1 #!/usr/bin/env python 2 # A tool to parse ASTMatchers.h and update the documentation in 3 # ../LibASTMatchersReference.html automatically. Run from the 4 # directory in which this file is located to update the docs. 5 6 import collections 7 import re 8 import urllib2 9 10 MATCHERS_FILE = '../../include/clang/ASTMatchers/ASTMatchers.h' 11 12 # Each matcher is documented in one row of the form: 13 # result | name | argA 14 # The subsequent row contains the documentation and is hidden by default, 15 # becoming visible via javascript when the user clicks the matcher name. 16 TD_TEMPLATE=""" 17 <tr><td>%(result)s</td><td class="name" onclick="toggle('%(id)s')"><a name="%(id)sAnchor">%(name)s</a></td><td>%(args)s</td></tr> 18 <tr><td colspan="4" class="doc" id="%(id)s"><pre>%(comment)s</pre></td></tr> 19 """ 20 21 # We categorize the matchers into these three categories in the reference: 22 node_matchers = {} 23 narrowing_matchers = {} 24 traversal_matchers = {} 25 26 # We output multiple rows per matcher if the matcher can be used on multiple 27 # node types. Thus, we need a new id per row to control the documentation 28 # pop-up. ids[name] keeps track of those ids. 29 ids = collections.defaultdict(int) 30 31 # Cache for doxygen urls we have already verified. 32 doxygen_probes = {} 33 34 def esc(text): 35 """Escape any html in the given text.""" 36 text = re.sub(r'&', '&', text) 37 text = re.sub(r'<', '<', text) 38 text = re.sub(r'>', '>', text) 39 def link_if_exists(m): 40 name = m.group(1) 41 url = 'http://clang.llvm.org/doxygen/classclang_1_1%s.html' % name 42 if url not in doxygen_probes: 43 try: 44 print 'Probing %s...' % url 45 urllib2.urlopen(url) 46 doxygen_probes[url] = True 47 except: 48 doxygen_probes[url] = False 49 if doxygen_probes[url]: 50 return r'Matcher<<a href="%s">%s</a>>' % (url, name) 51 else: 52 return m.group(0) 53 text = re.sub( 54 r'Matcher<([^\*&]+)>', link_if_exists, text) 55 return text 56 57 def extract_result_types(comment): 58 """Extracts a list of result types from the given comment. 59 60 We allow annotations in the comment of the matcher to specify what 61 nodes a matcher can match on. Those comments have the form: 62 Usable as: Any Matcher | (Matcher<T1>[, Matcher<t2>[, ...]]) 63 64 Returns ['*'] in case of 'Any Matcher', or ['T1', 'T2', ...]. 65 Returns the empty list if no 'Usable as' specification could be 66 parsed. 67 """ 68 result_types = [] 69 m = re.search(r'Usable as: Any Matcher[\s\n]*$', comment, re.S) 70 if m: 71 return ['*'] 72 while True: 73 m = re.match(r'^(.*)Matcher<([^>]+)>\s*,?[\s\n]*$', comment, re.S) 74 if not m: 75 if re.search(r'Usable as:\s*$', comment): 76 return result_types 77 else: 78 return None 79 result_types += [m.group(2)] 80 comment = m.group(1) 81 82 def strip_doxygen(comment): 83 """Returns the given comment without \-escaped words.""" 84 # If there is only a doxygen keyword in the line, delete the whole line. 85 comment = re.sub(r'^\\[^\s]+\n', r'', comment, flags=re.M) 86 # Delete the doxygen command and the following whitespace. 87 comment = re.sub(r'\\[^\s]+\s+', r'', comment) 88 return comment 89 90 def unify_arguments(args): 91 """Gets rid of anything the user doesn't care about in the argument list.""" 92 args = re.sub(r'internal::', r'', args) 93 args = re.sub(r'const\s+', r'', args) 94 args = re.sub(r'&', r' ', args) 95 args = re.sub(r'(^|\s)M\d?