• 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 // Compile regular expression to Prog.
6 //
7 // Prog and Inst are defined in prog.h.
8 // This file's external interface is just Regexp::CompileToProg.
9 // The Compiler class defined in this file is private.
10 
11 #include "re2/prog.h"
12 #include "re2/re2.h"
13 #include "re2/regexp.h"
14 #include "re2/walker-inl.h"
15 
16 namespace re2 {
17 
18 // List of pointers to Inst* that need to be filled in (patched).
19 // Because the Inst* haven't been filled in yet,
20 // we can use the Inst* word to hold the list's "next" pointer.
21 // It's kind of sleazy, but it works well in practice.
22 // See http://swtch.com/~rsc/regexp/regexp1.html for inspiration.
23 //
24 // Because the out and out1 fields in Inst are no longer pointers,
25 // we can't use pointers directly here either.  Instead, p refers
26 // to inst_[p>>1].out (p&1 == 0) or inst_[p>>1].out1 (p&1 == 1).
27 // p == 0 represents the NULL list.  This is okay because instruction #0
28 // is always the fail instruction, which never appears on a list.
29 
30 struct PatchList {
31   uint32 p;
32 
33   // Returns patch list containing just p.
34   static PatchList Mk(uint32 p);
35 
36   // Patches all the entries on l to have value v.
37   // Caller must not ever use patch list again.
38   static void Patch(Prog::Inst *inst0, PatchList l, uint32 v);
39 
40   // Deref returns the next pointer pointed at by p.
41   static PatchList Deref(Prog::Inst *inst0, PatchList l);
42 
43   // Appends two patch lists and returns result.
44   static PatchList Append(Prog::Inst *inst0, PatchList l1, PatchList l2);
45 };
46 
47 static PatchList nullPatchList;
48 
49 // Returns patch list containing just p.
Mk(uint32 p)50 PatchList PatchList::Mk(uint32 p) {
51   PatchList l;
52   l.p = p;
53   return l;
54 }
55 
56 // Returns the next pointer pointed at by l.
Deref(Prog::Inst * inst0,PatchList l)57 PatchList PatchList::Deref(Prog::Inst* inst0, PatchList l) {
58   Prog::Inst* ip = &inst0[l.p>>1];
59   if (l.p&1)
60     l.p = ip->out1();
61   else
62     l.p = ip->out();
63   return l;
64 }
65 
66 // Patches all the entries on l to have value v.
Patch(Prog::Inst * inst0,PatchList l,uint32 val)67 void PatchList::Patch(Prog::Inst *inst0, PatchList l, uint32 val) {
68   while (l.p != 0) {
69     Prog::Inst* ip = &inst0[l.p>>1];
70     if (l.p&1) {
71       l.p = ip->out1();
72       ip->out1_ = val;
73     } else {
74       l.p = ip->out();
75       ip->set_out(val);
76     }
77   }
78 }
79 
80 // Appends two patch lists and returns result.
Append(Prog::Inst * inst0,PatchList l1,PatchList l2)81 PatchList PatchList::Append(Prog::Inst* inst0, PatchList l1, PatchList l2) {
82   if (l1.p == 0)
83     return l2;
84   if (l2.p == 0)
85     return l1;
86 
87   PatchList l = l1;
88   for (;;) {
89     PatchList next = PatchList::Deref(inst0, l);
90     if (next.p == 0)
91       break;
92     l = next;
93   }
94 
95   Prog::Inst* ip = &inst0[l.p>>1];
96   if (l.p&1)
97     ip->out1_ = l2.p;
98   else
99     ip->set_out(l2.p);
100 
101   return l1;
102 }
103 
104 // Compiled program fragment.
105 struct Frag {
106   uint32 begin;
107   PatchList end;
108 
Fragre2::Frag109   explicit Frag(LinkerInitialized) {}
Fragre2::Frag110   Frag() : begin(0) { end.p = 0; }  // needed so Frag can go in vector
Fragre2::Frag111   Frag(uint32 begin, PatchList end) : begin(begin), end(end) {}
112 };
113 
114 static Frag kNullFrag(LINKER_INITIALIZED);
115 
116 // Input encodings.
117 enum Encoding {
118   kEncodingUTF8 = 1,  // UTF-8 (0-10FFFF)
119   kEncodingLatin1,    // Latin1 (0-FF)
120 };
121 
122 class Compiler : public Regexp::Walker<Frag> {
123  public:
124   explicit Compiler();
125   ~Compiler();
126 
127   // Compiles Regexp to a new Prog.
128   // Caller is responsible for deleting Prog when finished with it.
129   // If reversed is true, compiles for walking over the input
130   // string backward (reverses all concatenations).
131   static Prog *Compile(Regexp* re, bool reversed, int64 max_mem);
132 
133   // Compiles alternation of all the re to a new Prog.
134   // Each re has a match with an id equal to its index in the vector.
135   static Prog* CompileSet(const RE2::Options& options, RE2::Anchor anchor,
136                           Regexp* re);
137 
138   // Interface for Regexp::Walker, which helps traverse the Regexp.
139   // The walk is purely post-recursive: given the machines for the
140   // children, PostVisit combines them to create the machine for
141   // the current node.  The child_args are Frags.
142   // The Compiler traverses the Regexp parse tree, visiting
143   // each node in depth-first order.  It invokes PreVisit before
144   // visiting the node's children and PostVisit after visiting
145   // the children.
146   Frag PreVisit(Regexp* re, Frag parent_arg, bool* stop);
147   Frag PostVisit(Regexp* re, Frag parent_arg, Frag pre_arg, Frag* child_args,
148                  int nchild_args);
149   Frag ShortVisit(Regexp* re, Frag parent_arg);
150   Frag Copy(Frag arg);
151 
152   // Given fragment a, returns a+ or a+?; a* or a*?; a? or a??
