| 1 | #!/usr/bin/env python2
|
| 2 | """
|
| 3 | parse_cpython.py
|
| 4 | """
|
| 5 | from __future__ import print_function
|
| 6 |
|
| 7 | import errno
|
| 8 | import os
|
| 9 | import re
|
| 10 | import sys
|
| 11 |
|
| 12 | from mycpp.mylib import log
|
| 13 | # TODO: Could move these to a place where they don't depend on Oil
|
| 14 | from frontend.lexer_def import C, R
|
| 15 |
|
| 16 |
|
| 17 | C_DEF = [
|
| 18 | R(r'#.*', 'Comment'),
|
| 19 | R(r'[ \t\n]+', 'Whitespace'),
|
| 20 |
|
| 21 | # This could be more space-insensitive.
|
| 22 | R(r'static.*PyMethodDef (.*)\[\]\s*=\s*', 'BeginDef'),
|
| 23 | C(r'{', 'LBrace'),
|
| 24 | C(r'}', 'RBrace'),
|
| 25 | C(r',', 'Comma'),
|
| 26 | C(r';', 'Semi'),
|
| 27 | R(r'"([^"]*)"', 'Str'),
|
| 28 | C(r'FILE', 'FILE'),
|
| 29 | C(r'PyDoc_STR(', 'LDocStr'),
|
| 30 | C(r')', 'RDocStr'),
|
| 31 | R(r'[^,}\n]+', 'Opaque'),
|
| 32 | ]
|
| 33 |
|
| 34 |
|
| 35 | # NOTE: This is copied from osh/match.py because we don't have 're' there.
|
| 36 | def _CompileAll(pat_list):
|
| 37 | result = []
|
| 38 | for is_regex, pat, token_id in pat_list:
|
| 39 | if not is_regex:
|
| 40 | pat = re.escape(pat) # turn $ into \$
|
| 41 | result.append((re.compile(pat), token_id))
|
| 42 | return result
|
| 43 |
|
| 44 |
|
| 45 | class Lexer(object):
|
| 46 | def __init__(self, pat_list):
|
| 47 | self.pat_list = _CompileAll(pat_list)
|
| 48 |
|
| 49 | def Tokens(self, s):
|
| 50 | pos = 0
|
| 51 | n = len(s)
|
| 52 | while pos < n:
|
| 53 | for pat, id_ in self.pat_list:
|
| 54 | # FIRST MATCH
|
| 55 | m = pat.match(s, pos)
|
| 56 | if m:
|
| 57 | if m.groups():
|
| 58 | start, end = m.start(1), m.end(1)
|
| 59 | else:
|
| 60 | start, end = m.start(0), m.end(0)
|
| 61 | pos = m.end()
|
| 62 | break # found a match
|
| 63 | else:
|
| 64 | raise AssertionError(
|
| 65 | 'no token matched at position %r: %r' % ( pos, s[pos]))
|
| 66 |
|
| 67 | if id_ != 'Whitespace':
|
| 68 | yield id_, s[start:end], pos
|
| 69 | yield 'EOF', '', -1
|
| 70 |
|
| 71 |
|
| 72 | class Parser(object):
|
| 73 | """Parser for C PyMethodDef initializer lists."""
