michael@0: /* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
michael@0:  * vim: set ts=8 sts=4 et sw=4 tw=99:
michael@0:  *
michael@0:  * Copyright (C) 2009 Apple Inc. All rights reserved.
michael@0:  * Copyright (C) 2010 Peter Varga (pvarga@inf.u-szeged.hu), University of Szeged
michael@0:  *
michael@0:  * Redistribution and use in source and binary forms, with or without
michael@0:  * modification, are permitted provided that the following conditions
michael@0:  * are met:
michael@0:  * 1. Redistributions of source code must retain the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer.
michael@0:  * 2. Redistributions in binary form must reproduce the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer in the
michael@0:  *    documentation and/or other materials provided with the distribution.
michael@0:  *
michael@0:  * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY
michael@0:  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
michael@0:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
michael@0:  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL APPLE INC. OR
michael@0:  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
michael@0:  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
michael@0:  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
michael@0:  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
michael@0:  * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
michael@0:  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
michael@0:  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
michael@0:  */
michael@0: 
michael@0: #include "yarr/YarrPattern.h"
michael@0: 
michael@0: #include "yarr/Yarr.h"
michael@0: #include "yarr/YarrCanonicalizeUCS2.h"
michael@0: #include "yarr/YarrParser.h"
michael@0: 
michael@0: using namespace WTF;
michael@0: 
michael@0: namespace JSC { namespace Yarr {
michael@0: 
michael@0: #include "yarr/RegExpJitTables.h"
michael@0: 
michael@0: #if WTF_CPU_SPARC
michael@0: # define BASE_FRAME_SIZE 24
michael@0: #else
michael@0: # define BASE_FRAME_SIZE 0
michael@0: #endif
michael@0: 
michael@0: // Thanks, windows.h!
michael@0: #undef min
michael@0: #undef max
michael@0: 
michael@0: class CharacterClassConstructor {
michael@0: public:
michael@0:     CharacterClassConstructor(bool isCaseInsensitive = false)
michael@0:         : m_isCaseInsensitive(isCaseInsensitive)
michael@0:     {
michael@0:     }
michael@0: 
michael@0:     void reset()
michael@0:     {
michael@0:         m_matches.clear();
michael@0:         m_ranges.clear();
michael@0:         m_matchesUnicode.clear();
michael@0:         m_rangesUnicode.clear();
michael@0:     }
michael@0: 
michael@0:     void append(const CharacterClass* other)
michael@0:     {
michael@0:         for (size_t i = 0; i < other->m_matches.size(); ++i)
michael@0:             addSorted(m_matches, other->m_matches[i]);
michael@0:         for (size_t i = 0; i < other->m_ranges.size(); ++i)
michael@0:             addSortedRange(m_ranges, other->m_ranges[i].begin, other->m_ranges[i].end);
michael@0:         for (size_t i = 0; i < other->m_matchesUnicode.size(); ++i)
michael@0:             addSorted(m_matchesUnicode, other->m_matchesUnicode[i]);
michael@0:         for (size_t i = 0; i < other->m_rangesUnicode.size(); ++i)
michael@0:             addSortedRange(m_rangesUnicode, other->m_rangesUnicode[i].begin, other->m_rangesUnicode[i].end);
michael@0:     }
michael@0: 
michael@0:     void putChar(UChar ch)
michael@0:     {
michael@0:         // Handle ascii cases.
michael@0:         if (ch <= 0x7f) {
michael@0:             if (m_isCaseInsensitive && isASCIIAlpha(ch)) {
michael@0:                 addSorted(m_matches, toASCIIUpper(ch));
michael@0:                 addSorted(m_matches, toASCIILower(ch));
michael@0:             } else
michael@0:                 addSorted(m_matches, ch);
michael@0:             return;
michael@0:         }
michael@0: 
michael@0:         // Simple case, not a case-insensitive match.
michael@0:         if (!m_isCaseInsensitive) {
michael@0:             addSorted(m_matchesUnicode, ch);
michael@0:             return;
michael@0:         }
michael@0: 
michael@0:         // Add multiple matches, if necessary.