153   Frag Plus(Frag a, bool nongreedy);
154   Frag Star(Frag a, bool nongreedy);
155   Frag Quest(Frag a, bool nongreedy);
156 
157   // Given fragment a, returns (a) capturing as \n.
158   Frag Capture(Frag a, int n);
159 
160   // Given fragments a and b, returns ab; a|b
161   Frag Cat(Frag a, Frag b);
162   Frag Alt(Frag a, Frag b);
163 
164   // Returns a fragment that can't match anything.
165   Frag NoMatch();
166 
167   // Returns a fragment that matches the empty string.
168   Frag Match(int32 id);
169 
170   // Returns a no-op fragment.
171   Frag Nop();
172 
173   // Returns a fragment matching the byte range lo-hi.
174   Frag ByteRange(int lo, int hi, bool foldcase);
175 
176   // Returns a fragment matching an empty-width special op.
177   Frag EmptyWidth(EmptyOp op);
178 
179   // Adds n instructions to the program.
180   // Returns the index of the first one.
181   // Returns -1 if no more instructions are available.
182   int AllocInst(int n);
183 
184   // Deletes unused instructions.
185   void Trim();
186 
187   // Rune range compiler.
188 
189   // Begins a new alternation.
190   void BeginRange();
191 
192   // Adds a fragment matching the rune range lo-hi.
193   void AddRuneRange(Rune lo, Rune hi, bool foldcase);
194   void AddRuneRangeLatin1(Rune lo, Rune hi, bool foldcase);
195   void AddRuneRangeUTF8(Rune lo, Rune hi, bool foldcase);
196   void Add_80_10ffff();
197 
198   // New suffix that matches the byte range lo-hi, then goes to next.
199   int RuneByteSuffix(uint8 lo, uint8 hi, bool foldcase, int next);
200   int UncachedRuneByteSuffix(uint8 lo, uint8 hi, bool foldcase, int next);
201 
202   // Adds a suffix to alternation.
203   void AddSuffix(int id);
204 
205   // Returns the alternation of all the added suffixes.
206   Frag EndRange();
207 
208   // Single rune.
209   Frag Literal(Rune r, bool foldcase);
210 
211   void Setup(Regexp::ParseFlags, int64, RE2::Anchor);
212   Prog* Finish();
213 
214   // Returns .* where dot = any byte
215   Frag DotStar();
216 
217  private:
218   Prog* prog_;         // Program being built.
219   bool failed_;        // Did we give up compiling?
220   Encoding encoding_;  // Input encoding
221   bool reversed_;      // Should program run backward over text?
222 
223   int max_inst_;       // Maximum number of instructions.
224 
225   Prog::Inst* inst_;   // Pointer to first instruction.
226   int inst_len_;       // Number of instructions used.
227   int inst_cap_;       // Number of instructions allocated.
228 
229   int64 max_mem_;      // Total memory budget.
230 
231   map<uint64, int> rune_cache_;
232   Frag rune_range_;
233 
234   RE2::Anchor anchor_;  // anchor mode for RE2::Set
235 
236   DISALLOW_EVIL_CONSTRUCTORS(Compiler);
237 };
238 
Compiler()239 Compiler::Compiler() {
240   prog_ = new Prog();
241   failed_ = false;
242   encoding_ = kEncodingUTF8;
243   reversed_ = false;
244   inst_ = NULL;
245   inst_len_ = 0;
246   inst_cap_ = 0;
247   max_inst_ = 1;  // make AllocInst for fail instruction okay
248   max_mem_ = 0;
249   int fail = AllocInst(1);
250   inst_[fail].InitFail();
251   max_inst_ = 0;  // Caller must change
252 }
253 
~Compiler()254 Compiler::~Compiler() {
255   delete prog_;
256   delete[] inst_;
257 }
258 
AllocInst(int n)259 int Compiler::AllocInst(int n) {
260   if (failed_ || inst_len_ + n > max_inst_) {
261     failed_ = true;
262     return -1;
263   }
264 
265   if (inst_len_ + n > inst_cap_) {
266     if (inst_cap_ == 0)
267       inst_cap_ = 8;
268     while (inst_len_ + n > inst_cap_)
269       inst_cap_ *= 2;
270     Prog::Inst* ip = new Prog::Inst[inst_cap_];
271     memmove(ip, inst_, inst_len_ * sizeof ip[0]);
272     memset(ip + inst_len_, 0, (inst_cap_ - inst_len_) * sizeof ip[0]);
273     delete[] inst_;
274     inst_ = ip;
275   }
276   int id = inst_len_;
277   inst_len_ += n;
278   return id;
279 }
280 
Trim()281 void Compiler::Trim() {
282   if (inst_len_ < inst_cap_) {
283     Prog::Inst* ip = new Prog::Inst[inst_len_];
284     memmove(ip, inst_, inst_len_ * sizeof ip[0]);
285     delete[] inst_;
286     inst_ = ip;
287     inst_cap_ = inst_len_;
288   }
289 }
290 
291 // These routines are somewhat hard to visualize in text --
292 // see http://swtch.com/~rsc/regexp/regexp1.html for
293 // pictures explaining what is going on here.
294 
295 // Returns an unmatchable fragment.
NoMatch()296 Frag Compiler::NoMatch() {
297   return Frag(0, nullPatchList);
298 }
299 
300 // Is a an unmatchable fragment?
IsNoMatch(Frag a)301 static bool IsNoMatch(Frag a) {
302   return a.begin == 0;
303 }
304 
305 // Given fragments a and b, returns fragment for ab.
