/src/bloaty/third_party/re2/re2/walker-inl.h

Source (jump to first uncovered line)
// Copyright 2006 The RE2 Authors.  All Rights Reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

#ifndef RE2_WALKER_INL_H_
#define RE2_WALKER_INL_H_

// Helper class for traversing Regexps without recursion.
// Clients should declare their own subclasses that override
// the PreVisit and PostVisit methods, which are called before
// and after visiting the subexpressions.

// Not quite the Visitor pattern, because (among other things)
// the Visitor pattern is recursive.

#include <stack>

#include "util/logging.h"
#include "re2/regexp.h"

namespace re2 {

template<typename T> struct WalkState;

template<typename T> class Regexp::Walker {
 public:
  Walker();
  virtual ~Walker();

  // Virtual method called before visiting re's children.
  // PreVisit passes ownership of its return value to its caller.
  // The Arg* that PreVisit returns will be passed to PostVisit as pre_arg
  // and passed to the child PreVisits and PostVisits as parent_arg.
  // At the top-most Regexp, parent_arg is arg passed to walk.
  // If PreVisit sets *stop to true, the walk does not recurse
  // into the children.  Instead it behaves as though the return
  // value from PreVisit is the return value from PostVisit.
  // The default PreVisit returns parent_arg.
  virtual T PreVisit(Regexp* re, T parent_arg, bool* stop);

  // Virtual method called after visiting re's children.
  // The pre_arg is the T that PreVisit returned.
  // The child_args is a vector of the T that the child PostVisits returned.
  // PostVisit takes ownership of pre_arg.
  // PostVisit takes ownership of the Ts
  // in *child_args, but not the vector itself.
  // PostVisit passes ownership of its return value
  // to its caller.
  // The default PostVisit simply returns pre_arg.
  virtual T PostVisit(Regexp* re, T parent_arg, T pre_arg,
                      T* child_args, int nchild_args);

  // Virtual method called to copy a T,
  // when Walk notices that more than one child is the same re.
  virtual T Copy(T arg);

  // Virtual method called to do a "quick visit" of the re,
  // but not its children.  Only called once the visit budget
  // has been used up and we're trying to abort the walk
  // as quickly as possible.  Should return a value that
  // makes sense for the parent PostVisits still to be run.
  // This function is (hopefully) only called by
  // WalkExponential, but must be implemented by all clients,
  // just in case.
  virtual T ShortVisit(Regexp* re, T parent_arg) = 0;

  // Walks over a regular expression.
  // Top_arg is passed as parent_arg to PreVisit and PostVisit of re.
  // Returns the T returned by PostVisit on re.
  T Walk(Regexp* re, T top_arg);

  // Like Walk, but doesn't use Copy.  This can lead to
  // exponential runtimes on cross-linked Regexps like the
  // ones generated by Simplify.  To help limit this,
  // at most max_visits nodes will be visited and then
  // the walk will be cut off early.
  // If the walk *is* cut off early, ShortVisit(re)
  // will be called on regexps that cannot be fully
  // visited rather than calling PreVisit/PostVisit.
  T WalkExponential(Regexp* re, T top_arg, int max_visits);

  // Clears the stack.  Should never be necessary, since
  // Walk always enters and exits with an empty stack.
  // Logs DFATAL if stack is not already clear.
  void Reset();

  // Returns whether walk was cut off.
  bool stopped_early() { return stopped_early_; }

 private:
  // Walk state for the entire traversal.
  std::stack<WalkState<T> >* stack_;
  bool stopped_early_;
  int max_visits_;

  T WalkInternal(Regexp* re, T top_arg, bool use_copy);

  Walker(const Walker&) = delete;
  Walker& operator=(const Walker&) = delete;
};

template<typename T> T Regexp::Walker<T>::PreVisit(Regexp* re,
                                                   T parent_arg,
                                                   bool* stop) {
  return parent_arg;
}

template<typename T> T Regexp::Walker<T>::PostVisit(Regexp* re,
                                                    T parent_arg,
                                                    T pre_arg,
                                                    T* child_args,
                                                    int nchild_args) {
  return pre_arg;
}

template<typename T> T Regexp::Walker<T>::Copy(T arg) {
  return arg;
}

// State about a single level in the traversal.
template<typename T> struct WalkState {
  WalkState<T>(Regexp* re, T parent)
    : re(re),
      n(-1),
      parent_arg(parent),
      child_args(NULL) { }

