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 6import collections 7import re 8try: 9 from urllib.request import urlopen 10except ImportError: 11 from urllib2 import urlopen 12 13MATCHERS_FILE = '../../include/clang/ASTMatchers/ASTMatchers.h' 14 15# Each matcher is documented in one row of the form: 16# result | name | argA 17# The subsequent row contains the documentation and is hidden by default, 18# becoming visible via javascript when the user clicks the matcher name. 19TD_TEMPLATE=""" 20<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> 21<tr><td colspan="4" class="doc" id="%(id)s"><pre>%(comment)s</pre></td></tr> 22""" 23 24# We categorize the matchers into these three categories in the reference: 25node_matchers = {} 26narrowing_matchers = {} 27traversal_matchers = {} 28 29# We output multiple rows per matcher if the matcher can be used on multiple 30# node types. Thus, we need a new id per row to control the documentation 31# pop-up. ids[name] keeps track of those ids. 32ids = collections.defaultdict(int) 33 34# Cache for doxygen urls we have already verified. 35doxygen_probes = {} 36 37def esc(text): 38 """Escape any html in the given text.""" 39 text = re.sub(r'&', '&', text) 40 text = re.sub(r'<', '<', text) 41 text = re.sub(r'>', '>', text) 42 def link_if_exists(m): 43 name = m.group(1) 44 url = 'https://clang.llvm.org/doxygen/classclang_1_1%s.html' % name 45 if url not in doxygen_probes: 46 try: 47 print('Probing %s...' % url) 48 urlopen(url) 49 doxygen_probes[url] = True 50 except: 51 doxygen_probes[url] = False 52 if doxygen_probes[url]: 53 return r'Matcher<<a href="%s">%s</a>>' % (url, name) 54 else: 55 return m.group(0) 56 text = re.sub( 57 r'Matcher<([^\*&]+)>', link_if_exists, text) 58 return text 59 60def extract_result_types(comment): 61 """Extracts a list of result types from the given comment. 62 63 We allow annotations in the comment of the matcher to specify what 64 nodes a matcher can match on. Those comments have the form: 65 Usable as: Any Matcher | (Matcher<T1>[, Matcher<t2>[, ...]]) 66 67 Returns ['*'] in case of 'Any Matcher', or ['T1', 'T2', ...]. 68 Returns the empty list if no 'Usable as' specification could be 69 parsed. 70 """ 71 result_types = [] 72 m = re.search(r'Usable as: Any Matcher[\s\n]*$', comment, re.S) 73 if m: 74 return ['*'] 75 while True: 76 m = re.match(r'^(.*)Matcher<([^>]+)>\s*,?[\s\n]*$', comment, re.S) 77 if not m: 78 if re.search(r'Usable as:\s*$', comment): 79 return result_types 80 else: 81 return None 82 result_types += [m.group(2)] 83 comment = m.group(1) 84 85def strip_doxygen(comment): 86 """Returns the given comment without \-escaped words.""" 87 # If there is only a doxygen keyword in the line, delete the whole line. 88 comment = re.sub(r'^\\[^\s]+\n', r'', comment, flags=re.M) 89 90 # If there is a doxygen \see command, change the \see prefix into "See also:". 91 # FIXME: it would be better to turn this into a link to the target instead. 92 comment = re.sub(r'\\see', r'See also:', comment) 93 94 # Delete the doxygen command and the following whitespace. 95 comment = re.sub(r'\\[^\s]+\s+', r'', comment) 96 return comment 97 98def unify_arguments(args): 99 """Gets rid of anything the user doesn't care about in the argument list.""" 100 args = re.sub(r'internal::', r'', args) 101 args = re.sub(r'extern const\s+(.