|
| 74 |
|
| 75 | def __init__(self, tokens):
|
| 76 | self.tokens = tokens
|
| 77 | self.Next() # initialize
|
| 78 |
|
| 79 | def Next(self):
|
| 80 | while True:
|
| 81 | self.tok_id, self.tok_val, self.pos = self.tokens.next()
|
| 82 | if self.tok_id not in ('Comment', 'Whitespace'):
|
| 83 | break
|
| 84 | if 0:
|
| 85 | log('%s %r', self.tok_id, self.tok_val)
|
| 86 |
|
| 87 | def Eat(self, tok_id):
|
| 88 | if self.tok_id != tok_id:
|
| 89 | raise RuntimeError(
|
| 90 | 'Expected %r, got %r %r (byte offset %d)' %
|
| 91 | (tok_id, self.tok_id, self.tok_val, self.pos))
|
| 92 |
|
| 93 | self.Next()
|
| 94 |
|
| 95 | def ParseName(self):
|
| 96 | """
|
| 97 | Name = Str | Opaque('NULL') | Opaque('0')
|
| 98 | """
|
| 99 | if self.tok_id == 'Str':
|
| 100 | name = self.tok_val
|
| 101 | elif self.tok_id == 'Opaque':
|
| 102 | assert self.tok_val in ('NULL', '0')
|
| 103 | name = None
|
| 104 | else:
|
| 105 | raise RuntimeError('Unexpected token %r' % self.tok_id)
|
| 106 | self.Next()
|
| 107 | return name
|
| 108 |
|
| 109 | def ParseVal(self):
|
| 110 | """
|
| 111 | Val = Str
|
| 112 | | Opaque
|
| 113 | | LDocStr Str+ RDocStr # string concatenation happens
|
| 114 | """
|
| 115 | if self.tok_id == 'LDocStr':
|
| 116 | self.Next()
|
| 117 |
|
| 118 | val = self.tok_val
|
| 119 | self.Eat('Str')
|
| 120 | while self.tok_id == 'Str':
|
| 121 | val += self.tok_val
|
| 122 | self.Next()
|
| 123 |
|
| 124 | self.Eat('RDocStr')
|
| 125 |
|
| 126 | elif self.tok_id in ('Opaque', 'Str'):
|
| 127 | val = self.tok_val
|
| 128 | self.Next()
|
| 129 |
|
| 130 | else:
|
| 131 | raise RuntimeError('Unexpected token %r' % self.tok_id)
|
| 132 |
|
| 133 | return val
|
| 134 |
|
| 135 | def ParseItem(self):
|
| 136 | """
|
| 137 | Item = '{' Name (',' Val)+ '}' ','?
|
| 138 | """
|
| 139 | self.Eat('LBrace')
|
| 140 | name = self.ParseName()
|
| 141 |
|
| 142 | vals = []
|
| 143 | while self.tok_id == 'Comma':
|
| 144 | self.Next()
|
| 145 | vals.append(self.ParseVal())
|
| 146 |
|
| 147 | self.Eat('RBrace')
|
| 148 |
|
| 149 | if self.tok_id == 'Comma': # Optional
|
| 150 | self.Next()
|
| 151 |
|
| 152 | return name, vals
|
| 153 |
|
| 154 | def ParseDef(self):
|
| 155 | """
|
| 156 | Def = BeginDef '{' Item+ '}' ';'
|
| 157 | """
|
| 158 | def_name = self.tok_val
|
| 159 | self.Eat('BeginDef')
|
| 160 | self.Eat('LBrace')
|
| 161 |
|
| 162 | items = []
|
| 163 | while self.tok_id != 'RBrace':
|
| 164 | items.append(self.ParseItem())
|
| 165 |
|
| 166 | self.Next()
|
| 167 | self.Eat('Semi')
|
| 168 |
|
| 169 | return (def_name, items)
|
| 170 |
|
| 171 | def ParseHeader(self):
|
| 172 | self.Eat('FILE')
|
| 173 | path = self.tok_val
|
| 174 | self.Eat('Opaque')
|
| 175 | return path
|
| 176 |
|
| 177 | def ParseFile(self):
|
| 178 | """
|
| 179 | File = Header Def*
|
| 180 | """
|
| 181 | path = self.ParseHeader()
|
| 182 | defs = []
|
| 183 | while self.tok_id not in ('FILE', 'EOF'):
|
| 184 | defs.append(self.ParseDef())
|
| 185 |
|
| 186 | return path, defs
|
| 187 |
|
| 188 | def ParseStream(self):
|
| 189 | """
|
| 190 | Stream = File*
|
| 191 | """
|
| 192 | files = []
|
| 193 | while self.tok_id != 'EOF':
|
| 194 | files.append(self.ParseFile())
|
| 195 |
|
| 196 | return files
|
| 197 |
|
| 198 |
|
| 199 | def PrettyPrint(rel_path, def_name, entries, predicate, f, stats):
|
| 200 | def out(msg, *args):
|
| 201 | if args:
|
| 202 | msg = msg % args
|
| 203 | print(msg, file=f, end='')
|
| 204 |
|
| 205 | out('static PyMethodDef %s[] = {\n', def_name)
|
| 206 | for entry_name, vals in entries:
|
| 207 | if entry_name is None:
|
| 208 | out(' {0},\n') # null initializer
|
| 209 | continue
|
| 210 | stats['num_methods'] += 1
|
| 211 |
|
| 212 | if not predicate(rel_path, def_name, entry_name):
|
| 213 | stats['num_filtered'] += 1
|
| 214 | continue
|
| 215 |
|
| 216 | # Reprint the definition, but omit the docstring.