  Regexp* re;  // The regexp
  int n;  // The index of the next child to process; -1 means need to PreVisit
  T parent_arg;  // Accumulated arguments.
  T pre_arg;
  T child_arg;  // One-element buffer for child_args.
  T* child_args;
};

template<typename T> Regexp::Walker<T>::Walker() {
  stack_ = new std::stack<WalkState<T> >;
  stopped_early_ = false;
}

template<typename T> Regexp::Walker<T>::~Walker() {
  Reset();
  delete stack_;
}

// Clears the stack.  Should never be necessary, since
// Walk always enters and exits with an empty stack.
// Logs DFATAL if stack is not already clear.
template<typename T> void Regexp::Walker<T>::Reset() {
  if (stack_ && stack_->size() > 0) {
    LOG(DFATAL) << "Stack not empty.";
    while (stack_->size() > 0) {
      delete stack_->top().child_args;
      stack_->pop();
    }
  }
}

template<typename T> T Regexp::Walker<T>::WalkInternal(Regexp* re, T top_arg,
                                                       bool use_copy) {
  Reset();

  if (re == NULL) {
    LOG(DFATAL) << "Walk NULL";
    return top_arg;
  }

  stack_->push(WalkState<T>(re, top_arg));

  WalkState<T>* s;
  for (;;) {
    T t;
    s = &stack_->top();
    Regexp* re = s->re;
    switch (s->n) {
      case -1: {
        if (--max_visits_ < 0) {
          stopped_early_ = true;
          t = ShortVisit(re, s->parent_arg);
          break;
        }
        bool stop = false;
        s->pre_arg = PreVisit(re, s->parent_arg, &stop);
        if (stop) {
          t = s->pre_arg;
          break;
        }
        s->n = 0;
        s->child_args = NULL;
        if (re->nsub_ == 1)
          s->child_args = &s->child_arg;
        else if (re->nsub_ > 1)
          s->child_args = new T[re->nsub_];
        FALLTHROUGH_INTENDED;
      }
      default: {
        if (re->nsub_ > 0) {
          Regexp** sub = re->sub();
          if (s->n < re->nsub_) {
            if (use_copy && s->n > 0 && sub[s->n - 1] == sub[s->n]) {
              s->child_args[s->n] = Copy(s->child_args[s->n - 1]);
              s->n++;
            } else {
              stack_->push(WalkState<T>(sub[s->n], s->pre_arg));
            }
            continue;
          }
        }

        t = PostVisit(re, s->parent_arg, s->pre_arg, s->child_args, s->n);
        if (re->nsub_ > 1)
          delete[] s->child_args;
        break;
      }
    }

    // We've finished stack_->top().
    // Update next guy down.
    stack_->pop();
    if (stack_->size() == 0)
      return t;
    s = &stack_->top();
    if (s->child_args != NULL)
      s->child_args[s->n] = t;
    else
      s->child_arg = t;
    s->n++;
  }
}

template<typename T> T Regexp::Walker<T>::Walk(Regexp* re, T top_arg) {
  // Without the exponential walking behavior,
  // this budget should be more than enough for any
  // regexp, and yet not enough to get us in trouble
  // as far as CPU time.
  max_visits_ = 1000000;
  return WalkInternal(re, top_arg, true);
}

template<typename T> T Regexp::Walker<T>::WalkExponential(Regexp* re, T top_arg,
                                                          int max_visits) {
  max_visits_ = max_visits;
  return WalkInternal(re, top_arg, false);
}

