| 1 | #!/usr/bin/env python2
|
| 2 | """
|
| 3 | lazylex/html.py - Low-Level HTML Processing.
|
| 4 |
|
| 5 | See lazylex/README.md for details.
|
| 6 |
|
| 7 | TODO: This should be an Oil library eventually. It's a "lazily-parsed data
|
| 8 | structure" like TSV2.
|
| 9 | """
|
| 10 | from __future__ import print_function
|
| 11 |
|
| 12 | import cStringIO
|
| 13 | import re
|
| 14 | import sys
|
| 15 |
|
| 16 |
|
| 17 | def log(msg, *args):
|
| 18 | msg = msg % args
|
| 19 | print(msg, file=sys.stderr)
|
| 20 |
|
| 21 |
|
| 22 | class LexError(Exception):
|
| 23 | """For bad lexical elements like <> or && """
|
| 24 |
|
| 25 | def __init__(self, s, pos):
|
| 26 | self.s = s
|
| 27 | self.pos = pos
|
| 28 |
|
| 29 | def __str__(self):
|
| 30 | return '(LexError %r)' % (self.s[self.pos : self.pos + 20])
|
| 31 |
|
| 32 |
|
| 33 | class ParseError(Exception):
|
| 34 | """For errors in the tag structure."""
|
| 35 |
|
| 36 | def __init__(self, msg, *args):
|
| 37 | self.msg = msg
|
| 38 | self.args = args
|
| 39 |
|
| 40 | def __str__(self):
|
| 41 | return '(ParseError %s)' % (self.msg % self.args)
|
| 42 |
|
| 43 |
|
| 44 | class Output(object):
|
| 45 | """
|
| 46 | Takes an underlying input buffer and an output file. Maintains a position in
|
| 47 | the input buffer.
|
| 48 |
|
| 49 | Print FROM the input or print new text to the output.
|
| 50 | """
|
| 51 |
|
| 52 | def __init__(self, s, f, left_pos=0, right_pos=0):
|
| 53 | self.s = s
|
| 54 | self.f = f
|
| 55 | self.pos = left_pos
|
| 56 | if right_pos == 0:
|
| 57 | self.right_pos = len(s)
|
| 58 | else:
|
| 59 | self.right_pos = right_pos
|
| 60 |
|
| 61 | def SkipTo(self, pos):
|
| 62 | """Skip to a position."""
|
| 63 | self.pos = pos
|
| 64 |
|
| 65 | def PrintUntil(self, pos):
|
| 66 | """Print until a position."""
|
| 67 | piece = self.s[self.pos : pos]
|
| 68 | self.f.write(piece)
|
| 69 | self.pos = pos
|
| 70 |
|
| 71 | def PrintTheRest(self):
|
| 72 | """Print until the end of the string."""
|
| 73 | self.PrintUntil(self.right_pos)
|
| 74 |
|
| 75 | def Print(self, s):
|
| 76 | """Print text to the underlying buffer."""
|
| 77 | self.f.write(s)
|
| 78 |
|
| 79 |
|
| 80 | # HTML Tokens
|
| 81 | ( Decl, Comment, Processing,
|
| 82 | StartTag, StartEndTag, EndTag,
|
| 83 | DecChar, HexChar, CharEntity,
|
| 84 | RawData,
|
| 85 | Invalid, EndOfStream ) = range(12)
|
| 86 |
|
| 87 |
|
| 88 | def MakeLexer(rules):
|
| 89 | return [
|
| 90 | # DOTALL is for the comment
|
| 91 | (re.compile(pat, re.VERBOSE | re.DOTALL), i) for
|
| 92 | (pat, i) in rules
|
| 93 | ]
|
| 94 |
|
| 95 | #
|
| 96 | # Eggex
|
| 97 | #
|
| 98 | # Tag = / ~['>']+ /
|
| 99 |
|
| 100 | # Is this valid? A single character?
|
| 101 | # Tag = / ~'>'* /
|
| 102 |
|
| 103 | # Maybe better: / [NOT '>']+/
|
| 104 | # capital letters not allowed there?
