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'&', '&amp;', text)
40  text = re.sub(r'<', '&lt;', text)
41  text = re.sub(r'>', '&gt;', 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&lt;<a href="%s">%s</a>&gt;' % (url, name)
54    else:
55      return m.group(0)
56  text = re.sub(
57    r'Matcher&lt;([^\*&]+)&gt;', 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*$', comment):
79        return result_types
80      else:
81        return None
82    result_types += [m.group(2)]
83    comment = m.group(1)
84
85def strip_doxygen(comment):
86  """Returns the given comment without \-escaped words."""
87  # If there is only a doxygen keyword in the line, delete the whole line.
88  comment = re.sub(r'^\\[^\s]+\n', r'', comment, flags=re.M)
89
90  # If there is a doxygen \see command, change the \see prefix into "See also:".
91  # FIXME: it would be better to turn this into a link to the target instead.
92  comment = re.sub(r'\\see', r'See also:', comment)
93
94  # Delete the doxygen command and the following whitespace.
95  comment = re.sub(r'\\[^\s]+\s+', r'', comment)
96  return comment
97
98def unify_arguments(args):
99  """Gets rid of anything the user doesn't care about in the argument list."""
100  args = re.sub(r'internal::', r'', args)
101  args = re.sub(r'extern const\s+(.*)&', r'\1 ', args)
102  args = re.sub(r'&', r' ', args)
103  args = re.sub(r'(^|\s)M\d?(\s)', r'\1Matcher<*>\2', args)
104  args = re.sub(r'BindableMatcher', r'Matcher', args)
105  args = re.sub(r'const Matcher', r'Matcher', args)
106  return args
107
108def unify_type(result_type):
109  """Gets rid of anything the user doesn't care about in the type name."""
110  result_type = re.sub(r'^internal::(Bindable)?Matcher<([a-zA-Z_][a-zA-Z0-9_]*)>$', r'\2', result_type)
111  return result_type
112
113def add_matcher(result_type, name, args, comment, is_dyncast=False):
114  """Adds a matcher to one of our categories."""
115  if name == 'id':
116     # FIXME: Figure out whether we want to support the 'id' matcher.
117     return
118  matcher_id = '%s%d' % (name, ids[name])
119  ids[name] += 1
120  args = unify_arguments(args)
121  result_type = unify_type(result_type)
122  matcher_html = TD_TEMPLATE % {
123    'result': esc('Matcher<%s>' % result_type),
124    'name': name,
125    'args': esc(args),
126    'comment': esc(strip_doxygen(comment)),
127    'id': matcher_id,
128  }
129  if is_dyncast:
130    dict = node_matchers
131    lookup = result_type + name
132  # Use a heuristic to figure out whether a matcher is a narrowing or
133  # traversal matcher. By default, matchers that take other matchers as
134  # arguments (and are not node matchers) do traversal. We specifically
135  # exclude known narrowing matchers that also take other matchers as
136  # arguments.
137  elif ('Matcher<' not in args or
138        name in ['allOf', 'anyOf', 'anything', 'unless']):
139    dict = narrowing_matchers
140    lookup = result_type + name + esc(args)
141  else:
142    dict = traversal_matchers
143    lookup = result_type + name + esc(args)
144
145  if dict.get(lookup) is None or len(dict.get(lookup)) < len(matcher_html):
146    dict[lookup] = matcher_html
147
148def act_on_decl(declaration, comment, allowed_types):
149  """Parse the matcher out of the given declaration and comment.
150
151     If 'allowed_types' is set, it contains a list of node types the matcher
152     can match on, as extracted from the static type asserts in the matcher
153     definition.
154  """
155  if declaration.strip():
156
157    if re.match(r'^\s?(#|namespace|using)', declaration): return
158
159    # Node matchers are defined by writing:
160    #   VariadicDynCastAllOfMatcher<ResultType, ArgumentType> name;
161    m = re.match(r""".*Variadic(?:DynCast)?AllOfMatcher\s*<
162                       \s*([^\s,]+)\s*(?:,
163                       \s*([^\s>]+)\s*)?>
164                       \s*([^\s;]+)\s*;\s*$""", declaration, flags=re.X)
165    if m:
166      result, inner, name = m.groups()
167      if not inner:
168        inner = result
169      add_matcher(result, name, 'Matcher<%s>...' % inner,
170                  comment, is_dyncast=True)
171      return
172
173    # Special case of type matchers:
174    #   AstTypeMatcher<ArgumentType> name
175    m = re.match(r""".*AstTypeMatcher\s*<
176                       \s*([^\s>]+)\s*>
177                       \s*([^\s;]+)\s*;\s*$""", declaration, flags=re.X)
178    if m:
179      inner, name = m.groups()
180      add_matcher('Type', name, 'Matcher<%s>...' % inner,
181                  comment, is_dyncast=True)
182      # FIXME: re-enable once we have implemented casting on the TypeLoc
183      # hierarchy.