(\s)', r'\1Matcher<*>\2', args) 96 return args 97 98 def add_matcher(result_type, name, args, comment, is_dyncast=False): 99 """Adds a matcher to one of our categories.""" 100 if name == 'id': 101 # FIXME: Figure out whether we want to support the 'id' matcher. 102 return 103 matcher_id = '%s%d' % (name, ids[name]) 104 ids[name] += 1 105 args = unify_arguments(args) 106 matcher_html = TD_TEMPLATE % { 107 'result': esc('Matcher<%s>' % result_type), 108 'name': name, 109 'args': esc(args), 110 'comment': esc(strip_doxygen(comment)), 111 'id': matcher_id, 112 } 113 if is_dyncast: 114 node_matchers[result_type + name] = matcher_html 115 # Use a heuristic to figure out whether a matcher is a narrowing or 116 # traversal matcher. By default, matchers that take other matchers as 117 # arguments (and are not node matchers) do traversal. We specifically 118 # exclude known narrowing matchers that also take other matchers as 119 # arguments. 120 elif ('Matcher<' not in args or 121 name in ['allOf', 'anyOf', 'anything', 'unless']): 122 narrowing_matchers[result_type + name + esc(args)] = matcher_html 123 else: 124 traversal_matchers[result_type + name + esc(args)] = matcher_html 125 126 def act_on_decl(declaration, comment, allowed_types): 127 """Parse the matcher out of the given declaration and comment. 128 129 If 'allowed_types' is set, it contains a list of node types the matcher 130 can match on, as extracted from the static type asserts in the matcher 131 definition. 132 """ 133 if declaration.strip(): 134 # Node matchers are defined by writing: 135 # VariadicDynCastAllOfMatcher<ResultType, ArgumentType> name; 136 m = re.match(r""".*Variadic(?:DynCast)?AllOfMatcher\s*< 137 \s*([^\s,]+)\s*(?:, 138 \s*([^\s>]+)\s*)?> 139 \s*([^\s;]+)\s*;\s*$""", declaration, flags=re.X) 140 if m: 141 result, inner, name = m.groups() 142 if not inner: 143 inner = result 144 add_matcher(result, name, 'Matcher<%s>...' % inner, 145 comment, is_dyncast=True) 146 return 147 148 # Parse the various matcher definition macros. 149 m = re.match(""".*AST_TYPE_MATCHER\( 150 \s*([^\s,]+\s*), 151 \s*([^\s,]+\s*) 152 \)\s*;\s*$""", declaration, flags=re.X) 153 if m: 154 inner, name = m.groups() 155 add_matcher('Type', name, 'Matcher<%s>...' % inner, 156 comment, is_dyncast=True) 157 # FIXME: re-enable once we have implemented casting on the TypeLoc 158 # hierarchy. 159 # add_matcher('TypeLoc', '%sLoc' % name, 'Matcher<%sLoc>...' % inner, 160 # comment, is_dyncast=True) 161 return 162 163 m = re.match(""".*AST_TYPE(LOC)?_TRAVERSE_MATCHER\( 164 \s*([^\s,]+\s*), 165 \s*(?:[^\s,]+\s*), 166 \s*AST_POLYMORPHIC_SUPPORTED_TYPES\(([^)]*)\) 167 \)\s*;\s*$""", declaration, flags=re.X) 168 if m: 169 loc, name, results = m.groups()[0:3] 170 result_types = [r.strip() for r in results.split(',')] 171 172 comment_result_types = extract_result_types(comment) 173 if (comment_result_types and 174 sorted(result_types) != sorted(comment_result_types)): 175 raise Exception('Inconsistent documentation for: %s' % name) 176 for result_type in result_types: 177 add_matcher(result_type, name, 'Matcher<Type>', comment) 178 if loc: 179 add_matcher('%sLoc' % result_type, '%sLoc' % name, 'Matcher<TypeLoc>', 180 comment) 181 return 182 183 m = re.match(r"""^\s*AST_POLYMORPHIC_MATCHER(_P)?