• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2007 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_PROG_H_
6 #define RE2_PROG_H_
7 
8 // Compiled representation of regular expressions.
9 // See regexp.h for the Regexp class, which represents a regular
10 // expression symbolically.
11 
12 #include <stdint.h>
13 #include <functional>
14 #include <mutex>
15 #include <string>
16 #include <vector>
17 #include <type_traits>
18 
19 #include "util/util.h"
20 #include "util/logging.h"
21 #include "re2/pod_array.h"
22 #include "re2/re2.h"
23 #include "re2/sparse_array.h"
24 #include "re2/sparse_set.h"
25 
26 namespace re2 {
27 
28 // Opcodes for Inst
29 enum InstOp {
30   kInstAlt = 0,      // choose between out_ and out1_
31   kInstAltMatch,     // Alt: out_ is [00-FF] and back, out1_ is match; or vice versa.
32   kInstByteRange,    // next (possible case-folded) byte must be in [lo_, hi_]
33   kInstCapture,      // capturing parenthesis number cap_
34   kInstEmptyWidth,   // empty-width special (^ $ ...); bit(s) set in empty_
35   kInstMatch,        // found a match!
36   kInstNop,          // no-op; occasionally unavoidable
37   kInstFail,         // never match; occasionally unavoidable
38   kNumInst,
39 };
40 
41 // Bit flags for empty-width specials
42 enum EmptyOp {
43   kEmptyBeginLine        = 1<<0,      // ^ - beginning of line
44   kEmptyEndLine          = 1<<1,      // $ - end of line
45   kEmptyBeginText        = 1<<2,      // \A - beginning of text
46   kEmptyEndText          = 1<<3,      // \z - end of text
47   kEmptyWordBoundary     = 1<<4,      // \b - word boundary
48   kEmptyNonWordBoundary  = 1<<5,      // \B - not \b
49   kEmptyAllFlags         = (1<<6)-1,
50 };
51 
52 class DFA;
53 class Regexp;
54 
55 // Compiled form of regexp program.
56 class Prog {
57  public:
58   Prog();
59   ~Prog();
60 
61   // Single instruction in regexp program.
62   class Inst {
63    public:
64     // See the assertion below for why this is so.
65     Inst() = default;
66 
67     // Copyable.
68     Inst(const Inst&) = default;
69     Inst& operator=(const Inst&) = default;
70 
71     // Constructors per opcode
72     void InitAlt(uint32_t out, uint32_t out1);
73     void InitByteRange(int lo, int hi, int foldcase, uint32_t out);
74     void InitCapture(int cap, uint32_t out);
75     void InitEmptyWidth(EmptyOp empty, uint32_t out);
76     void InitMatch(int id);
77     void InitNop(uint32_t out);
78     void InitFail();
79 
80     // Getters
id(Prog * p)81     int id(Prog* p) { return static_cast<int>(this - p->inst_.data()); }
opcode()82     InstOp opcode() { return static_cast<InstOp>(out_opcode_&7); }
last()83     int last()      { return (out_opcode_>>3)&1; }
out()84     int out()       { return out_opcode_>>4; }
out1()85     int out1()      { DCHECK(opcode() == kInstAlt || opcode() == kInstAltMatch); return out1_; }
cap()86     int cap()       { DCHECK_EQ(opcode(), kInstCapture); return cap_; }
lo()87     int lo()        { DCHECK_EQ(opcode(), kInstByteRange); return lo_; }
hi()88     int hi()        { DCHECK_EQ(opcode(), kInstByteRange); return hi_; }
foldcase()89     int foldcase()  { DCHECK_EQ(opcode(), kInstByteRange); return hint_foldcase_&1; }
hint()90     int hint()      { DCHECK_EQ(opcode(), kInstByteRange); return hint_foldcase_>>1; }
match_id()91     int match_id()  { DCHECK_EQ(opcode(), kInstMatch); return match_id_; }
empty()92     EmptyOp empty() { DCHECK_EQ(opcode(), kInstEmptyWidth); return empty_; }
93 
greedy(Prog * p)94     bool greedy(Prog* p) {
95       DCHECK_EQ(opcode(), kInstAltMatch);
96       return p->inst(out())->opcode() == kInstByteRange ||
97              (p->inst(out())->opcode() == kInstNop &&
98               p->inst(p->inst(out())->out())->opcode() == kInstByteRange);
99     }
100 
101     // Does this inst (an kInstByteRange) match c?