Cat(Frag a,Frag b)306 Frag Compiler::Cat(Frag a, Frag b) {
307   if (IsNoMatch(a) || IsNoMatch(b))
308     return NoMatch();
309 
310   // Elide no-op.
311   Prog::Inst* begin = &inst_[a.begin];
312   if (begin->opcode() == kInstNop &&
313       a.end.p == (a.begin << 1) &&
314       begin->out() == 0) {
315     PatchList::Patch(inst_, a.end, b.begin);  // in case refs to a somewhere
316     return b;
317   }
318 
319   // To run backward over string, reverse all concatenations.
320   if (reversed_) {
321     PatchList::Patch(inst_, b.end, a.begin);
322     return Frag(b.begin, a.end);
323   }
324 
325   PatchList::Patch(inst_, a.end, b.begin);
326   return Frag(a.begin, b.end);
327 }
328 
329 // Given fragments for a and b, returns fragment for a|b.
Alt(Frag a,Frag b)330 Frag Compiler::Alt(Frag a, Frag b) {
331   // Special case for convenience in loops.
332   if (IsNoMatch(a))
333     return b;
334   if (IsNoMatch(b))
335     return a;
336 
337   int id = AllocInst(1);
338   if (id < 0)
339     return NoMatch();
340 
341   inst_[id].InitAlt(a.begin, b.begin);
342   return Frag(id, PatchList::Append(inst_, a.end, b.end));
343 }
344 
345 // When capturing submatches in like-Perl mode, a kOpAlt Inst
346 // treats out_ as the first choice, out1_ as the second.
347 //
348 // For *, +, and ?, if out_ causes another repetition,
349 // then the operator is greedy.  If out1_ is the repetition
350 // (and out_ moves forward), then the operator is non-greedy.
351 
352 // Given a fragment a, returns a fragment for a* or a*? (if nongreedy)
Star(Frag a,bool nongreedy)353 Frag Compiler::Star(Frag a, bool nongreedy) {
354   int id = AllocInst(1);
355   if (id < 0)
356     return NoMatch();
357   inst_[id].InitAlt(0, 0);
358   PatchList::Patch(inst_, a.end, id);
359   if (nongreedy) {
360     inst_[id].out1_ = a.begin;
361     return Frag(id, PatchList::Mk(id << 1));
362   } else {
363     inst_[id].set_out(a.begin);
364     return Frag(id, PatchList::Mk((id << 1) | 1));
365   }
366 }
367 
368 // Given a fragment for a, returns a fragment for a+ or a+? (if nongreedy)
Plus(Frag a,bool nongreedy)369 Frag Compiler::Plus(Frag a, bool nongreedy) {
370   // a+ is just a* with a different entry point.
371   Frag f = Star(a, nongreedy);
372   return Frag(a.begin, f.end);
373 }
374 
375 // Given a fragment for a, returns a fragment for a? or a?? (if nongreedy)
Quest(Frag a,bool nongreedy)376 Frag Compiler::Quest(Frag a, bool nongreedy) {
377   int id = AllocInst(1);
378   if (id < 0)
379     return NoMatch();
380   PatchList pl;
381   if (nongreedy) {
382     inst_[id].InitAlt(0, a.begin);
383     pl = PatchList::Mk(id << 1);
384   } else {
385     inst_[id].InitAlt(a.begin, 0);
386     pl = PatchList::Mk((id << 1) | 1);
387   }
388   return Frag(id, PatchList::Append(inst_, pl, a.end));
389 }
390 
391 // Returns a fragment for the byte range lo-hi.
ByteRange(int lo,int hi,bool foldcase)392 Frag Compiler::ByteRange(int lo, int hi, bool foldcase) {
393   int id = AllocInst(1);
394   if (id < 0)
395     return NoMatch();
396   inst_[id].InitByteRange(lo, hi, foldcase, 0);
397   prog_->byte_inst_count_++;
398   prog_->MarkByteRange(lo, hi);
399   if (foldcase && lo <= 'z' && hi >= 'a') {
400     if (lo < 'a')
401       lo = 'a';
402     if (hi > 'z')
403       hi = 'z';
404     if (lo <= hi)
405       prog_->MarkByteRange(lo + 'A' - 'a', hi + 'A' - 'a');
406   }
407   return Frag(id, PatchList::Mk(id << 1));
408 }
409 
410 // Returns a no-op fragment.  Sometimes unavoidable.
Nop()411 Frag Compiler::Nop() {
412   int id = AllocInst(1);
413   if (id < 0)
414     return NoMatch();
415   inst_[id].InitNop(0);
416   return Frag(id, PatchList::Mk(id << 1));
417 }
418 
419 // Returns a fragment that signals a match.
Match(int32 match_id)420 Frag Compiler::Match(int32 match_id) {
421   int id = AllocInst(1);
422   if (id < 0)
423     return NoMatch();
424   inst_[id].InitMatch(match_id);
425   return Frag(id, nullPatchList);
426 }
427 
428 // Returns a fragment matching a particular empty-width op (like ^ or $)
EmptyWidth(EmptyOp empty)429 Frag Compiler::EmptyWidth(EmptyOp empty) {
430   int id = AllocInst(1);
431   if (id < 0)
432     return NoMatch();
433   inst_[id].InitEmptyWidth(empty, 0);
434   if (empty & (kEmptyBeginLine|kEmptyEndLine))
435     prog_->MarkByteRange('\n', '\n');
436   if (empty & (kEmptyWordBoundary|kEmptyNonWordBoundary)) {
437     int j;
438     for (int i = 0; i < 256; i = j) {
439       for (j = i+1; j < 256 && Prog::IsWordChar(i) == Prog::IsWordChar(j); j++)
440         ;
441       prog_->MarkByteRange(i, j-1);
442     }
443   }
444   return Frag(id, PatchList::Mk(id << 1));
445 }
446 
447 // Given a fragment a, returns a fragment with capturing parens around a.