(.?)(?:_OVERLOAD)?\( 184 \s*([^\s,]+)\s*, 185 \s*AST_POLYMORPHIC_SUPPORTED_TYPES\(([^)]*)\) 186 (?:,\s*([^\s,]+)\s* 187 ,\s*([^\s,]+)\s*)? 188 (?:,\s*([^\s,]+)\s* 189 ,\s*([^\s,]+)\s*)? 190 (?:,\s*\d+\s*)? 191 \)\s*{\s*$""", declaration, flags=re.X) 192 193 if m: 194 p, n, name, results = m.groups()[0:4] 195 args = m.groups()[4:] 196 result_types = [r.strip() for r in results.split(',')] 197 if allowed_types and allowed_types != result_types: 198 raise Exception('Inconsistent documentation for: %s' % name) 199 if n not in ['', '2']: 200 raise Exception('Cannot parse "%s"' % declaration) 201 args = ', '.join('%s %s' % (args[i], args[i+1]) 202 for i in range(0, len(args), 2) if args[i]) 203 for result_type in result_types: 204 add_matcher(result_type, name, args, comment) 205 return 206 207 m = re.match(r"""^\s*AST_MATCHER_FUNCTION(_P)?(.?)(?:_OVERLOAD)?\( 208 (?:\s*([^\s,]+)\s*,)? 209 \s*([^\s,]+)\s* 210 (?:,\s*([^\s,]+)\s* 211 ,\s*([^\s,]+)\s*)? 212 (?:,\s*([^\s,]+)\s* 213 ,\s*([^\s,]+)\s*)? 214 (?:,\s*\d+\s*)? 215 \)\s*{\s*$""", declaration, flags=re.X) 216 if m: 217 p, n, result, name = m.groups()[0:4] 218 args = m.groups()[4:] 219 if n not in ['', '2']: 220 raise Exception('Cannot parse "%s"' % declaration) 221 args = ', '.join('%s %s' % (args[i], args[i+1]) 222 for i in range(0, len(args), 2) if args[i]) 223 add_matcher(result, name, args, comment) 224 return 225 226 m = re.match(r"""^\s*AST_MATCHER(_P)?(.?)(?:_OVERLOAD)?\( 227 (?:\s*([^\s,]+)\s*,)? 228 \s*([^\s,]+)\s* 229 (?:,\s*([^\s,]+)\s* 230 ,\s*([^\s,]+)\s*)? 231 (?:,\s*([^\s,]+)\s* 232 ,\s*([^\s,]+)\s*)? 233 (?:,\s*\d+\s*)? 234 \)\s*{\s*$""", declaration, flags=re.X) 235 if m: 236 p, n, result, name = m.groups()[0:4] 237 args = m.groups()[4:] 238 if not result: 239 if not allowed_types: 240 raise Exception('Did not find allowed result types for: %s' % name) 241 result_types = allowed_types 242 else: 243 result_types = [result] 244 if n not in ['', '2']: 245 raise Exception('Cannot parse "%s"' % declaration) 246 args = ', '.join('%s %s' % (args[i], args[i+1]) 247 for i in range(0, len(args), 2) if args[i]) 248 for result_type in result_types: 249 add_matcher(result_type, name, args, comment) 250 return 251 252 # Parse ArgumentAdapting matchers. 253 m = re.match( 254 r"""^.*ArgumentAdaptingMatcherFunc<.*>\s*(?:LLVM_ATTRIBUTE_UNUSED\s*) 255 ([a-zA-Z]*)\s*=\s*{};$""", 256 declaration, flags=re.X) 257 if m: 258 name = m.groups()[0] 259 add_matcher('*', name, 'Matcher<*>', comment) 260 return 261 262 # Parse Variadic operator matchers. 263 m = re.match( 264 r"""^.*VariadicOperatorMatcherFunc\s*<\s*([^,]+),\s*([^\s>]+)\s*>\s* 265 ([a-zA-Z]*)\s*=\s*{.*};$""", 266 declaration, flags=re.X) 267 if m: 268 min_args, max_args, name = m.groups()[:3] 269 if max_args == '1': 270 add_matcher('*', name, 'Matcher<*>', comment) 271 return 272 elif max_args == 'UINT_MAX': 273 add_matcher('*', name, 'Matcher<*>, ..., Matcher<*>', comment) 274 return 275 276 277 # Parse free standing matcher functions, like: 278 # Matcher<ResultType> Name(Matcher<ArgumentType> InnerMatcher) { 279 m = re.match(r"""^\s*(.