|
| 105 | #
|
| 106 | # But then this is confusing:
|
| 107 | # / [NOT ~digit]+/
|
| 108 | #
|
| 109 | # / [NOT digit] / is [^\d]
|
| 110 | # / ~digit / is \D
|
| 111 | #
|
| 112 | # Or maybe:
|
| 113 | #
|
| 114 | # / [~ digit]+ /
|
| 115 | # / [~ '>']+ /
|
| 116 | # / [NOT '>']+ /
|
| 117 |
|
| 118 | # End = / '</' Tag '>' /
|
| 119 | # StartEnd = / '<' Tag '/>' /
|
| 120 | # Start = / '<' Tag '>' /
|
| 121 | #
|
| 122 | # EntityRef = / '&' dot{* N} ';' /
|
| 123 |
|
| 124 |
|
| 125 | LEXER = [
|
| 126 | # TODO: instead of nongreedy matches, the loop can just do .find('-->') and
|
| 127 | # .find('?>')
|
| 128 |
|
| 129 | # Actually non-greedy matches are regular and can be matched in linear time
|
| 130 | # with RE2.
|
| 131 | #
|
| 132 | # https://news.ycombinator.com/item?id=27099798
|
| 133 | #
|
| 134 | # Maybe try combining all of these for speed.
|
| 135 |
|
| 136 | (r'<!-- .*? -->', Comment),
|
| 137 | (r'<\? .*? \?>', Processing),
|
| 138 |
|
| 139 | # NOTE: < is allowed in these.
|
| 140 | (r'<! [^>]+ >', Decl), # <!DOCTYPE html>
|
| 141 |
|
| 142 | (r'</ [^>]+ >', EndTag), # self-closing <br/> comes FIRST
|
| 143 | (r'< [^>]+ />', StartEndTag), # end </a>
|
| 144 | (r'< [^>]+ >', StartTag), # start <a>
|
| 145 |
|
| 146 | (r'&\# [0-9]+ ;', DecChar),
|
| 147 | (r'&\# x[0-9a-fA-F]+ ;', HexChar),
|
| 148 | (r'& [a-zA-Z]+ ;', CharEntity),
|
| 149 |
|
| 150 | # Note: > is allowed in raw data.
|
| 151 | # https://stackoverflow.com/questions/10462348/right-angle-bracket-in-html
|
| 152 | (r'[^&<]+', RawData),
|
| 153 |
|
| 154 | (r'.', Invalid), # error!
|
| 155 | ]
|
| 156 |
|
| 157 | LEXER = MakeLexer(LEXER)
|
| 158 |
|
| 159 |
|
| 160 | def _Tokens(s, left_pos, right_pos):
|
| 161 | """
|
| 162 | Args:
|
| 163 | s: string to parse
|
| 164 | left_pos, right_pos: Optional span boundaries.
|
| 165 | """
|
| 166 | pos = left_pos
|
| 167 | if right_pos == 0:
|
| 168 | n = len(s)
|
| 169 | else:
|
| 170 | n = right_pos
|
| 171 |
|
| 172 | while pos < n:
|
| 173 | # Find the FIRST pattern that matches.
|
| 174 | for pat, tok_id in LEXER:
|
| 175 | m = pat.match(s, pos)
|
| 176 | if m:
|
| 177 | end_pos = m.end()
|
| 178 | yield tok_id, end_pos
|
| 179 | pos = end_pos
|
| 180 | break
|
| 181 |
|
| 182 | # Zero length sentinel
|
| 183 | yield EndOfStream, pos
|
| 184 |
|
| 185 |
|
| 186 | def ValidTokens(s, left_pos=0, right_pos=0):
|
| 187 | """
|
| 188 | Wrapper around _Tokens to prevent callers from having to handle Invalid.
|
| 189 |
|
| 190 | I'm not combining the two functions because I might want to do a 'yield'
|
| 191 | transformation on Tokens()? Exceptions might complicate the issue?
|
| 192 | """
|
| 193 | pos = left_pos
|
| 194 | for tok_id, end_pos in _Tokens(s, left_pos, right_pos):
|
| 195 | if tok_id == Invalid:
|
| 196 | raise LexError(s, pos)
|
| 197 | yield tok_id, end_pos
|
| 198 | pos = end_pos
|
| 199 |
|
| 200 |
|
| 201 | # To match <a or </a
|
| 202 | # <h2 but not <2h ?