}  // namespace re2

#endif  // RE2_WALKER_INL_H_

Coverage Report

Created: 2022-08-24 06:07

Line	Count	Source (jump to first uncovered line)
1		// Copyright 2006 The RE2 Authors. All Rights Reserved.
2		// Use of this source code is governed by a BSD-style
3		// license that can be found in the LICENSE file.
4
5		#ifndef RE2_WALKER_INL_H_
6		#define RE2_WALKER_INL_H_
7
8		// Helper class for traversing Regexps without recursion.
9		// Clients should declare their own subclasses that override
10		// the PreVisit and PostVisit methods, which are called before
11		// and after visiting the subexpressions.
12
13		// Not quite the Visitor pattern, because (among other things)
14		// the Visitor pattern is recursive.
15
16		#include <stack>
17
18		#include "util/logging.h"
19		#include "re2/regexp.h"
20
21		namespace re2 {
22
23		template<typename T> struct WalkState;
24
25		template<typename T> class Regexp::Walker {
26		public:
27		Walker();
28		virtual ~Walker();
29
30		// Virtual method called before visiting re's children.
31		// PreVisit passes ownership of its return value to its caller.
32		// The Arg* that PreVisit returns will be passed to PostVisit as pre_arg
33		// and passed to the child PreVisits and PostVisits as parent_arg.
34		// At the top-most Regexp, parent_arg is arg passed to walk.
35		// If PreVisit sets *stop to true, the walk does not recurse
36		// into the children. Instead it behaves as though the return
37		// value from PreVisit is the return value from PostVisit.
38		// The default PreVisit returns parent_arg.
39		virtual T PreVisit(Regexp* re, T parent_arg, bool* stop);
40
41		// Virtual method called after visiting re's children.
42		// The pre_arg is the T that PreVisit returned.
43		// The child_args is a vector of the T that the child PostVisits returned.
44		// PostVisit takes ownership of pre_arg.
45		// PostVisit takes ownership of the Ts
46		// in *child_args, but not the vector itself.
47		// PostVisit passes ownership of its return value
48		// to its caller.
49		// The default PostVisit simply returns pre_arg.
50		virtual T PostVisit(Regexp* re, T parent_arg, T pre_arg,
51		T* child_args, int nchild_args);
52
53		// Virtual method called to copy a T,
54		// when Walk notices that more than one child is the same re.
55		virtual T Copy(T arg);
56
57		// Virtual method called to do a "quick visit" of the re,
58		// but not its children. Only called once the visit budget
59		// has been used up and we're trying to abort the walk
60		// as quickly as possible. Should return a value that
61		// makes sense for the parent PostVisits still to be run.
62		// This function is (hopefully) only called by
63		// WalkExponential, but must be implemented by all clients,
64		// just in case.
65		virtual T ShortVisit(Regexp* re, T parent_arg) = 0;
66
67		// Walks over a regular expression.
68		// Top_arg is passed as parent_arg to PreVisit and PostVisit of re.
69		// Returns the T returned by PostVisit on re.
70		T Walk(Regexp* re, T top_arg);
71
72		// Like Walk, but doesn't use Copy. This can lead to
73		// exponential runtimes on cross-linked Regexps like the
74		// ones generated by Simplify. To help limit this,
75		// at most max_visits nodes will be visited and then
76		// the walk will be cut off early.
77		// If the walk is cut off early, ShortVisit(re)
78		// will be called on regexps that cannot be fully
79		// visited rather than calling PreVisit/PostVisit.