184      # add_matcher('TypeLoc', '%sLoc' % name, 'Matcher<%sLoc>...' % inner,
185      #             comment, is_dyncast=True)
186      return
187
188    # Parse the various matcher definition macros.
189    m = re.match(""".*AST_TYPE(LOC)?_TRAVERSE_MATCHER(?:_DECL)?\(
190                       \s*([^\s,]+\s*),
191                       \s*(?:[^\s,]+\s*),
192                       \s*AST_POLYMORPHIC_SUPPORTED_TYPES\(([^)]*)\)
193                     \)\s*;\s*$""", declaration, flags=re.X)
194    if m:
195      loc, name, results = m.groups()[0:3]
196      result_types = [r.strip() for r in results.split(',')]
197
198      comment_result_types = extract_result_types(comment)
199      if (comment_result_types and
200          sorted(result_types) != sorted(comment_result_types)):
201        raise Exception('Inconsistent documentation for: %s' % name)
202      for result_type in result_types:
203        add_matcher(result_type, name, 'Matcher<Type>', comment)
204        # if loc:
205        #   add_matcher('%sLoc' % result_type, '%sLoc' % name, 'Matcher<TypeLoc>',
206        #               comment)
207      return
208
209    m = re.match(r"""^\s*AST_POLYMORPHIC_MATCHER(_P)?(.?)(?:_OVERLOAD)?\(
210                          \s*([^\s,]+)\s*,
211                          \s*AST_POLYMORPHIC_SUPPORTED_TYPES\(([^)]*)\)
212                       (?:,\s*([^\s,]+)\s*
213                          ,\s*([^\s,]+)\s*)?
214                       (?:,\s*([^\s,]+)\s*
215                          ,\s*([^\s,]+)\s*)?
216                       (?:,\s*\d+\s*)?
217                      \)\s*{\s*$""", declaration, flags=re.X)
218
219    if m:
220      p, n, name, results = m.groups()[0:4]
221      args = m.groups()[4:]
222      result_types = [r.strip() for r in results.split(',')]
223      if allowed_types and allowed_types != result_types:
224        raise Exception('Inconsistent documentation for: %s' % name)
225      if n not in ['', '2']:
226        raise Exception('Cannot parse "%s"' % declaration)
227      args = ', '.join('%s %s' % (args[i], args[i+1])
228                       for i in range(0, len(args), 2) if args[i])
229      for result_type in result_types:
230        add_matcher(result_type, name, args, comment)
231      return
232
233    m = re.match(r"""^\s*AST_POLYMORPHIC_MATCHER_REGEX(?:_OVERLOAD)?\(
234                          \s*([^\s,]+)\s*,
235                          \s*AST_POLYMORPHIC_SUPPORTED_TYPES\(([^)]*)\),
236                          \s*([^\s,]+)\s*
237                       (?:,\s*\d+\s*)?
238                      \)\s*{\s*$""", declaration, flags=re.X)
239
240    if m:
241      name, results, arg_name = m.groups()[0:3]
242      result_types = [r.strip() for r in results.split(',')]
243      if allowed_types and allowed_types != result_types:
244        raise Exception('Inconsistent documentation for: %s' % name)
245      arg = "StringRef %s, Regex::RegexFlags Flags = NoFlags" % arg_name
246      comment += """
247If the matcher is used in clang-query, RegexFlags parameter
248should be passed as a quoted string. e.g: "NoFlags".
249Flags can be combined with '|' example \"IgnoreCase | BasicRegex\"
250"""
251      for result_type in result_types:
252        add_matcher(result_type, name, arg, comment)
253      return
254
255    m = re.match(r"""^\s*AST_MATCHER_FUNCTION(_P)?(.?)(?:_OVERLOAD)?\(
256                       (?:\s*([^\s,]+)\s*,)?
257                          \s*([^\s,]+)\s*
258                       (?:,\s*([^\s,]+)\s*
259                          ,\s*([^\s,]+)\s*)?
260                       (?:,\s*([^\s,]+)\s*
261                          ,\s*([^\s,]+)\s*)?
262                       (?:,\s*\d+\s*)?