|
| 203 | _TAG_RE = re.compile(r'/? \s* ([a-zA-Z][a-zA-Z0-9]*)', re.VERBOSE)
|
| 204 |
|
| 205 | # To match href="foo"
|
| 206 |
|
| 207 | _ATTR_RE = re.compile(r'''
|
| 208 | \s+ # Leading whitespace is required
|
| 209 | ([a-z]+) # Attribute name
|
| 210 | (?: # Optional attribute value
|
| 211 | \s* = \s*
|
| 212 | (?:
|
| 213 | " ([^>"]*) " # double quoted value
|
| 214 | | ([a-zA-Z0-9_\-]+) # Just allow unquoted "identifiers"
|
| 215 | # TODO: relax this? for href=$foo
|
| 216 | )
|
| 217 | )?
|
| 218 | ''', re.VERBOSE)
|
| 219 |
|
| 220 |
|
| 221 | TagName, AttrName, UnquotedValue, QuotedValue = range(4)
|
| 222 |
|
| 223 | class TagLexer(object):
|
| 224 | """
|
| 225 | Given a tag like <a href="..."> or <link type="..." />, the TagLexer
|
| 226 | provides a few operations:
|
| 227 |
|
| 228 | - What is the tag?
|
| 229 | - Iterate through the attributes, giving (name, value_start_pos, value_end_pos)
|
| 230 | """
|
| 231 | def __init__(self, s):
|
| 232 | self.s = s
|
| 233 | self.start_pos = -1 # Invalid
|
| 234 | self.end_pos = -1
|
| 235 |
|
| 236 | def Reset(self, start_pos, end_pos):
|
| 237 | self.start_pos = start_pos
|
| 238 | self.end_pos = end_pos
|
| 239 |
|
| 240 | def TagString(self):
|
| 241 | return self.s[self.start_pos : self.end_pos]
|
| 242 |
|
| 243 | def TagName(self):
|
| 244 | # First event
|
| 245 | tok_id, start, end = next(self.Tokens())
|
| 246 | return self.s[start : end]
|
| 247 |
|
| 248 | def GetSpanForAttrValue(self, attr_name):
|
| 249 | # Algorithm: search for QuotedValue or UnquotedValue after AttrName
|
| 250 | # TODO: Could also cache these
|
| 251 |
|
| 252 | events = self.Tokens()
|
| 253 | val = (-1, -1)
|
| 254 | try:
|
| 255 | while True:
|
| 256 | tok_id, start, end = next(events)
|
| 257 | if tok_id == AttrName:
|
| 258 | name = self.s[start:end]
|
| 259 | if name == attr_name:
|
| 260 | # For HasAttr()
|
| 261 | #val = True
|
| 262 |
|
| 263 | # Now try to get a real value
|
| 264 | tok_id, start, end = next(events)
|
| 265 | if tok_id in (QuotedValue, UnquotedValue):
|
| 266 |
|
| 267 | # TODO: Unescape this with htmlentitydefs
|
| 268 | # I think we need another lexer!
|
| 269 | #
|
| 270 | # We could make a single pass?
|
| 271 | # Shortcut: 'if '&' in substring'
|
| 272 | # Then we need to unescape it
|
| 273 |
|
| 274 | val = start, end
|
| 275 | break
|
| 276 |
|
| 277 | except StopIteration:
|
| 278 | pass
|
| 279 | return val
|
| 280 |
|
| 281 | def GetAttr(self, attr_name):
|
| 282 | # Algorithm: search for QuotedValue or UnquotedValue after AttrName
|
| 283 | # TODO: Could also cache these
|
| 284 | start, end = self.GetSpanForAttrValue(attr_name)
|
| 285 | if start == -1:
|
| 286 | return None
|
| 287 | return self.s[start : end]
|
| 288 |
|
| 289 | def Tokens(self):
|
| 290 | """
|
| 291 | Yields a sequence of tokens: Tag (AttrName AttrValue?)*
|
| 292 |
|
| 293 | Where each Token is (Type, start_pos, end_pos)
|
| 294 |
|
| 295 | Note that start and end are NOT redundant! We skip over some unwanted
|
| 296 | characters.
|
| 297 | """
|
| 298 | m = _TAG_RE.match(self.s, self.start_pos+1)
|
| 299 | if not m:
|
| 300 | raise RuntimeError('Invalid HTML tag: %r' % self.TagString())
|
| 301 | yield TagName, m.start(1), m.end(1)
|
| 302 |
|
| 303 | pos = m.end(0)
|
| 304 |
|
| 305 | while True:
|
| 306 | # don't search past the end
|
| 307 | m = _ATTR_RE.match(self.s, pos, self.end_pos)
|
| 308 | if not m:
|
| 309 | # A validating parser would check that > or /> is next -- there's no junk
|
| 310 | break
|
| 311 |
|
| 312 | yield AttrName, m.start(1), m.end(1)
|
| 313 |
|
| 314 | # Quoted is group 2, unquoted is group 3.