Capture(Frag a,int n)448 Frag Compiler::Capture(Frag a, int n) {
449   int id = AllocInst(2);
450   if (id < 0)
451     return NoMatch();
452   inst_[id].InitCapture(2*n, a.begin);
453   inst_[id+1].InitCapture(2*n+1, 0);
454   PatchList::Patch(inst_, a.end, id+1);
455 
456   return Frag(id, PatchList::Mk((id+1) << 1));
457 }
458 
459 // A Rune is a name for a Unicode code point.
460 // Returns maximum rune encoded by UTF-8 sequence of length len.
MaxRune(int len)461 static int MaxRune(int len) {
462   int b;  // number of Rune bits in len-byte UTF-8 sequence (len < UTFmax)
463   if (len == 1)
464     b = 7;
465   else
466     b = 8-(len+1) + 6*(len-1);
467   return (1<<b) - 1;   // maximum Rune for b bits.
468 }
469 
470 // The rune range compiler caches common suffix fragments,
471 // which are very common in UTF-8 (e.g., [80-bf]).
472 // The fragment suffixes are identified by their start
473 // instructions.  NULL denotes the eventual end match.
474 // The Frag accumulates in rune_range_.  Caching common
475 // suffixes reduces the UTF-8 "." from 32 to 24 instructions,
476 // and it reduces the corresponding one-pass NFA from 16 nodes to 8.
477 
BeginRange()478 void Compiler::BeginRange() {
479   rune_cache_.clear();
480   rune_range_.begin = 0;
481   rune_range_.end = nullPatchList;
482 }
483 
UncachedRuneByteSuffix(uint8 lo,uint8 hi,bool foldcase,int next)484 int Compiler::UncachedRuneByteSuffix(uint8 lo, uint8 hi, bool foldcase,
485                                      int next) {
486   Frag f = ByteRange(lo, hi, foldcase);
487   if (next != 0) {
488     PatchList::Patch(inst_, f.end, next);
489   } else {
490     rune_range_.end = PatchList::Append(inst_, rune_range_.end, f.end);
491   }
492   return f.begin;
493 }
494 
RuneByteSuffix(uint8 lo,uint8 hi,bool foldcase,int next)495 int Compiler::RuneByteSuffix(uint8 lo, uint8 hi, bool foldcase, int next) {
496   // In Latin1 mode, there's no point in caching.
497   // In forward UTF-8 mode, only need to cache continuation bytes.
498   if (encoding_ == kEncodingLatin1 ||
499       (encoding_ == kEncodingUTF8 &&
500        !reversed_ &&
501        !(0x80 <= lo && hi <= 0xbf))) {
502     return UncachedRuneByteSuffix(lo, hi, foldcase, next);
503   }
504 
505   uint64 key = ((uint64)next << 17) | (lo<<9) | (hi<<1) | foldcase;
506   map<uint64, int>::iterator it = rune_cache_.find(key);
507   if (it != rune_cache_.end())
508     return it->second;
509   int id = UncachedRuneByteSuffix(lo, hi, foldcase, next);
510   rune_cache_[key] = id;
511   return id;
512 }
513 
AddSuffix(int id)514 void Compiler::AddSuffix(int id) {
515   if (rune_range_.begin == 0) {
516     rune_range_.begin = id;
517     return;
518   }
519 
520   int alt = AllocInst(1);
521   if (alt < 0) {
522     rune_range_.begin = 0;
523     return;
524   }
525   inst_[alt].InitAlt(rune_range_.begin, id);
526   rune_range_.begin = alt;
527 }
528 
EndRange()529 Frag Compiler::EndRange() {
530   return rune_range_;
531 }
532 
533 // Converts rune range lo-hi into a fragment that recognizes
534 // the bytes that would make up those runes in the current
535 // encoding (Latin 1 or UTF-8).
536 // This lets the machine work byte-by-byte even when
537 // using multibyte encodings.
538 
AddRuneRange(Rune lo,Rune hi,bool foldcase)539 void Compiler::AddRuneRange(Rune lo, Rune hi, bool foldcase) {
540   switch (encoding_) {
541     default:
542     case kEncodingUTF8:
543       AddRuneRangeUTF8(lo, hi, foldcase);
544       break;
545     case kEncodingLatin1:
546       AddRuneRangeLatin1(lo, hi, foldcase);
547       break;
548   }
549 }
550 
AddRuneRangeLatin1(Rune lo,Rune hi,bool foldcase)551 void Compiler::AddRuneRangeLatin1(Rune lo, Rune hi, bool foldcase) {
552   // Latin1 is easy: runes *are* bytes.
553   if (lo > hi || lo > 0xFF)
554     return;
555   if (hi > 0xFF)
556     hi = 0xFF;
557   AddSuffix(RuneByteSuffix(lo, hi, foldcase, 0));
558 }
559 
560 // Table describing how to make a UTF-8 matching machine
561 // for the rune range 80-10FFFF (Runeself-Runemax).
562 // This range happens frequently enough (for example /./ and /[^a-z]/)
563 // and the rune_cache_ map is slow enough that this is worth
564 // special handling.  Makes compilation of a small expression
565 // with a dot in it about 10% faster.
566 // The * in the comments below mark whole sequences.
