Tokens():
|
1017 | if tok_id == h8_id.Invalid:
|
1018 | raise LexError(self.s, pos)
|
1019 | pos = end_pos
|
1020 | #log('pos %d', pos)
|
1021 | num_tokens += 1
|
1022 | return num_tokens
|
1023 |
|
1024 | def Tokens(self):
|
1025 | # type: () -> Iterator[Tuple[h8_id_t, int]]
|
1026 | pos = self.start_pos
|
1027 | while pos < self.end_pos:
|
1028 | # Find the first match, like above.
|
1029 | # Note: frontend/match.py uses _LongestMatch(), which is different!
|
1030 | # TODO: reconcile them. This lexer should be expressible in re2c.
|
1031 | for pat, tok_id in ATTR_VALUE_LEX_COMPILED:
|
1032 | m = pat.match(self.s, pos)
|
1033 | if m:
|
1034 | if 0:
|
1035 | tok_str = m.group(0)
|
1036 | log('token = %r', tok_str)
|
1037 |
|
1038 | end_pos = m.end(0)
|
1039 | yield tok_id, end_pos
|
1040 | pos = end_pos
|
1041 | break
|
1042 | else:
|
1043 | raise AssertionError('h8_id.Invalid rule should have matched')
|