*)&', r'\1 ', args) 102 args = re.sub(r'&', r' ', args) 103 args = re.sub(r'(^|\s)M\d?(\s)', r'\1Matcher<*>\2', args) 104 return args 105 106def add_matcher(result_type, name, args, comment, is_dyncast=False): 107 """Adds a matcher to one of our categories.""" 108 if name == 'id': 109 # FIXME: Figure out whether we want to support the 'id' matcher. 110 return 111 matcher_id = '%s%d' % (name, ids[name]) 112 ids[name] += 1 113 args = unify_arguments(args) 114 matcher_html = TD_TEMPLATE % { 115 'result': esc('Matcher<%s>' % result_type), 116 'name': name, 117 'args': esc(args), 118 'comment': esc(strip_doxygen(comment)), 119 'id': matcher_id, 120 } 121 if is_dyncast: 122 node_matchers[result_type + name] = matcher_html 123 # Use a heuristic to figure out whether a matcher is a narrowing or 124 # traversal matcher. By default, matchers that take other matchers as 125 # arguments (and are not node matchers) do traversal. We specifically 126 # exclude known narrowing matchers that also take other matchers as 127 # arguments. 128 elif ('Matcher<' not in args or 129 name in ['allOf', 'anyOf', 'anything', 'unless']): 130 narrowing_matchers[result_type + name + esc(args)] = matcher_html 131 else: 132 traversal_matchers[result_type + name + esc(args)] = matcher_html 133 134def act_on_decl(declaration, comment, allowed_types): 135 """Parse the matcher out of the given declaration and comment. 136 137 If 'allowed_types' is set, it contains a list of node types the matcher 138 can match on, as extracted from the static type asserts in the matcher 139 definition. 140 """ 141 if declaration.strip(): 142 # Node matchers are defined by writing: 143 # VariadicDynCastAllOfMatcher<ResultType, ArgumentType> name; 144 m = re.match(r""".*Variadic(?:DynCast)?AllOfMatcher\s*< 145 \s*([^\s,]+)\s*(?:, 146 \s*([^\s>]+)\s*)?> 147 \s*([^\s;]+)\s*;\s*$""", declaration, flags=re.X) 148 if m: 149 result, inner, name = m.groups() 150 if not inner: 151 inner = result 152 add_matcher(result, name, 'Matcher<%s>...' % inner, 153 comment, is_dyncast=True) 154 return 155 156 # Special case of type matchers: 157 # AstTypeMatcher<ArgumentType> name 158 m = re.match(r""".*AstTypeMatcher\s*< 159 \s*([^\s>]+)\s*> 160 \s*([^\s;]+)\s*;\s*$""", declaration, flags=re.X) 161 if m: 162 inner, name = m.groups() 163 add_matcher('Type', name, 'Matcher<%s>...' % inner, 164 comment, is_dyncast=True) 165 # FIXME: re-enable once we have implemented casting on the TypeLoc 166 # hierarchy. 167 # add_matcher('TypeLoc', '%sLoc' % name, 'Matcher<%sLoc>...' % inner, 168 # comment, is_dyncast=True) 169 return 170 171 # Parse the various matcher definition macros. 172 m = re.match(""".*AST_TYPE(LOC)?_TRAVERSE_MATCHER(?:_DECL)?\( 173 \s*([^\s,]+\s*), 174 \s*(?:[^\s,]+\s*), 175 \s*AST_POLYMORPHIC_SUPPORTED_TYPES\(([^)]*)\) 176 \)\s*;\s*$""", declaration, flags=re.X) 177 if m: 178 loc, name, results = m.groups()[0:3] 179 result_types = [r.strip() for r in results.split(',')] 180 181 comment_result_types = extract_result_types(comment) 182 if (comment_result_types and 183 sorted(result_types) != sorted(comment_result_types)): 184 raise Exception('Inconsistent documentation for: %s' % name) 185 for result_type in result_types: 186 add_matcher(result_type, name, 'Matcher<Type>', comment) 187 # if loc: 188 # add_matcher('%sLoc' % result_type, '%sLoc' % name, 'Matcher<TypeLoc>', 189 # comment) 190 return 191 192 m = re.match(r"""^\s*AST_POLYMORPHIC_MATCHER(_P)?