|
| 217 | out(' {"%s", ', entry_name)
|
| 218 | out(vals[0]) # The C function
|
| 219 | out(', ')
|
| 220 | out(vals[1]) # The flags
|
| 221 | out('},\n')
|
| 222 | out('};\n')
|
| 223 |
|
| 224 |
|
| 225 | MODULES_TO_FILTER = [
|
| 226 | # My Own
|
| 227 | 'libc.c',
|
| 228 | 'fastlex.c',
|
| 229 | 'line_input.c',
|
| 230 |
|
| 231 | 'import.c',
|
| 232 | 'marshal.c', # additional filters below
|
| 233 | #'zipimport.c', # Cannot filter this because find_module called from C!
|
| 234 |
|
| 235 | # Types for Builtins
|
| 236 | 'enumobject.c',
|
| 237 | 'rangeobject.c',
|
| 238 |
|
| 239 | # Interpreter types
|
| 240 | 'descrobject.c',
|
| 241 | 'exceptions.c',
|
| 242 | 'structseq.c',
|
| 243 | '_warnings.c',
|
| 244 |
|
| 245 | # Control flow
|
| 246 | 'frameobject.c',
|
| 247 | 'genobject.c',
|
| 248 | 'iterobject.c',
|
| 249 |
|
| 250 | # GC
|
| 251 | '_weakref.c',
|
| 252 | 'weakrefobject.c',
|
| 253 | 'gcmodule.c',
|
| 254 |
|
| 255 | # "Data types"
|
| 256 | #'boolobject.c', # No defs
|
| 257 | 'cStringIO.c',
|
| 258 | 'dictobject.c',
|
| 259 | 'fileobject.c',
|
| 260 | 'floatobject.c',
|
| 261 | 'intobject.c',
|
| 262 | 'listobject.c',
|
| 263 | 'longobject.c',
|
| 264 | #'moduleobject.c', # No defs
|
| 265 | 'setobject.c',
|
| 266 | 'stringobject.c',
|
| 267 | 'tupleobject.c',
|
| 268 | 'sliceobject.c',
|
| 269 | 'typeobject.c',
|
| 270 |
|
| 271 | # Builtins
|
| 272 | 'bltinmodule.c', # additional filters below
|
| 273 | #'sysmodule.c', # Filtered below
|
| 274 |
|
| 275 | # Libraries
|
| 276 | 'errnomodule.c', # has no methods, but include it for completeness
|
| 277 | 'fcntlmodule.c',
|
| 278 | 'posixmodule.c',
|
| 279 | 'pwdmodule.c',
|
| 280 | 'readline.c',
|
| 281 | 'resource.c',
|
| 282 | 'signalmodule.c',
|
| 283 | 'timemodule.c',
|
| 284 | 'termios.c',
|
| 285 | 'mathmodule.c',
|
| 286 | ]
|
| 287 |
|
| 288 |
|
| 289 | class OilMethodFilter(object):
|
| 290 |
|
| 291 | def __init__(self, py_names):
|
| 292 | self.py_names = py_names
|
| 293 |
|
| 294 | def __call__(self, rel_path, def_name, method_name):
|
| 295 | basename = os.path.basename(rel_path)
|
| 296 |
|
| 297 | if method_name == 'count': # False positive for {str,list,tuple}.count()
|
| 298 | return False
|
| 299 |
|
| 300 | if method_name == 'collect': # False positive: pyannotate and gcmodule.c
|
| 301 | return False
|
| 302 |
|
| 303 | # enter/exit needed for 'with open'. __length_hint__ is an optimization.
|
| 304 | if method_name in ('__enter__', '__exit__', '__length_hint__'):
|
| 305 | return True
|
| 306 | # Notes:
|
| 307 | # - __reduce__ and __setstate__ are for pickle. And I think
|
| 308 | # __getnewargs__.
|
| 309 | # - Do we need __sizeof__? Is that for sys.getsizeof()?
|
| 310 |
|
| 311 | # 5/2022: avoid regression? Not sure why this was getting deleted
|
| 312 | if method_name == '__getitem__':
|
| 313 | return True
|
| 314 |
|
| 315 | # NOTE: LoadYshGrammar needs marshal.loads().
|
| 316 | # False positive for yajl.dumps() and load()
|
| 317 | if basename == 'marshal.c' and method_name in ('dump', 'dumps', 'load'):
|
| 318 | return False
|
| 319 |
|
| 320 | # Auto-filtering gave false-positives here.
|
| 321 | # We don't need top-level next(). The method should be good enough.