263                      \)\s*{\s*$""", declaration, flags=re.X)
264    if m:
265      p, n, result, name = m.groups()[0:4]
266      args = m.groups()[4:]
267      if n not in ['', '2']:
268        raise Exception('Cannot parse "%s"' % declaration)
269      args = ', '.join('%s %s' % (args[i], args[i+1])
270                       for i in range(0, len(args), 2) if args[i])
271      add_matcher(result, name, args, comment)
272      return
273
274    m = re.match(r"""^\s*AST_MATCHER(_P)?(.?)(?:_OVERLOAD)?\(
275                       (?:\s*([^\s,]+)\s*,)?
276                          \s*([^\s,]+)\s*
277                       (?:,\s*([^,]+)\s*
278                          ,\s*([^\s,]+)\s*)?
279                       (?:,\s*([^\s,]+)\s*
280                          ,\s*([^\s,]+)\s*)?
281                       (?:,\s*\d+\s*)?
282                      \)\s*{""", declaration, flags=re.X)
283    if m:
284      p, n, result, name = m.groups()[0:4]
285      args = m.groups()[4:]
286      if not result:
287        if not allowed_types:
288          raise Exception('Did not find allowed result types for: %s' % name)
289        result_types = allowed_types
290      else:
291        result_types = [result]
292      if n not in ['', '2']:
293        raise Exception('Cannot parse "%s"' % declaration)
294      args = ', '.join('%s %s' % (args[i], args[i+1])
295                       for i in range(0, len(args), 2) if args[i])
296      for result_type in result_types:
297        add_matcher(result_type, name, args, comment)
298      return
299
300    m = re.match(r"""^\s*AST_MATCHER_REGEX(?:_OVERLOAD)?\(
301                       \s*([^\s,]+)\s*,
302                       \s*([^\s,]+)\s*,
303                       \s*([^\s,]+)\s*
304                       (?:,\s*\d+\s*)?
305                      \)\s*{""", declaration, flags=re.X)
306    if m:
307      result, name, arg_name = m.groups()[0:3]
308      if not result:
309        if not allowed_types:
310          raise Exception('Did not find allowed result types for: %s' % name)
311        result_types = allowed_types
312      else:
313        result_types = [result]
314      arg = "StringRef %s, Regex::RegexFlags Flags = NoFlags" % arg_name
315      comment += """
316If the matcher is used in clang-query, RegexFlags parameter
317should be passed as a quoted string. e.g: "NoFlags".
318Flags can be combined with '|' example \"IgnoreCase | BasicRegex\"
319"""
320
321      for result_type in result_types:
322        add_matcher(result_type, name, arg, comment)
323      return
324
325    # Parse ArgumentAdapting matchers.
326    m = re.match(
327        r"""^.*ArgumentAdaptingMatcherFunc<.*>\s*
328              ([a-zA-Z]*);$""",
329        declaration, flags=re.X)
330    if m:
331      name = m.groups()[0]
332      add_matcher('*', name, 'Matcher<*>', comment)
333      return
334
335    # Parse Variadic functions.
336    m = re.match(
337        r"""^.*internal::VariadicFunction\s*<\s*([^,]+),\s*([^,]+),\s*[^>]+>\s*
338              ([a-zA-Z]*);$""",
339        declaration, flags=re.X)
340    if m:
341      result, arg, name = m.groups()[:3]
342      add_matcher(result, name, '%s, ..., %s' % (arg, arg), comment)
343      return
344
345    m = re.match(
346        r"""^.*internal::VariadicFunction\s*<\s*
347              internal::PolymorphicMatcherWithParam1<[\S\s]+
348              AST_POLYMORPHIC_SUPPORTED_TYPES\(([^)]*)\)>,\s*([^,]+),
349              \s*[^>]+>\s*([a-zA-Z]*);$""",
350        declaration, flags=re.X)
351
352    if m:
353      results, arg, name = m.groups()[:3]
354
355      result_types = [r.strip() for r in results.split(',')]
356      for result_type in result_types:
357        add_matcher(result_type, name, '%s, ..., %s' % (arg, arg), comment)
358      return
359
360
361    # Parse Variadic operator matchers.
362    m = re.match(
363        r"""^.*VariadicOperatorMatcherFunc\s*<\s*([^,]+),\s*([^\s]+)\s*>\s*
364              ([a-zA-Z]*);$""",
365        declaration, flags=re.X)
366    if m:
367      min_args, max_args, name = m.groups()[:3]
368      if max_args == '1':
369        add_matcher('*', name, 'Matcher<*>', comment)
370        return
371      elif max_args == 'std::numeric_limits<unsigned>::max()':
372        add_matcher('*', name, 'Matcher<*>, ..., Matcher<*>', comment)
373        return
374
375
376    # Parse free standing matcher functions, like:
377    #   Matcher<ResultType> Name(Matcher<ArgumentType> InnerMatcher) {
378    m = re.match(r"""^\s*(?:template\s+<\s*(?:class|typename)\s+(.+)\s*>\s+)?