(.?)(?:_OVERLOAD)?\( 193 \s*([^\s,]+)\s*, 194 \s*AST_POLYMORPHIC_SUPPORTED_TYPES\(([^)]*)\) 195 (?:,\s*([^\s,]+)\s* 196 ,\s*([^\s,]+)\s*)? 197 (?:,\s*([^\s,]+)\s* 198 ,\s*([^\s,]+)\s*)? 199 (?:,\s*\d+\s*)? 200 \)\s*{\s*$""", declaration, flags=re.X) 201 202 if m: 203 p, n, name, results = m.groups()[0:4] 204 args = m.groups()[4:] 205 result_types = [r.strip() for r in results.split(',')] 206 if allowed_types and allowed_types != result_types: 207 raise Exception('Inconsistent documentation for: %s' % name) 208 if n not in ['', '2']: 209 raise Exception('Cannot parse "%s"' % declaration) 210 args = ', '.join('%s %s' % (args[i], args[i+1]) 211 for i in range(0, len(args), 2) if args[i]) 212 for result_type in result_types: 213 add_matcher(result_type, name, args, comment) 214 return 215 216 m = re.match(r"""^\s*AST_MATCHER_FUNCTION(_P)?(.?)(?:_OVERLOAD)?\( 217 (?:\s*([^\s,]+)\s*,)? 218 \s*([^\s,]+)\s* 219 (?:,\s*([^\s,]+)\s* 220 ,\s*([^\s,]+)\s*)? 221 (?:,\s*([^\s,]+)\s* 222 ,\s*([^\s,]+)\s*)? 223 (?:,\s*\d+\s*)? 224 \)\s*{\s*$""", declaration, flags=re.X) 225 if m: 226 p, n, result, name = m.groups()[0:4] 227 args = m.groups()[4:] 228 if n not in ['', '2']: 229 raise Exception('Cannot parse "%s"' % declaration) 230 args = ', '.join('%s %s' % (args[i], args[i+1]) 231 for i in range(0, len(args), 2) if args[i]) 232 add_matcher(result, name, args, comment) 233 return 234 235 m = re.match(r"""^\s*AST_MATCHER(_P)?(.?)(?:_OVERLOAD)?\( 236 (?:\s*([^\s,]+)\s*,)? 237 \s*([^\s,]+)\s* 238 (?:,\s*([^,]+)\s* 239 ,\s*([^\s,]+)\s*)? 240 (?:,\s*([^\s,]+)\s* 241 ,\s*([^\s,]+)\s*)? 242 (?:,\s*\d+\s*)? 243 \)\s*{""", declaration, flags=re.X) 244 if m: 245 p, n, result, name = m.groups()[0:4] 246 args = m.groups()[4:] 247 if not result: 248 if not allowed_types: 249 raise Exception('Did not find allowed result types for: %s' % name) 250 result_types = allowed_types 251 else: 252 result_types = [result] 253 if n not in ['', '2']: 254 raise Exception('Cannot parse "%s"' % declaration) 255 args = ', '.join('%s %s' % (args[i], args[i+1]) 256 for i in range(0, len(args), 2) if args[i]) 257 for result_type in result_types: 258 add_matcher(result_type, name, args, comment) 259 return 260 261 # Parse ArgumentAdapting matchers. 262 m = re.match( 263 r"""^.*ArgumentAdaptingMatcherFunc<.*>\s* 264 ([a-zA-Z]*);$""", 265 declaration, flags=re.X) 266 if m: 267 name = m.groups()[0] 268 add_matcher('*', name, 'Matcher<*>', comment) 269 return 270 271 # Parse Variadic functions. 272 m = re.match( 273 r"""^.*internal::VariadicFunction\s*<\s*([^,]+),\s*([^,]+),\s*[^>]+>\s* 274 ([a-zA-Z]*);$""", 275 declaration, flags=re.X) 276 if m: 277 result, arg, name = m.groups()[:3] 278 add_matcher(result, name, '%s, ..., %s' % (arg, arg), comment) 279 return 280 281 # Parse Variadic operator matchers. 282 m = re.match( 283 r"""^.*VariadicOperatorMatcherFunc\s*<\s*([^,]+),\s*([^\s]+)\s*>\s* 284 ([a-zA-Z]*);$""", 285 declaration, flags=re.X) 286 if m: 287 min_args, max_args, name = m.groups()[:3] 288 if max_args == '1': 289 add_matcher('*', name, 'Matcher<*>', comment) 290 return 291 elif max_args == 'std::numeric_limits<unsigned>::max()': 292 add_matcher('*', name, 'Matcher<*>, ..., Matcher<*>', comment) 293 return 294 295 296 # Parse free standing matcher functions, like: 297 # Matcher<ResultType> Name(Matcher<ArgumentType> InnerMatcher) { 298 m = re.match(r"""^\s*(.