567 static struct ByteRangeProg {
568   int next;
569   int lo;
570   int hi;
571 } prog_80_10ffff[] = {
572   // Two-byte
573   { -1, 0x80, 0xBF, },  // 0:  80-BF
574   {  0, 0xC2, 0xDF, },  // 1:  C2-DF 80-BF*
575 
576   // Three-byte
577   {  0, 0xA0, 0xBF, },  // 2:  A0-BF 80-BF
578   {  2, 0xE0, 0xE0, },  // 3:  E0 A0-BF 80-BF*
579   {  0, 0x80, 0xBF, },  // 4:  80-BF 80-BF
580   {  4, 0xE1, 0xEF, },  // 5:  E1-EF 80-BF 80-BF*
581 
582   // Four-byte
583   {  4, 0x90, 0xBF, },  // 6:  90-BF 80-BF 80-BF
584   {  6, 0xF0, 0xF0, },  // 7:  F0 90-BF 80-BF 80-BF*
585   {  4, 0x80, 0xBF, },  // 8:  80-BF 80-BF 80-BF
586   {  8, 0xF1, 0xF3, },  // 9: F1-F3 80-BF 80-BF 80-BF*
587   {  4, 0x80, 0x8F, },  // 10: 80-8F 80-BF 80-BF
588   { 10, 0xF4, 0xF4, },  // 11: F4 80-8F 80-BF 80-BF*
589 };
590 
Add_80_10ffff()591 void Compiler::Add_80_10ffff() {
592   int inst[arraysize(prog_80_10ffff)] = { 0 }; // does not need to be initialized; silences gcc warning
593   for (int i = 0; i < arraysize(prog_80_10ffff); i++) {
594     const ByteRangeProg& p = prog_80_10ffff[i];
595     int next = 0;
596     if (p.next >= 0)
597       next = inst[p.next];
598     inst[i] = UncachedRuneByteSuffix(p.lo, p.hi, false, next);
599     if ((p.lo & 0xC0) != 0x80)
600       AddSuffix(inst[i]);
601   }
602 }
603 
AddRuneRangeUTF8(Rune lo,Rune hi,bool foldcase)604 void Compiler::AddRuneRangeUTF8(Rune lo, Rune hi, bool foldcase) {
605   if (lo > hi)
606     return;
607 
608   // Pick off 80-10FFFF as a common special case
609   // that can bypass the slow rune_cache_.
610   if (lo == 0x80 && hi == 0x10ffff && !reversed_) {
611     Add_80_10ffff();
612     return;
613   }
614 
615   // Split range into same-length sized ranges.
616   for (int i = 1; i < UTFmax; i++) {
617     Rune max = MaxRune(i);
618     if (lo <= max && max < hi) {
619       AddRuneRangeUTF8(lo, max, foldcase);
620       AddRuneRangeUTF8(max+1, hi, foldcase);
621       return;
622     }
623   }
624 
625   // ASCII range is always a special case.
626   if (hi < Runeself) {
627     AddSuffix(RuneByteSuffix(lo, hi, foldcase, 0));
628     return;
629   }
630 
631   // Split range into sections that agree on leading bytes.
632   for (int i = 1; i < UTFmax; i++) {
633     uint m = (1<<(6*i)) - 1;  // last i bytes of a UTF-8 sequence
634     if ((lo & ~m) != (hi & ~m)) {
635       if ((lo & m) != 0) {
636         AddRuneRangeUTF8(lo, lo|m, foldcase);
637         AddRuneRangeUTF8((lo|m)+1, hi, foldcase);
638         return;
639       }
640       if ((hi & m) != m) {
641         AddRuneRangeUTF8(lo, (hi&~m)-1, foldcase);
642         AddRuneRangeUTF8(hi&~m, hi, foldcase);
643         return;
644       }
645     }
646   }
647 
648   // Finally.  Generate byte matching equivalent for lo-hi.
649   uint8 ulo[UTFmax], uhi[UTFmax];
650   int n = runetochar(reinterpret_cast<char*>(ulo), &lo);
651   int m = runetochar(reinterpret_cast<char*>(uhi), &hi);
652   (void)m;  // USED(m)
653   DCHECK_EQ(n, m);
654 
655   int id = 0;
656   if (reversed_) {
657     for (int i = 0; i < n; i++)
658       id = RuneByteSuffix(ulo[i], uhi[i], false, id);
659   } else {
660     for (int i = n-1; i >= 0; i--)
661       id = RuneByteSuffix(ulo[i], uhi[i], false, id);
662   }
663   AddSuffix(id);
664 }
665 
666 // Should not be called.
Copy(Frag arg)667 Frag Compiler::Copy(Frag arg) {
668   // We're using WalkExponential; there should be no copying.
669   LOG(DFATAL) << "Compiler::Copy called!";
670   failed_ = true;
671   return NoMatch();
672 }
673 
674 // Visits a node quickly; called once WalkExponential has
675 // decided to cut this walk short.
ShortVisit(Regexp * re,Frag)676 Frag Compiler::ShortVisit(Regexp* re, Frag) {
677   failed_ = true;
678   return NoMatch();
679 }
680 
681 // Called before traversing a node's children during the walk.
PreVisit(Regexp * re,Frag,bool * stop)682 Frag Compiler::PreVisit(Regexp* re, Frag, bool* stop) {
683   // Cut off walk if we've already failed.
684   if (failed_)
685     *stop = true;
686 
687   return kNullFrag;  // not used by caller
688 }
689 
Literal(Rune r,bool foldcase)690 Frag Compiler::Literal(Rune r, bool foldcase) {
691   switch (encoding_) {
692     default:
693       return kNullFrag;
694 
695     case kEncodingLatin1:
696       return ByteRange(r, r, foldcase);
697 
698     case kEncodingUTF8: {
699       if (r < Runeself)  // Make common case fast.