*)\s+ 280 ([^\s\(]+)\s*\( 281 (.*) 282 \)\s*{""", declaration, re.X) 283 if m: 284 result, name, args = m.groups() 285 args = ', '.join(p.strip() for p in args.split(',')) 286 m = re.match(r'.*\s+internal::(Bindable)?Matcher<([^>]+)>$', result) 287 if m: 288 result_types = [m.group(2)] 289 else: 290 result_types = extract_result_types(comment) 291 if not result_types: 292 if not comment: 293 # Only overloads don't have their own doxygen comments; ignore those. 294 print 'Ignoring "%s"' % name 295 else: 296 print 'Cannot determine result type for "%s"' % name 297 else: 298 for result_type in result_types: 299 add_matcher(result_type, name, args, comment) 300 else: 301 print '*** Unparsable: "' + declaration + '" ***' 302 303 def sort_table(matcher_type, matcher_map): 304 """Returns the sorted html table for the given row map.""" 305 table = '' 306 for key in sorted(matcher_map.keys()): 307 table += matcher_map[key] + '\n' 308 return ('<!-- START_%(type)s_MATCHERS -->\n' + 309 '%(table)s' + 310 '<!--END_%(type)s_MATCHERS -->') % { 311 'type': matcher_type, 312 'table': table, 313 } 314 315 # Parse the ast matchers. 316 # We alternate between two modes: 317 # body = True: We parse the definition of a matcher. We need 318 # to parse the full definition before adding a matcher, as the 319 # definition might contain static asserts that specify the result 320 # type. 321 # body = False: We parse the comments and declaration of the matcher. 322 comment = '' 323 declaration = '' 324 allowed_types = [] 325 body = False 326 for line in open(MATCHERS_FILE).read().splitlines(): 327 if body: 328 if line.strip() and line[0] == '}': 329 if declaration: 330 act_on_decl(declaration, comment, allowed_types) 331 comment = '' 332 declaration = '' 333 allowed_types = [] 334 body = False 335 else: 336 m = re.search(r'is_base_of<([^,]+), NodeType>', line) 337 if m and m.group(1): 338 allowed_types += [m.group(1)] 339 continue 340 if line.strip() and line.lstrip()[0] == '/': 341 comment += re.sub(r'/+\s?', '', line) + '\n' 342 else: 343 declaration += ' ' + line 344 if ((not line.strip()) or 345 line.rstrip()[-1] == ';' or 346 (line.rstrip()[-1] == '{' and line.rstrip()[-3:] != '= {')): 347 if line.strip() and line.rstrip()[-1] == '{': 348 body = True 349 else: 350 act_on_decl(declaration, comment, allowed_types) 351 comment = '' 352 declaration = '' 353 allowed_types = [] 354 355 node_matcher_table = sort_table('DECL', node_matchers) 356 narrowing_matcher_table = sort_table('NARROWING', narrowing_matchers) 357 traversal_matcher_table = sort_table('TRAVERSAL', traversal_matchers) 358 359 reference = open('../LibASTMatchersReference.html').read() 360 reference = re.sub(r'<!-- START_DECL_MATCHERS.*END_DECL_MATCHERS -->', 361 '%s', reference, flags=re.S) % node_matcher_table 362 reference = re.sub(r'<!-- START_NARROWING_MATCHERS.*END_NARROWING_MATCHERS -->', 363 '%s', reference, flags=re.S) % narrowing_matcher_table 364 reference = re.sub(r'<!-- START_TRAVERSAL_MATCHERS.*END_TRAVERSAL_MATCHERS -->', 365 '%s', reference, flags=re.S) % traversal_matcher_table 366 367 with open('../LibASTMatchersReference.html', 'wb') as output: 368 output.write(reference) 369 370