Matches(int c)102     inline bool Matches(int c) {
103       DCHECK_EQ(opcode(), kInstByteRange);
104       if (foldcase() && 'A' <= c && c <= 'Z')
105         c += 'a' - 'A';
106       return lo_ <= c && c <= hi_;
107     }
108 
109     // Returns string representation for debugging.
110     std::string Dump();
111 
112     // Maximum instruction id.
113     // (Must fit in out_opcode_. PatchList/last steal another bit.)
114     static const int kMaxInst = (1<<28) - 1;
115 
116    private:
set_opcode(InstOp opcode)117     void set_opcode(InstOp opcode) {
118       out_opcode_ = (out()<<4) | (last()<<3) | opcode;
119     }
120 
set_last()121     void set_last() {
122       out_opcode_ = (out()<<4) | (1<<3) | opcode();
123     }
124 
set_out(int out)125     void set_out(int out) {
126       out_opcode_ = (out<<4) | (last()<<3) | opcode();
127     }
128 
set_out_opcode(int out,InstOp opcode)129     void set_out_opcode(int out, InstOp opcode) {
130       out_opcode_ = (out<<4) | (last()<<3) | opcode;
131     }
132 
133     uint32_t out_opcode_;  // 28 bits: out, 1 bit: last, 3 (low) bits: opcode
134     union {                // additional instruction arguments:
135       uint32_t out1_;      // opcode == kInstAlt
136                            //   alternate next instruction
137 
138       int32_t cap_;        // opcode == kInstCapture
139                            //   Index of capture register (holds text
140                            //   position recorded by capturing parentheses).
141                            //   For \n (the submatch for the nth parentheses),
142                            //   the left parenthesis captures into register 2*n
143                            //   and the right one captures into register 2*n+1.
144 
145       int32_t match_id_;   // opcode == kInstMatch
146                            //   Match ID to identify this match (for re2::Set).
147 
148       struct {             // opcode == kInstByteRange
149         uint8_t lo_;       //   byte range is lo_-hi_ inclusive
150         uint8_t hi_;       //
151         uint16_t hint_foldcase_;  // 15 bits: hint, 1 (low) bit: foldcase
152                            //   hint to execution engines: the delta to the
153                            //   next instruction (in the current list) worth
154                            //   exploring iff this instruction matched; 0
155                            //   means there are no remaining possibilities,
156                            //   which is most likely for character classes.
157                            //   foldcase: A-Z -> a-z before checking range.
158       };
159 
160       EmptyOp empty_;       // opcode == kInstEmptyWidth
161                             //   empty_ is bitwise OR of kEmpty* flags above.
162     };
163 
164     friend class Compiler;
165     friend struct PatchList;
166     friend class Prog;
167   };
168 
169   // Inst must be trivial so that we can freely clear it with memset(3).
170   // Arrays of Inst are initialised by copying the initial elements with
171   // memmove(3) and then clearing any remaining elements with memset(3).
172   static_assert(std::is_trivial<Inst>::value, "Inst must be trivial");
173 
174   // Whether to anchor the search.
175   enum Anchor {
176     kUnanchored,  // match anywhere
177     kAnchored,    // match only starting at beginning of text
178   };
179 
180   // Kind of match to look for (for anchor != kFullMatch)
181   //
182   // kLongestMatch mode finds the overall longest
183   // match but still makes its submatch choices the way
184   // Perl would, not in the way prescribed by POSIX.
185   // The POSIX rules are much more expensive to implement,
186   // and no one has needed them.