379                     (.*)\s+
380                     ([^\s\(]+)\s*\(
381                     (.*)
382                     \)\s*{""", declaration, re.X)
383    if m:
384      template_name, result, name, args = m.groups()
385      if template_name:
386        matcherTemplateArgs = re.findall(r'Matcher<\s*(%s)\s*>' % template_name, args)
387        templateArgs = re.findall(r'(?:^|[\s,<])(%s)(?:$|[\s,>])' % template_name, args)
388        if len(matcherTemplateArgs) < len(templateArgs):
389          # The template name is used naked, so don't replace with `*`` later on
390          template_name = None
391        else :
392          args = re.sub(r'(^|[\s,<])%s($|[\s,>])' % template_name, r'\1*\2', args)
393      args = ', '.join(p.strip() for p in args.split(','))
394      m = re.match(r'(?:^|.*\s+)internal::(?:Bindable)?Matcher<([^>]+)>$', result)
395      if m:
396        result_types = [m.group(1)]
397        if template_name and len(result_types) is 1 and result_types[0] == template_name:
398          result_types = ['*']
399      else:
400        result_types = extract_result_types(comment)
401      if not result_types:
402        if not comment:
403          # Only overloads don't have their own doxygen comments; ignore those.
404          print('Ignoring "%s"' % name)
405        else:
406          print('Cannot determine result type for "%s"' % name)
407      else:
408        for result_type in result_types:
409          add_matcher(result_type, name, args, comment)
410    else:
411      print('*** Unparsable: "' + declaration + '" ***')
412
413def sort_table(matcher_type, matcher_map):
414  """Returns the sorted html table for the given row map."""
415  table = ''
416  for key in sorted(matcher_map.keys()):
417    table += matcher_map[key] + '\n'
418  return ('<!-- START_%(type)s_MATCHERS -->\n' +
419          '%(table)s' +
420          '<!--END_%(type)s_MATCHERS -->') % {
421    'type': matcher_type,
422    'table': table,
423  }
424
425# Parse the ast matchers.
426# We alternate between two modes:
427# body = True: We parse the definition of a matcher. We need
428#   to parse the full definition before adding a matcher, as the
429#   definition might contain static asserts that specify the result
430#   type.
431# body = False: We parse the comments and declaration of the matcher.
432comment = ''
433declaration = ''
434allowed_types = []
435body = False
436for line in open(MATCHERS_FILE).read().splitlines():
437  if body:
438    if line.strip() and line[0] == '}':
439      if declaration:
440        act_on_decl(declaration, comment, allowed_types)
441        comment = ''
442        declaration = ''
443        allowed_types = []
444      body = False
445    else:
446      m = re.search(r'is_base_of<([^,]+), NodeType>', line)
447      if m and m.group(1):
448        allowed_types += [m.group(1)]
449    continue
450  if line.strip() and line.lstrip()[0] == '/':
451    comment += re.sub(r'^/+\s?', '', line) + '\n'
452  else:
453    declaration += ' ' + line
454    if ((not line.strip()) or
455        line.rstrip()[-1] == ';' or
456        (line.rstrip()[-1] == '{' and line.rstrip()[-3:] != '= {')):
457      if line.strip() and line.rstrip()[-1] == '{':
458        body = True
459      else:
460        act_on_decl(declaration, comment, allowed_types)
461        comment = ''
462        declaration = ''
463        allowed_types = []
464
465node_matcher_table = sort_table('DECL', node_matchers)
466narrowing_matcher_table = sort_table('NARROWING', narrowing_matchers)
467traversal_matcher_table = sort_table('TRAVERSAL', traversal_matchers)
468
469reference = open('../LibASTMatchersReference.html').read()
470reference = re.sub(r'<!-- START_DECL_MATCHERS.*END_DECL_MATCHERS -->',
471                   node_matcher_table, reference, flags=re.S)
472reference = re.sub(r'<!-- START_NARROWING_MATCHERS.*END_NARROWING_MATCHERS -->',
473                   narrowing_matcher_table, reference, flags=re.S)
474reference = re.sub(r'<!-- START_TRAVERSAL_MATCHERS.*END_TRAVERSAL_MATCHERS -->',
475                   traversal_matcher_table, reference, flags=re.S)
476
477with open('../LibASTMatchersReference.html', 'wb') as output:
478  output.write(reference)
479
480