80		T WalkExponential(Regexp* re, T top_arg, int max_visits);
81
82		// Clears the stack. Should never be necessary, since
83		// Walk always enters and exits with an empty stack.
84		// Logs DFATAL if stack is not already clear.
85		void Reset();
86
87		// Returns whether walk was cut off.
88	0	bool stopped_early() { return stopped_early_; }
89
90		private:
91		// Walk state for the entire traversal.
92		std::stack<WalkState<T> >* stack_;
93		bool stopped_early_;
94		int max_visits_;
95
96		T WalkInternal(Regexp* re, T top_arg, bool use_copy);
97
98		Walker(const Walker&) = delete;
99		Walker& operator=(const Walker&) = delete;
100		};
101
102		template<typename T> T Regexp::Walker<T>::PreVisit(Regexp* re,
103		T parent_arg,
104	0	bool* stop) {
105	0	return parent_arg;
106	0	} Unexecuted instantiation: re2::Regexp::Walker<int>::PreVisit(re2::Regexp, int, bool) Unexecuted instantiation: re2::Regexp::Walker<re2::Frag>::PreVisit(re2::Regexp, re2::Frag, bool) Unexecuted instantiation: re2::Regexp::Walker<re2::Regexp>::PreVisit(re2::Regexp, re2::Regexp, bool)
107
108		template<typename T> T Regexp::Walker<T>::PostVisit(Regexp* re,
109		T parent_arg,
110		T pre_arg,
111		T* child_args,
112	0	int nchild_args) {
113	0	return pre_arg;
114	0	} Unexecuted instantiation: re2::Regexp::Walker<int>::PostVisit(re2::Regexp, int, int, int, int) Unexecuted instantiation: re2::Regexp::Walker<re2::Frag>::PostVisit(re2::Regexp, re2::Frag, re2::Frag, re2::Frag, int) Unexecuted instantiation: re2::Regexp::Walker<re2::Regexp>::PostVisit(re2::Regexp, re2::Regexp, re2::Regexp, re2::Regexp**, int)
115
116	0	template<typename T> T Regexp::Walker<T>::Copy(T arg) {
117	0	return arg;
118	0	} Unexecuted instantiation: re2::Regexp::Walker<int>::Copy(int) Unexecuted instantiation: re2::Regexp::Walker<re2::Frag>::Copy(re2::Frag) Unexecuted instantiation: re2::Regexp::Walker<re2::Regexp>::Copy(re2::Regexp)
119
120		// State about a single level in the traversal.
121		template<typename T> struct WalkState {
122		WalkState<T>(Regexp* re, T parent)
123		: re(re),
124		n(-1),
125		parent_arg(parent),
126	0	child_args(NULL) { } Unexecuted instantiation: re2::WalkState<int>::WalkState(re2::Regexp, int) Unexecuted instantiation: re2::WalkState<re2::Frag>::WalkState(re2::Regexp, re2::Frag) Unexecuted instantiation: re2::WalkState<re2::Regexp>::WalkState(re2::Regexp, re2::Regexp*)
127
128		Regexp* re; // The regexp
129		int n; // The index of the next child to process; -1 means need to PreVisit
130		T parent_arg; // Accumulated arguments.
131		T pre_arg;
132		T child_arg; // One-element buffer for child_args.
133		T* child_args;
134		};
135
136	0	template<typename T> Regexp::Walker<T>::Walker() {
137	0	stack_ = new std::stack<WalkState<T> >;
138	0	stopped_early_ = false;
139	0	} Unexecuted instantiation: re2::Regexp::Walker<int>::Walker() Unexecuted instantiation: re2::Regexp::Walker<re2::Frag>::Walker() Unexecuted instantiation: re2::Regexp::Walker<re2::Regexp*>::Walker()
140
141	0	template<typename T> Regexp::Walker<T>::~Walker() {
142	0	Reset();
143	0	delete stack_;
144	0	} Unexecuted instantiation: re2::Regexp::Walker<int>::~Walker() Unexecuted instantiation: re2::Regexp::Walker<re2::Frag>::~Walker() Unexecuted instantiation: re2::Regexp::Walker<re2::Regexp*>::~Walker()
145
146		// Clears the stack. Should never be necessary, since
147		// Walk always enters and exits with an empty stack.
148		// Logs DFATAL if stack is not already clear.