187   //
188   // kFullMatch is not strictly necessary -- we could use
189   // kLongestMatch and then check the length of the match -- but
190   // the matching code can run faster if it knows to consider only
191   // full matches.
192   enum MatchKind {
193     kFirstMatch,     // like Perl, PCRE
194     kLongestMatch,   // like egrep or POSIX
195     kFullMatch,      // match only entire text; implies anchor==kAnchored
196     kManyMatch       // for SearchDFA, records set of matches
197   };
198 
inst(int id)199   Inst *inst(int id) { return &inst_[id]; }
start()200   int start() { return start_; }
start_unanchored()201   int start_unanchored() { return start_unanchored_; }
set_start(int start)202   void set_start(int start) { start_ = start; }
set_start_unanchored(int start)203   void set_start_unanchored(int start) { start_unanchored_ = start; }
size()204   int size() { return size_; }
reversed()205   bool reversed() { return reversed_; }
set_reversed(bool reversed)206   void set_reversed(bool reversed) { reversed_ = reversed; }
list_count()207   int list_count() { return list_count_; }
inst_count(InstOp op)208   int inst_count(InstOp op) { return inst_count_[op]; }
list_heads()209   uint16_t* list_heads() { return list_heads_.data(); }
set_dfa_mem(int64_t dfa_mem)210   void set_dfa_mem(int64_t dfa_mem) { dfa_mem_ = dfa_mem; }
dfa_mem()211   int64_t dfa_mem() { return dfa_mem_; }
flags()212   int flags() { return flags_; }
set_flags(int flags)213   void set_flags(int flags) { flags_ = flags; }
anchor_start()214   bool anchor_start() { return anchor_start_; }
set_anchor_start(bool b)215   void set_anchor_start(bool b) { anchor_start_ = b; }
anchor_end()216   bool anchor_end() { return anchor_end_; }
set_anchor_end(bool b)217   void set_anchor_end(bool b) { anchor_end_ = b; }
bytemap_range()218   int bytemap_range() { return bytemap_range_; }
bytemap()219   const uint8_t* bytemap() { return bytemap_; }
220 
221   // Lazily computed.
222   int first_byte();
223 
224   // Returns string representation of program for debugging.
225   std::string Dump();
226   std::string DumpUnanchored();
227   std::string DumpByteMap();
228 
229   // Returns the set of kEmpty flags that are in effect at
230   // position p within context.
231   static uint32_t EmptyFlags(const StringPiece& context, const char* p);
232 
233   // Returns whether byte c is a word character: ASCII only.
234   // Used by the implementation of \b and \B.
235   // This is not right for Unicode, but:
236   //   - it's hard to get right in a byte-at-a-time matching world
237   //     (the DFA has only one-byte lookahead).
238   //   - even if the lookahead were possible, the Progs would be huge.
239   // This crude approximation is the same one PCRE uses.
IsWordChar(uint8_t c)240   static bool IsWordChar(uint8_t c) {
241     return ('A' <= c && c <= 'Z') ||
242            ('a' <= c && c <= 'z') ||
243            ('0' <= c && c <= '9') ||
244            c == '_';
245   }
246 
247   // Execution engines.  They all search for the regexp (run the prog)
248   // in text, which is in the larger context (used for ^ $ \b etc).
249   // Anchor and kind control the kind of search.
250   // Returns true if match found, false if not.
251   // If match found, fills match[0..nmatch-1] with submatch info.
252   // match[0] is overall match, match[1] is first set of parens, etc.
253   // If a particular submatch is not matched during the regexp match,
254   // it is set to NULL.
255   //
256   // Matching text == StringPiece(NULL, 0) is treated as any other empty
257   // string, but note that on return, it will not be possible to distinguish
258   // submatches that matched that empty string from submatches that didn't
259   // match anything.  Either way, match[i] == NULL.
260 
261   // Search using NFA: can find submatches but kind of slow.