|
| 322 | # iter is a field name
|
| 323 | if (basename == 'bltinmodule.c' and method_name in
|
| 324 | ('compile', 'format', 'next', 'vars', 'iter', 'eval', 'bin')):
|
| 325 | return False
|
| 326 | if basename == 'bltinmodule.c':
|
| 327 | # Get "bootstrapping error" without this.
|
| 328 | if method_name == '__import__':
|
| 329 | return True
|
| 330 |
|
| 331 | if basename == '_warnings.c' and method_name == 'warn':
|
| 332 | return False
|
| 333 |
|
| 334 | if basename == 'dictobject.c' and method_name in (
|
| 335 | 'iterkeys', 'itervalues', 'copy', 'fromkeys', 'popitem', 'setdefault'):
|
| 336 | return False
|
| 337 |
|
| 338 | if basename == 'tupleobject.c' and method_name == 'index':
|
| 339 | return False
|
| 340 |
|
| 341 | if basename == 'stringobject.c' and method_name == 'translate':
|
| 342 | # false positive from arg.translate
|
| 343 | return False
|
| 344 |
|
| 345 | if basename == 'setobject.c' and method_name in ('pop', 'copy'):
|
| 346 | return False
|
| 347 |
|
| 348 | if basename == 'frozensetobject.c' and method_name == 'copy':
|
| 349 | return False
|
| 350 |
|
| 351 | if basename == 'sliceobject.c' and method_name == 'indices':
|
| 352 | return False
|
| 353 |
|
| 354 | # Shadowed by fanos.send(), posix.close(), etc.
|
| 355 | if basename == 'genobject.c' and method_name in ('send', 'close'):
|
| 356 | return False
|
| 357 |
|
| 358 | # We're using list.remove()
|
| 359 | if basename == 'posixmodule.c' and method_name == 'remove': # Shadowed
|
| 360 | return False
|
| 361 |
|
| 362 | # We're using dict.clear() and list.remove()
|
| 363 | if basename == 'setobject.c' and method_name in ('clear', 'remove'):
|
| 364 | return False
|
| 365 |
|
| 366 | # Do custom filtering here.
|
| 367 | if (basename == 'sysmodule.c' and method_name not in self.py_names):
|
| 368 | # These can't be removed or they cause assertions!
|
| 369 | if method_name not in ('displayhook', 'excepthook'):
|
| 370 | return False
|
| 371 |
|
| 372 | # This one is called from C.
|
| 373 | if basename == 'signalmodule.c' and method_name == 'default_int_handler':
|
| 374 | return True
|
| 375 |
|
| 376 | # Name collisions
|
| 377 | if basename == 'mathmodule.c' and method_name in ('exp', 'log'):
|
| 378 | return False
|
| 379 |
|
| 380 | # segfault without this
|
| 381 | if basename == 'typeobject.c' and method_name == '__new__':
|
| 382 | return True
|
| 383 |
|
| 384 | if basename == 'descrobject.c':
|
| 385 | # Apparently used for dir() on class namespace, as in dir(Id).
|
| 386 | if method_name == 'keys':
|
| 387 | return True
|
| 388 | return False
|
| 389 |
|
| 390 | # Try just filtering {time,pwd,posix}module.c, etc.
|
| 391 | if basename in MODULES_TO_FILTER and method_name not in self.py_names:
|
| 392 | return False
|
| 393 |
|
| 394 | #log('= %s %s', def_name, method_name)
|
| 395 |
|
| 396 | # If it doesn't appear in the .py source, it can't be used. (Exception: it
|
| 397 | # could be used in C source with dynamic lookup? But I don't think CPython
|
| 398 | # does that.)