149	0	template<typename T> void Regexp::Walker<T>::Reset() {
150	0	if (stack_ && stack_->size() > 0) {
151	0	LOG(DFATAL) << "Stack not empty.";
152	0	while (stack_->size() > 0) {
153	0	delete stack_->top().child_args;
154	0	stack_->pop();
155	0	}
156	0	}
157	0	} Unexecuted instantiation: re2::Regexp::Walker<int>::Reset() Unexecuted instantiation: re2::Regexp::Walker<re2::Frag>::Reset() Unexecuted instantiation: re2::Regexp::Walker<re2::Regexp*>::Reset()
158
159		template<typename T> T Regexp::Walker<T>::WalkInternal(Regexp* re, T top_arg,
160	0	bool use_copy) {
161	0	Reset();
162
163	0	if (re == NULL) {
164	0	LOG(DFATAL) << "Walk NULL";
165	0	return top_arg;
166	0	}
167
168	0	stack_->push(WalkState<T>(re, top_arg));
169
170	0	WalkState<T>* s;
171	0	for (;;) {
172	0	T t;
173	0	s = &stack_->top();
174	0	Regexp* re = s->re;
175	0	switch (s->n) {
176	0	case -1: {
177	0	if (--max_visits_ < 0) {
178	0	stopped_early_ = true;
179	0	t = ShortVisit(re, s->parent_arg);
180	0	break;
181	0	}
182	0	bool stop = false;
183	0	s->pre_arg = PreVisit(re, s->parent_arg, &stop);
184	0	if (stop) {
185	0	t = s->pre_arg;
186	0	break;
187	0	}
188	0	s->n = 0;
189	0	s->child_args = NULL;
190	0	if (re->nsub_ == 1)
191	0	s->child_args = &s->child_arg;
192	0	else if (re->nsub_ > 1)
193	0	s->child_args = new T[re->nsub_];
194	0	FALLTHROUGH_INTENDED;
195	0	}
196	0	default: {
197	0	if (re->nsub_ > 0) {
198	0	Regexp** sub = re->sub();
199	0	if (s->n < re->nsub_) {
200	0	if (use_copy && s->n > 0 && sub[s->n - 1] == sub[s->n]) {
201	0	s->child_args[s->n] = Copy(s->child_args[s->n - 1]);
202	0	s->n++;
203	0	} else {
204	0	stack_->push(WalkState<T>(sub[s->n], s->pre_arg));
205	0	}
206	0	continue;
207	0	}
208	0	}
209
210	0	t = PostVisit(re, s->parent_arg, s->pre_arg, s->child_args, s->n);
211	0	if (re->nsub_ > 1)
212	0	delete[] s->child_args;
213	0	break;
214	0	}
215	0	}
216
217		// We've finished stack_->top().
218		// Update next guy down.
219	0	stack_->pop();
220	0	if (stack_->size() == 0)
221	0	return t;
222	0	s = &stack_->top();
223	0	if (s->child_args != NULL)
224	0	s->child_args[s->n] = t;
225	0	else
226	0	s->child_arg = t;
227	0	s->n++;
228	0	}
229	0	} Unexecuted instantiation: re2::Regexp::Walker<int>::WalkInternal(re2::Regexp, int, bool) Unexecuted instantiation: re2::Regexp::Walker<re2::Frag>::WalkInternal(re2::Regexp, re2::Frag, bool) Unexecuted instantiation: re2::Regexp::Walker<re2::Regexp>::WalkInternal(re2::Regexp, re2::Regexp*, bool)
230
231	0	template<typename T> T Regexp::Walker<T>::Walk(Regexp* re, T top_arg) {
232		// Without the exponential walking behavior,
233		// this budget should be more than enough for any
234		// regexp, and yet not enough to get us in trouble
235		// as far as CPU time.
236	0	max_visits_ = 1000000;
237	0	return WalkInternal(re, top_arg, true);
238	0	} Unexecuted instantiation: re2::Regexp::Walker<int>::Walk(re2::Regexp, int) Unexecuted instantiation: re2::Regexp::Walker<re2::Regexp>::Walk(re2::Regexp, re2::Regexp)
239
240		template<typename T> T Regexp::Walker<T>::WalkExponential(Regexp* re, T top_arg,
241	0	int max_visits) {
242	0	max_visits_ = max_visits;
243	0	return WalkInternal(re, top_arg, false);
244	0	} Unexecuted instantiation: re2::Regexp::Walker<re2::Frag>::WalkExponential(re2::Regexp, re2::Frag, int) Unexecuted instantiation: re2::Regexp::Walker<int>::WalkExponential(re2::Regexp, int, int)
245
246		} // namespace re2
247
248		#endif // RE2_WALKER_INL_H_