262   bool SearchNFA(const StringPiece& text, const StringPiece& context,
263                  Anchor anchor, MatchKind kind,
264                  StringPiece* match, int nmatch);
265 
266   // Search using DFA: much faster than NFA but only finds
267   // end of match and can use a lot more memory.
268   // Returns whether a match was found.
269   // If the DFA runs out of memory, sets *failed to true and returns false.
270   // If matches != NULL and kind == kManyMatch and there is a match,
271   // SearchDFA fills matches with the match IDs of the final matching state.
272   bool SearchDFA(const StringPiece& text, const StringPiece& context,
273                  Anchor anchor, MatchKind kind, StringPiece* match0,
274                  bool* failed, SparseSet* matches);
275 
276   // The callback issued after building each DFA state with BuildEntireDFA().
277   // If next is null, then the memory budget has been exhausted and building
278   // will halt. Otherwise, the state has been built and next points to an array
279   // of bytemap_range()+1 slots holding the next states as per the bytemap and
280   // kByteEndText. The number of the state is implied by the callback sequence:
281   // the first callback is for state 0, the second callback is for state 1, ...
282   // match indicates whether the state is a matching state.
283   using DFAStateCallback = std::function<void(const int* next, bool match)>;
284 
285   // Build the entire DFA for the given match kind.
286   // Usually the DFA is built out incrementally, as needed, which
287   // avoids lots of unnecessary work.
288   // If cb is not empty, it receives one callback per state built.
289   // Returns the number of states built.
290   // FOR TESTING OR EXPERIMENTAL PURPOSES ONLY.
291   int BuildEntireDFA(MatchKind kind, const DFAStateCallback& cb);
292 
293   // Controls whether the DFA should bail out early if the NFA would be faster.
294   // FOR TESTING ONLY.
295   static void TEST_dfa_should_bail_when_slow(bool b);
296 
297   // Compute bytemap.
298   void ComputeByteMap();
299 
300   // Computes whether all matches must begin with the same first
301   // byte, and if so, returns that byte.  If not, returns -1.
302   int ComputeFirstByte();
303 
304   // Run peep-hole optimizer on program.
305   void Optimize();
306 
307   // One-pass NFA: only correct if IsOnePass() is true,
308   // but much faster than NFA (competitive with PCRE)
309   // for those expressions.
310   bool IsOnePass();
311   bool SearchOnePass(const StringPiece& text, const StringPiece& context,
312                      Anchor anchor, MatchKind kind,
313                      StringPiece* match, int nmatch);
314 
315   // Bit-state backtracking.  Fast on small cases but uses memory
316   // proportional to the product of the list count and the text size.
CanBitState()317   bool CanBitState() { return list_heads_.data() != NULL; }
318   bool SearchBitState(const StringPiece& text, const StringPiece& context,
319                       Anchor anchor, MatchKind kind,
320                       StringPiece* match, int nmatch);
321 
322   static const int kMaxOnePassCapture = 5;  // $0 through $4
323 
324   // Backtracking search: the gold standard against which the other
325   // implementations are checked.  FOR TESTING ONLY.
326   // It allocates a ton of memory to avoid running forever.
327   // It is also recursive, so can't use in production (will overflow stacks).
328   // The name "Unsafe" here is supposed to be a flag that
329   // you should not be using this function.
330   bool UnsafeSearchBacktrack(const StringPiece& text,
331                              const StringPiece& context,
332                              Anchor anchor, MatchKind kind,
333                              StringPiece* match, int nmatch);
334 
335   // Computes range for any strings matching regexp. The min and max can in
336   // some cases be arbitrarily precise, so the caller gets to specify the
337   // maximum desired length of string returned.
338   //
339   // Assuming PossibleMatchRange(&min, &max, N) returns successfully, any
340   // string s that is an anchored match for this regexp satisfies
341   //   min <= s && s <= max.
342   //
343   // Note that PossibleMatchRange() will only consider the first copy of an
344   // infinitely repeated element (i.e., any regexp element followed by a '*' or
345   // '+' operator). Regexps with "{N}" constructions are not affected, as those
346   // do not compile down to infinite repetitions.