700         return ByteRange(r, r, foldcase);
701       uint8 buf[UTFmax];
702       int n = runetochar(reinterpret_cast<char*>(buf), &r);
703       Frag f = ByteRange((uint8)buf[0], buf[0], false);
704       for (int i = 1; i < n; i++)
705         f = Cat(f, ByteRange((uint8)buf[i], buf[i], false));
706       return f;
707     }
708   }
709 }
710 
711 // Called after traversing the node's children during the walk.
712 // Given their frags, build and return the frag for this re.
PostVisit(Regexp * re,Frag,Frag,Frag * child_frags,int nchild_frags)713 Frag Compiler::PostVisit(Regexp* re, Frag, Frag, Frag* child_frags,
714                          int nchild_frags) {
715   // If a child failed, don't bother going forward, especially
716   // since the child_frags might contain Frags with NULLs in them.
717   if (failed_)
718     return NoMatch();
719 
720   // Given the child fragments, return the fragment for this node.
721   switch (re->op()) {
722     case kRegexpRepeat:
723       // Should not see; code at bottom of function will print error
724       break;
725 
726     case kRegexpNoMatch:
727       return NoMatch();
728 
729     case kRegexpEmptyMatch:
730       return Nop();
731 
732     case kRegexpHaveMatch: {
733       Frag f = Match(re->match_id());
734       // Remember unanchored match to end of string.
735       if (anchor_ != RE2::ANCHOR_BOTH)
736         f = Cat(DotStar(), Cat(EmptyWidth(kEmptyEndText), f));
737       return f;
738     }
739 
740     case kRegexpConcat: {
741       Frag f = child_frags[0];
742       for (int i = 1; i < nchild_frags; i++)
743         f = Cat(f, child_frags[i]);
744       return f;
745     }
746 
747     case kRegexpAlternate: {
748       Frag f = child_frags[0];
749       for (int i = 1; i < nchild_frags; i++)
750         f = Alt(f, child_frags[i]);
751       return f;
752     }
753 
754     case kRegexpStar:
755       return Star(child_frags[0], re->parse_flags()&Regexp::NonGreedy);
756 
757     case kRegexpPlus:
758       return Plus(child_frags[0], re->parse_flags()&Regexp::NonGreedy);
759 
760     case kRegexpQuest:
761       return Quest(child_frags[0], re->parse_flags()&Regexp::NonGreedy);
762 
763     case kRegexpLiteral:
764       return Literal(re->rune(), re->parse_flags()&Regexp::FoldCase);
765 
766     case kRegexpLiteralString: {
767       // Concatenation of literals.
768       if (re->nrunes() == 0)
769         return Nop();
770       Frag f;
771       for (int i = 0; i < re->nrunes(); i++) {
772         Frag f1 = Literal(re->runes()[i], re->parse_flags()&Regexp::FoldCase);
773         if (i == 0)
774           f = f1;
775         else
776           f = Cat(f, f1);
777       }
778       return f;
779     }
780 
781     case kRegexpAnyChar:
782       BeginRange();
783       AddRuneRange(0, Runemax, false);
784       return EndRange();
785 
786     case kRegexpAnyByte:
787       return ByteRange(0x00, 0xFF, false);
788 
789     case kRegexpCharClass: {
790       CharClass* cc = re->cc();
791       if (cc->empty()) {
792         // This can't happen.
793         LOG(DFATAL) << "No ranges in char class";
794         failed_ = true;
795         return NoMatch();
796       }
797 
798       // ASCII case-folding optimization: if the char class
799       // behaves the same on A-Z as it does on a-z,
800       // discard any ranges wholly contained in A-Z
801       // and mark the other ranges as foldascii.
802       // This reduces the size of a program for
803       // (?i)abc from 3 insts per letter to 1 per letter.
804       bool foldascii = cc->FoldsASCII();
805 
806       // Character class is just a big OR of the different
807       // character ranges in the class.
808       BeginRange();
809       for (CharClass::iterator i = cc->begin(); i != cc->end(); ++i) {
810         // ASCII case-folding optimization (see above).
811         if (foldascii && 'A' <= i->lo && i->hi <= 'Z')
812           continue;
813 
814         // If this range contains all of A-Za-z or none of it,
815         // the fold flag is unnecessary; don't bother.
816         bool fold = foldascii;
817         if ((i->lo <= 'A' && 'z' <= i->hi) || i->hi < 'A' || 'z' < i->lo)
818           fold = false;
819 
820         AddRuneRange(i->lo, i->hi, fold);
821       }
822       return EndRange();
823     }
824 
825     case kRegexpCapture:
826       // If this is a non-capturing parenthesis -- (?:foo) --
827       // just use the inner expression.
828       if (re->cap() < 0)
829         return child_frags[0];
830       return Capture(child_frags[0], re->cap());
831 
832     case kRegexpBeginLine:
833       return EmptyWidth(reversed_ ? kEmptyEndLine : kEmptyBeginLine);
834 
835     case kRegexpEndLine:
836       return EmptyWidth(reversed_ ? kEmptyBeginLine : kEmptyEndLine);
837 
838     case kRegexpBeginText:
839       return EmptyWidth(reversed_ ? kEmptyEndText : kEmptyBeginText);
840 
841     case kRegexpEndText:
842       return EmptyWidth(reversed_ ? kEmptyBeginText : kEmptyEndText);
843 
844     case kRegexpWordBoundary:
845       return EmptyWidth(kEmptyWordBoundary);
846 
847     case kRegexpNoWordBoundary:
848       return EmptyWidth(kEmptyNonWordBoundary);
849   }
850   LOG(DFATAL) << "Missing case in Compiler: " << re->op();
851   failed_ = true;
852   return NoMatch();
853 }
854 
855 // Is this regexp required to start at the beginning of the text?
856 // Only approximate; can return false for complicated regexps like (\Aa|\Ab),
857 // but handles (\A(a|b)).  Could use the Walker to write a more exact one.