|
| 315 | if m.group(2) is not None:
|
| 316 | yield QuotedValue, m.start(2), m.end(2)
|
| 317 | elif m.group(3) is not None:
|
| 318 | yield UnquotedValue, m.start(3), m.end(3)
|
| 319 |
|
| 320 | # Skip past the "
|
| 321 | pos = m.end(0)
|
| 322 |
|
| 323 |
|
| 324 | def ReadUntilStartTag(it, tag_lexer, tag_name):
|
| 325 | """Find the next <foo>.
|
| 326 |
|
| 327 | tag_lexer is RESET.
|
| 328 | """
|
| 329 | pos = 0
|
| 330 | while True:
|
| 331 | try:
|
| 332 | tok_id, end_pos = next(it)
|
| 333 | except StopIteration:
|
| 334 | break
|
| 335 | tag_lexer.Reset(pos, end_pos)
|
| 336 | if tok_id == StartTag and tag_lexer.TagName() == tag_name:
|
| 337 | return pos, end_pos
|
| 338 |
|
| 339 | pos = end_pos
|
| 340 |
|
| 341 | raise ParseError('No start tag %r', tag_name)
|
| 342 |
|
| 343 |
|
| 344 | def ReadUntilEndTag(it, tag_lexer, tag_name):
|
| 345 | """Find the next </foo>.
|
| 346 |
|
| 347 | tag_lexer is RESET.
|
| 348 | """
|
| 349 | pos = 0
|
| 350 | while True:
|
| 351 | try:
|
| 352 | tok_id, end_pos = next(it)
|
| 353 | except StopIteration:
|
| 354 | break
|
| 355 | tag_lexer.Reset(pos, end_pos)
|
| 356 | if tok_id == EndTag and tag_lexer.TagName() == tag_name:
|
| 357 | return pos, end_pos
|
| 358 |
|
| 359 | pos = end_pos
|
| 360 |
|
| 361 | raise ParseError('No end tag %r', tag_name)
|
| 362 |
|
| 363 |
|
| 364 | CHAR_ENTITY = {
|
| 365 | 'amp': '&',
|
| 366 | 'lt': '<',
|
| 367 | 'gt': '>',
|
| 368 | 'quot': '"',
|
| 369 | }
|
| 370 |
|
| 371 |
|
| 372 | def ToText(s, left_pos=0, right_pos=0):
|
| 373 | """
|
| 374 | Given HTML, return text by unquoting > and < etc.
|
| 375 |
|
| 376 | Used by:
|
| 377 | doctools/oils_doc.py: PygmentsPlugin
|
| 378 | doctool/make_help.py: HelpIndexCards
|
| 379 |
|
| 380 | In the latter case, we cold process some tags, like:
|
| 381 |
|
| 382 | - Blue Link (not clickable, but still useful)
|
| 383 | - Red X
|
| 384 |
|
| 385 | That should be html.ToAnsi.
|
| 386 | """
|
| 387 | f = cStringIO.StringIO()
|
| 388 | out = Output(s, f, left_pos, right_pos)
|
| 389 |
|
| 390 | pos = left_pos
|
| 391 | for tok_id, end_pos in ValidTokens(s, left_pos, right_pos):
|
| 392 | if tok_id == RawData:
|
| 393 | out.SkipTo(pos)
|
| 394 | out.PrintUntil(end_pos)
|
| 395 |
|
| 396 | elif tok_id == CharEntity: # &
|
| 397 |
|
| 398 | entity = s[pos+1 : end_pos-1]
|
| 399 |
|
| 400 | out.SkipTo(pos)
|
| 401 | out.Print(CHAR_ENTITY[entity])
|
| 402 | out.SkipTo(end_pos)
|
| 403 |
|
| 404 | # Not handling these yet
|
| 405 | elif tok_id == HexChar:
|
| 406 | raise AssertionError('Hex Char %r' % s[pos : pos + 20])
|
| 407 |
|
| 408 | elif tok_id == DecChar:
|
| 409 | raise AssertionError('Dec Char %r' % s[pos : pos + 20])
|
| 410 |
|
| 411 | pos = end_pos
|
| 412 |
|
| 413 | out.PrintTheRest()
|
| 414 | return f.getvalue()
|
| 415 |
|
| 416 |
|