*)\s+ 299 ([^\s\(]+)\s*\( 300 (.*) 301 \)\s*{""", declaration, re.X) 302 if m: 303 result, name, args = m.groups() 304 args = ', '.join(p.strip() for p in args.split(',')) 305 m = re.match(r'.*\s+internal::(Bindable)?Matcher<([^>]+)>$', result) 306 if m: 307 result_types = [m.group(2)] 308 else: 309 result_types = extract_result_types(comment) 310 if not result_types: 311 if not comment: 312 # Only overloads don't have their own doxygen comments; ignore those. 313 print('Ignoring "%s"' % name) 314 else: 315 print('Cannot determine result type for "%s"' % name) 316 else: 317 for result_type in result_types: 318 add_matcher(result_type, name, args, comment) 319 else: 320 print('*** Unparsable: "' + declaration + '" ***') 321 322def sort_table(matcher_type, matcher_map): 323 """Returns the sorted html table for the given row map.""" 324 table = '' 325 for key in sorted(matcher_map.keys()): 326 table += matcher_map[key] + '\n' 327 return ('<!-- START_%(type)s_MATCHERS -->\n' + 328 '%(table)s' + 329 '<!--END_%(type)s_MATCHERS -->') % { 330 'type': matcher_type, 331 'table': table, 332 } 333 334# Parse the ast matchers. 335# We alternate between two modes: 336# body = True: We parse the definition of a matcher. We need 337# to parse the full definition before adding a matcher, as the 338# definition might contain static asserts that specify the result 339# type. 340# body = False: We parse the comments and declaration of the matcher. 341comment = '' 342declaration = '' 343allowed_types = [] 344body = False 345for line in open(MATCHERS_FILE).read().splitlines(): 346 if body: 347 if line.strip() and line[0] == '}': 348 if declaration: 349 act_on_decl(declaration, comment, allowed_types) 350 comment = '' 351 declaration = '' 352 allowed_types = [] 353 body = False 354 else: 355 m = re.search(r'is_base_of<([^,]+), NodeType>', line) 356 if m and m.group(1): 357 allowed_types += [m.group(1)] 358 continue 359 if line.strip() and line.lstrip()[0] == '/': 360 comment += re.sub(r'^/+\s?', '', line) + '\n' 361 else: 362 declaration += ' ' + line 363 if ((not line.strip()) or 364 line.rstrip()[-1] == ';' or 365 (line.rstrip()[-1] == '{' and line.rstrip()[-3:] != '= {')): 366 if line.strip() and line.rstrip()[-1] == '{': 367 body = True 368 else: 369 act_on_decl(declaration, comment, allowed_types) 370 comment = '' 371 declaration = '' 372 allowed_types = [] 373 374node_matcher_table = sort_table('DECL', node_matchers) 375narrowing_matcher_table = sort_table('NARROWING', narrowing_matchers) 376traversal_matcher_table = sort_table('TRAVERSAL', traversal_matchers) 377 378reference = open('../LibASTMatchersReference.html').read() 379reference = re.sub(r'<!-- START_DECL_MATCHERS.*END_DECL_MATCHERS -->', 380 node_matcher_table, reference, flags=re.S) 381reference = re.sub(r'<!-- START_NARROWING_MATCHERS.*END_NARROWING_MATCHERS -->', 382 narrowing_matcher_table, reference, flags=re.S) 383reference = re.sub(r'<!-- START_TRAVERSAL_MATCHERS.*END_TRAVERSAL_MATCHERS -->', 384 traversal_matcher_table, reference, flags=re.S) 385 386with open('../LibASTMatchersReference.html', 'wb') as output: 387 output.write(reference) 388 389