IsAnchorStart(Regexp ** pre,int depth)858 static bool IsAnchorStart(Regexp** pre, int depth) {
859   Regexp* re = *pre;
860   Regexp* sub;
861   // The depth limit makes sure that we don't overflow
862   // the stack on a deeply nested regexp.  As the comment
863   // above says, IsAnchorStart is conservative, so returning
864   // a false negative is okay.  The exact limit is somewhat arbitrary.
865   if (re == NULL || depth >= 4)
866     return false;
867   switch (re->op()) {
868     default:
869       break;
870     case kRegexpConcat:
871       if (re->nsub() > 0) {
872         sub = re->sub()[0]->Incref();
873         if (IsAnchorStart(&sub, depth+1)) {
874           Regexp** subcopy = new Regexp*[re->nsub()];
875           subcopy[0] = sub;  // already have reference
876           for (int i = 1; i < re->nsub(); i++)
877             subcopy[i] = re->sub()[i]->Incref();
878           *pre = Regexp::Concat(subcopy, re->nsub(), re->parse_flags());
879           delete[] subcopy;
880           re->Decref();
881           return true;
882         }
883         sub->Decref();
884       }
885       break;
886     case kRegexpCapture:
887       sub = re->sub()[0]->Incref();
888       if (IsAnchorStart(&sub, depth+1)) {
889         *pre = Regexp::Capture(sub, re->parse_flags(), re->cap());
890         re->Decref();
891         return true;
892       }
893       sub->Decref();
894       break;
895     case kRegexpBeginText:
896       *pre = Regexp::LiteralString(NULL, 0, re->parse_flags());
897       re->Decref();
898       return true;
899   }
900   return false;
901 }
902 
903 // Is this regexp required to start at the end of the text?
904 // Only approximate; can return false for complicated regexps like (a\z|b\z),
905 // but handles ((a|b)\z).  Could use the Walker to write a more exact one.
IsAnchorEnd(Regexp ** pre,int depth)906 static bool IsAnchorEnd(Regexp** pre, int depth) {
907   Regexp* re = *pre;
908   Regexp* sub;
909   // The depth limit makes sure that we don't overflow
910   // the stack on a deeply nested regexp.  As the comment
911   // above says, IsAnchorEnd is conservative, so returning
912   // a false negative is okay.  The exact limit is somewhat arbitrary.
913   if (re == NULL || depth >= 4)
914     return false;
915   switch (re->op()) {
916     default:
917       break;
918     case kRegexpConcat:
919       if (re->nsub() > 0) {
920         sub = re->sub()[re->nsub() - 1]->Incref();
921         if (IsAnchorEnd(&sub, depth+1)) {
922           Regexp** subcopy = new Regexp*[re->nsub()];
923           subcopy[re->nsub() - 1] = sub;  // already have reference
924           for (int i = 0; i < re->nsub() - 1; i++)
925             subcopy[i] = re->sub()[i]->Incref();
926           *pre = Regexp::Concat(subcopy, re->nsub(), re->parse_flags());
927           delete[] subcopy;
928           re->Decref();
929           return true;
930         }
931         sub->Decref();
932       }
933       break;
934     case kRegexpCapture:
935       sub = re->sub()[0]->Incref();
936       if (IsAnchorEnd(&sub, depth+1)) {
937         *pre = Regexp::Capture(sub, re->parse_flags(), re->cap());
938         re->Decref();
939         return true;
940       }
941       sub->Decref();
942       break;
943     case kRegexpEndText:
944       *pre = Regexp::LiteralString(NULL, 0, re->parse_flags());
945       re->Decref();
946       return true;
947   }
948   return false;
949 }
950 
Setup(Regexp::ParseFlags flags,int64 max_mem,RE2::Anchor anchor)951 void Compiler::Setup(Regexp::ParseFlags flags, int64 max_mem,
952                      RE2::Anchor anchor) {
953   prog_->set_flags(flags);
954 
955   if (flags & Regexp::Latin1)
956     encoding_ = kEncodingLatin1;
957   max_mem_ = max_mem;
958   if (max_mem <= 0) {
959     max_inst_ = 100000;  // more than enough
960   } else if (max_mem <= sizeof(Prog)) {
961     // No room for anything.
962     max_inst_ = 0;
963   } else {
964     int64 m = (max_mem - sizeof(Prog)) / sizeof(Prog::Inst);
965     // Limit instruction count so that inst->id() fits nicely in an int.
966     // SparseArray also assumes that the indices (inst->id()) are ints.
967     // The call to WalkExponential uses 2*max_inst_ below,
968     // and other places in the code use 2 or 3 * prog->size().
969     // Limiting to 2^24 should avoid overflow in those places.
970     // (The point of allowing more than 32 bits of memory is to
971     // have plenty of room for the DFA states, not to use it up
972     // on the program.)
973     if (m >= 1<<24)
974       m = 1<<24;
975 
976     // Inst imposes its own limit (currently bigger than 2^24 but be safe).
977     if (m > Prog::Inst::kMaxInst)
978       m = Prog::Inst::kMaxInst;
979 
980     max_inst_ = m;
981   }
982 
983   anchor_ = anchor;
984 }
985 
986 // Compiles re, returning program.
987 // Caller is responsible for deleting prog_.
988 // If reversed is true, compiles a program that expects
989 // to run over the input string backward (reverses all concatenations).
990 // The reversed flag is also recorded in the returned program.
Compile(Regexp * re,bool reversed,int64 max_mem)991 Prog* Compiler::Compile(Regexp* re, bool reversed, int64 max_mem) {
992   Compiler c;
993 
994   c.Setup(re->parse_flags(), max_mem, RE2::ANCHOR_BOTH /* unused */);
995   c.reversed_ = reversed;
996 
997   // Simplify to remove things like counted repetitions
998   // and character classes like \d.