|
| 399 | #if method_name not in self.py_names:
|
| 400 | if 0:
|
| 401 | log('Omitting %r', method_name)
|
| 402 | return False
|
| 403 |
|
| 404 | return True
|
| 405 |
|
| 406 |
|
| 407 | def main(argv):
|
| 408 | action = argv[1]
|
| 409 |
|
| 410 | try:
|
| 411 | py_names_path = argv[2]
|
| 412 | except IndexError:
|
| 413 | method_filter = None
|
| 414 | else:
|
| 415 | py_names = set()
|
| 416 | with open(py_names_path) as f:
|
| 417 | for line in f:
|
| 418 | py_names.add(line.strip())
|
| 419 | method_filter = OilMethodFilter(py_names)
|
| 420 |
|
| 421 | if action == 'filtered':
|
| 422 | tokens = None
|
| 423 | else:
|
| 424 | tokens = Lexer(C_DEF).Tokens(sys.stdin.read())
|
| 425 |
|
| 426 | if action == 'lex': # for debugging
|
| 427 | while True:
|
| 428 | id_, value, pos = tokens.next()
|
| 429 | print('%s\t%r' % (id_, value))
|
| 430 | if id_ == 'EOF':
|
| 431 | break
|
| 432 |
|
| 433 | elif action == 'audit': # show after filtering, for debugging
|
| 434 | p = Parser(tokens)
|
| 435 | files = p.ParseStream()
|
| 436 | for rel_path, defs in files:
|
| 437 | basename = os.path.basename(rel_path)
|
| 438 |
|
| 439 | print(rel_path)
|
| 440 | for def_name, entries in defs:
|
| 441 | print('\t' + def_name)
|
| 442 | for method_name, vals in entries:
|
| 443 | if method_name is None:
|
| 444 | continue
|
| 445 | if not method_filter(rel_path, def_name, method_name):
|
| 446 | continue
|
| 447 | print('\t\t%s %s' % (method_name, vals))
|
| 448 |
|
| 449 | elif action == 'filter': # for slimming the build down
|
| 450 | out_dir = argv[3]
|
| 451 |
|
| 452 | p = Parser(tokens)
|
| 453 | files = p.ParseStream()
|
| 454 |
|
| 455 | # Print to files.
|
| 456 |
|
| 457 | stats = {'num_methods': 0, 'num_defs': 0, 'num_filtered': 0}
|
| 458 | for rel_path, defs in files:
|
| 459 | # Make a directory for each .c file! Each file is a def.
|
| 460 | c_dir = os.path.join(out_dir, rel_path)
|
| 461 | try:
|
| 462 | os.makedirs(c_dir)
|
| 463 | except OSError as e:
|
| 464 | if e.errno != errno.EEXIST:
|
| 465 | raise
|
| 466 |
|
| 467 | for def_name, entries in defs:
|
| 468 | out_path = os.path.join(c_dir, '%s.def' % def_name)
|
| 469 |
|
| 470 | # TODO: Write a separate file here for each one. We have to include a
|
| 471 | # different file at each definition.
|
| 472 |
|
| 473 | with open(out_path, 'w') as f:
|
| 474 | print('// %s' % rel_path, file=f)
|
| 475 | print('', file=f)
|
| 476 | PrettyPrint(rel_path, def_name, entries, method_filter, f, stats)
|
| 477 |
|
| 478 | stats['num_defs'] += 1
|
| 479 | log('Wrote %s', out_path)
|
| 480 |
|
| 481 | stats['num_left'] = stats['num_methods'] - stats['num_filtered']
|
| 482 | log('cpython_defs.py: Filtered %(num_filtered)d of %(num_methods)d methods, '
|
| 483 | 'leaving %(num_left)d (from %(num_defs)d definitions)' % stats)
|
| 484 |
|
| 485 | elif action == 'tsv':
|
| 486 | p = Parser(tokens)
|
| 487 | files = p.ParseStream()
|
| 488 | header = [
|
| 489 | 'file', 'def_name', 'py_method_name', 'c_symbol_name', 'flags',
|
| 490 | 'used'
|
| 491 | ]
|
| 492 | print('\t'.join(header))
|
| 493 | for rel_path, defs in files:
|
| 494 | for def_name, entries in defs:
|
| 495 | for method_name, vals in entries:
|
| 496 | if method_name is None:
|
| 497 | continue
|
| 498 | b = method_filter(rel_path, def_name, method_name)
|
| 499 | used = 'T' if b else 'F'
|
| 500 |
|
| 501 | # TODO: The c_symbol_name could be parsed better. It sometimes has
|
| 502 | # "(PyCFunction)" on the front of it.
|
| 503 |
|
| 504 | row = [rel_path, def_name, method_name, vals[0], vals[1], used]
|
| 505 | print('\t'.join(row))
|
| 506 |
|
| 507 | elif action == 'filtered':
|
| 508 | for name in MODULES_TO_FILTER:
|
| 509 | print(name)
|
| 510 |
|
| 511 | else:
|
| 512 | raise RuntimeError('Invalid action %r' % action)
|
| 513 |
|
| 514 |
|
| 515 | if __name__ == '__main__':
|
| 516 | try:
|
| 517 | main(sys.argv)
|
| 518 | except RuntimeError as e:
|
| 519 | print('FATAL: %s' % e, file=sys.stderr)
|
| 520 | sys.exit(1)
|