347   //
348   // Returns true on success, false on error.
349   bool PossibleMatchRange(std::string* min, std::string* max, int maxlen);
350 
351   // EXPERIMENTAL! SUBJECT TO CHANGE!
352   // Outputs the program fanout into the given sparse array.
353   void Fanout(SparseArray<int>* fanout);
354 
355   // Compiles a collection of regexps to Prog.  Each regexp will have
356   // its own Match instruction recording the index in the output vector.
357   static Prog* CompileSet(Regexp* re, RE2::Anchor anchor, int64_t max_mem);
358 
359   // Flattens the Prog from "tree" form to "list" form. This is an in-place
360   // operation in the sense that the old instructions are lost.
361   void Flatten();
362 
363   // Walks the Prog; the "successor roots" or predecessors of the reachable
364   // instructions are marked in rootmap or predmap/predvec, respectively.
365   // reachable and stk are preallocated scratch structures.
366   void MarkSuccessors(SparseArray<int>* rootmap,
367                       SparseArray<int>* predmap,
368                       std::vector<std::vector<int>>* predvec,
369                       SparseSet* reachable, std::vector<int>* stk);
370 
371   // Walks the Prog from the given "root" instruction; the "dominator root"
372   // of the reachable instructions (if such exists) is marked in rootmap.
373   // reachable and stk are preallocated scratch structures.
374   void MarkDominator(int root, SparseArray<int>* rootmap,
375                      SparseArray<int>* predmap,
376                      std::vector<std::vector<int>>* predvec,
377                      SparseSet* reachable, std::vector<int>* stk);
378 
379   // Walks the Prog from the given "root" instruction; the reachable
380   // instructions are emitted in "list" form and appended to flat.
381   // reachable and stk are preallocated scratch structures.
382   void EmitList(int root, SparseArray<int>* rootmap,
383                 std::vector<Inst>* flat,
384                 SparseSet* reachable, std::vector<int>* stk);
385 
386   // Computes hints for ByteRange instructions in [begin, end).
387   void ComputeHints(std::vector<Inst>* flat, int begin, int end);
388 
389  private:
390   friend class Compiler;
391 
392   DFA* GetDFA(MatchKind kind);
393   void DeleteDFA(DFA* dfa);
394 
395   bool anchor_start_;       // regexp has explicit start anchor
396   bool anchor_end_;         // regexp has explicit end anchor
397   bool reversed_;           // whether program runs backward over input
398   bool did_flatten_;        // has Flatten been called?
399   bool did_onepass_;        // has IsOnePass been called?
400 
401   int start_;               // entry point for program
402   int start_unanchored_;    // unanchored entry point for program
403   int size_;                // number of instructions
404   int bytemap_range_;       // bytemap_[x] < bytemap_range_
405   int first_byte_;          // required first byte for match, or -1 if none
406   int flags_;               // regexp parse flags
407 
408   int list_count_;                 // count of lists (see above)
409   int inst_count_[kNumInst];       // count of instructions by opcode
410   PODArray<uint16_t> list_heads_;  // sparse array enumerating list heads
411                                    // not populated if size_ is overly large
412 
413   PODArray<Inst> inst_;              // pointer to instruction array
414   PODArray<uint8_t> onepass_nodes_;  // data for OnePass nodes
415 
416   int64_t dfa_mem_;         // Maximum memory for DFAs.
417   DFA* dfa_first_;          // DFA cached for kFirstMatch/kManyMatch
418   DFA* dfa_longest_;        // DFA cached for kLongestMatch/kFullMatch
419 
420   uint8_t bytemap_[256];    // map from input bytes to byte classes
421 
422   std::once_flag first_byte_once_;
423   std::once_flag dfa_first_once_;
424   std::once_flag dfa_longest_once_;
425 
426   Prog(const Prog&) = delete;
427   Prog& operator=(const Prog&) = delete;
428 };
429 
430 }  // namespace re2
431 
432 #endif  // RE2_PROG_H_
433