999   Regexp* sre = re->Simplify();
1000   if (sre == NULL)
1001     return NULL;
1002 
1003   // Record whether prog is anchored, removing the anchors.
1004   // (They get in the way of other optimizations.)
1005   bool is_anchor_start = IsAnchorStart(&sre, 0);
1006   bool is_anchor_end = IsAnchorEnd(&sre, 0);
1007 
1008   // Generate fragment for entire regexp.
1009   Frag f = c.WalkExponential(sre, kNullFrag, 2*c.max_inst_);
1010   sre->Decref();
1011   if (c.failed_)
1012     return NULL;
1013 
1014   // Success!  Finish by putting Match node at end, and record start.
1015   // Turn off c.reversed_ (if it is set) to force the remaining concatenations
1016   // to behave normally.
1017   c.reversed_ = false;
1018   Frag all = c.Cat(f, c.Match(0));
1019   c.prog_->set_start(all.begin);
1020 
1021   if (reversed) {
1022     c.prog_->set_anchor_start(is_anchor_end);
1023     c.prog_->set_anchor_end(is_anchor_start);
1024   } else {
1025     c.prog_->set_anchor_start(is_anchor_start);
1026     c.prog_->set_anchor_end(is_anchor_end);
1027   }
1028 
1029   // Also create unanchored version, which starts with a .*? loop.
1030   if (c.prog_->anchor_start()) {
1031     c.prog_->set_start_unanchored(c.prog_->start());
1032   } else {
1033     Frag unanchored = c.Cat(c.DotStar(), all);
1034     c.prog_->set_start_unanchored(unanchored.begin);
1035   }
1036 
1037   c.prog_->set_reversed(reversed);
1038 
1039   // Hand ownership of prog_ to caller.
1040   return c.Finish();
1041 }
1042 
Finish()1043 Prog* Compiler::Finish() {
1044   if (failed_)
1045     return NULL;
1046 
1047   if (prog_->start() == 0 && prog_->start_unanchored() == 0) {
1048     // No possible matches; keep Fail instruction only.
1049     inst_len_ = 1;
1050   }
1051 
1052   // Trim instruction to minimum array and transfer to Prog.
1053   Trim();
1054   prog_->inst_ = inst_;
1055   prog_->size_ = inst_len_;
1056   inst_ = NULL;
1057 
1058   // Compute byte map.
1059   prog_->ComputeByteMap();
1060 
1061   prog_->Optimize();
1062 
1063   // Record remaining memory for DFA.
1064   if (max_mem_ <= 0) {
1065     prog_->set_dfa_mem(1<<20);
1066   } else {
1067     int64 m = max_mem_ - sizeof(Prog) - inst_len_*sizeof(Prog::Inst);
1068     if (m < 0)
1069       m = 0;
1070     prog_->set_dfa_mem(m);
1071   }
1072 
1073   Prog* p = prog_;
1074   prog_ = NULL;
1075   return p;
1076 }
1077 
1078 // Converts Regexp to Prog.
CompileToProg(int64 max_mem)1079 Prog* Regexp::CompileToProg(int64 max_mem) {
1080   return Compiler::Compile(this, false, max_mem);
1081 }
1082 
CompileToReverseProg(int64 max_mem)1083 Prog* Regexp::CompileToReverseProg(int64 max_mem) {
1084   return Compiler::Compile(this, true, max_mem);
1085 }
1086 
DotStar()1087 Frag Compiler::DotStar() {
1088   return Star(ByteRange(0x00, 0xff, false), true);
1089 }
1090 
1091 // Compiles RE set to Prog.
CompileSet(const RE2::Options & options,RE2::Anchor anchor,Regexp * re)1092 Prog* Compiler::CompileSet(const RE2::Options& options, RE2::Anchor anchor,
1093                            Regexp* re) {
1094   Compiler c;
1095 
1096   Regexp::ParseFlags pf = static_cast<Regexp::ParseFlags>(options.ParseFlags());
1097   c.Setup(pf, options.max_mem(), anchor);
1098 
1099   // Compile alternation of fragments.
1100   Frag all = c.WalkExponential(re, kNullFrag, 2*c.max_inst_);
1101   re->Decref();
1102   if (c.failed_)
1103     return NULL;
1104 
1105   if (anchor == RE2::UNANCHORED) {
1106     // The trailing .* was added while handling kRegexpHaveMatch.
1107     // We just have to add the leading one.
1108     all = c.Cat(c.DotStar(), all);
1109   }
1110 
1111   c.prog_->set_start(all.begin);
1112   c.prog_->set_start_unanchored(all.begin);
1113   c.prog_->set_anchor_start(true);
1114   c.prog_->set_anchor_end(true);
1115 
1116   Prog* prog = c.Finish();
1117   if (prog == NULL)
1118     return NULL;
1119 
1120   // Make sure DFA has enough memory to operate,
1121   // since we're not going to fall back to the NFA.
1122   bool failed;
1123   StringPiece sp = "hello, world";
1124   prog->SearchDFA(sp, sp, Prog::kAnchored, Prog::kManyMatch,
1125                   NULL, &failed, NULL);
1126   if (failed) {
1127     delete prog;
1128     return NULL;
1129   }
1130 
1131   return prog;
1132 }
1133 
CompileSet(const RE2::Options & options,RE2::Anchor anchor,Regexp * re)1134 Prog* Prog::CompileSet(const RE2::Options& options, RE2::Anchor anchor,
1135                        Regexp* re) {
1136   return Compiler::CompileSet(options, anchor, re);
1137 }
1138 
1139 }  // namespace re2
1140