michael@0:         const UCS2CanonicalizationRange* info = rangeInfoFor(ch);
michael@0:         if (info->type == CanonicalizeUnique)
michael@0:             addSorted(m_matchesUnicode, ch);
michael@0:         else
michael@0:             putUnicodeIgnoreCase(ch, info);
michael@0:     }
michael@0: 
michael@0:     void putUnicodeIgnoreCase(UChar ch, const UCS2CanonicalizationRange* info)
michael@0:     {
michael@0:         ASSERT(m_isCaseInsensitive);
michael@0:         ASSERT(ch > 0x7f);
michael@0:         ASSERT(ch >= info->begin && ch <= info->end);
michael@0:         ASSERT(info->type != CanonicalizeUnique);
michael@0:         if (info->type == CanonicalizeSet) {
michael@0:             for (const uint16_t* set = characterSetInfo[info->value]; (ch = *set); ++set)
michael@0:                 addSorted(m_matchesUnicode, ch);
michael@0:         } else {
michael@0:             addSorted(m_matchesUnicode, ch);
michael@0:             addSorted(m_matchesUnicode, getCanonicalPair(info, ch));
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     void putRange(UChar lo, UChar hi)
michael@0:     {
michael@0:         if (lo <= 0x7f) {
michael@0:             char asciiLo = lo;
michael@0:             char asciiHi = std::min(hi, (UChar)0x7f);
michael@0:             addSortedRange(m_ranges, lo, asciiHi);
michael@0: 
michael@0:             if (m_isCaseInsensitive) {
michael@0:                 if ((asciiLo <= 'Z') && (asciiHi >= 'A'))
michael@0:                     addSortedRange(m_ranges, std::max(asciiLo, 'A')+('a'-'A'), std::min(asciiHi, 'Z')+('a'-'A'));
michael@0:                 if ((asciiLo <= 'z') && (asciiHi >= 'a'))
michael@0:                     addSortedRange(m_ranges, std::max(asciiLo, 'a')+('A'-'a'), std::min(asciiHi, 'z')+('A'-'a'));
michael@0:             }
michael@0:         }
michael@0:         if (hi <= 0x7f)
michael@0:             return;
michael@0: 
michael@0:         lo = std::max(lo, (UChar)0x80);
michael@0:         addSortedRange(m_rangesUnicode, lo, hi);
michael@0: 
michael@0:         if (!m_isCaseInsensitive)
michael@0:             return;
michael@0: 
michael@0:         const UCS2CanonicalizationRange* info = rangeInfoFor(lo);
michael@0:         while (true) {
michael@0:             // Handle the range [lo .. end]
michael@0:             UChar end = std::min<UChar>(info->end, hi);
michael@0: 
michael@0:             switch (info->type) {
michael@0:             case CanonicalizeUnique:
michael@0:                 // Nothing to do - no canonical equivalents.
michael@0:                 break;
michael@0:             case CanonicalizeSet: {
michael@0:                 UChar ch;
michael@0:                 for (const uint16_t* set = characterSetInfo[info->value]; (ch = *set); ++set)
michael@0:                     addSorted(m_matchesUnicode, ch);
michael@0:                 break;
michael@0:             }
michael@0:             case CanonicalizeRangeLo:
michael@0:                 addSortedRange(m_rangesUnicode, lo + info->value, end + info->value);
michael@0:                 break;
michael@0:             case CanonicalizeRangeHi:
michael@0:                 addSortedRange(m_rangesUnicode, lo - info->value, end - info->value);
michael@0:                 break;
michael@0:             case CanonicalizeAlternatingAligned:
michael@0:                 // Use addSortedRange since there is likely an abutting range to combine with.
michael@0:                 if (lo & 1)
michael@0:                     addSortedRange(m_rangesUnicode, lo - 1, lo - 1);
michael@0:                 if (!(end & 1))
michael@0:                     addSortedRange(m_rangesUnicode, end + 1, end + 1);
michael@0:                 break;
michael@0:             case CanonicalizeAlternatingUnaligned:
michael@0:                 // Use addSortedRange since there is likely an abutting range to combine with.
michael@0:                 if (!(lo & 1))
michael@0:                     addSortedRange(m_rangesUnicode, lo - 1, lo - 1);
michael@0:                 if (end & 1)
michael@0:                     addSortedRange(m_rangesUnicode, end + 1, end + 1);
michael@0:                 break;
michael@0:             }
michael@0: 
michael@0:             if (hi == end)
michael@0:                 return;
michael@0: 
michael@0:             ++info;
michael@0:             lo = info->begin;
michael@0:         };
michael@0: 
michael@0:     }
michael@0: 
michael@0:     CharacterClass* charClass()
michael@0:     {
michael@0:         CharacterClass* characterClass = newOrCrash<CharacterClass>();
michael@0: 
michael@0:         characterClass->m_matches.swap(m_matches);
michael@0:         characterClass->m_ranges.swap(m_ranges);
michael@0:         characterClass->m_matchesUnicode.swap(m_matchesUnicode);
michael@0:         characterClass->m_rangesUnicode.swap(m_rangesUnicode);
michael@0: 
michael@0:         return characterClass;
michael@0:     }
michael@0: 
michael@0: private:
michael@0:     void addSorted(Vector<UChar>& matches, UChar ch)
michael@0:     {
michael@0:         unsigned pos = 0;
michael@0:         unsigned range = matches.size();
michael@0: 
michael@0:         // binary chop, find position to insert char.
michael@0:         while (range) {
michael@0:             unsigned index = range >> 1;
michael@0: 
michael@0:             int val = matches[pos+index] - ch;
michael@0:             if (!val)
michael@0:                 return;
michael@0:             else if (val > 0)
michael@0:                 range = index;
michael@0:             else {
michael@0:                 pos += (index+1);
michael@0:                 range -= (index+1);
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         if (pos == matches.size())
michael@0:             matches.append(ch);
michael@0:         else
michael@0:             matches.insert(pos, ch);
michael@0:     }
michael@0: 
michael@0:     void addSortedRange(Vector<CharacterRange>& ranges, UChar lo, UChar hi)
michael@0:     {
michael@0:         unsigned end = ranges.size();
michael@0: 
michael@0:         // Simple linear scan - I doubt there are that many ranges anyway...
michael@0:         // feel free to fix this with something faster (eg binary chop).
michael@0:         for (unsigned i = 0; i < end; ++i) {
michael@0:             // does the new range fall before the current position in the array
michael@0:             if (hi < ranges[i].begin) {
michael@0:                 // optional optimization: concatenate appending ranges? - may not be worthwhile.
michael@0:                 if (hi == (ranges[i].begin - 1)) {
michael@0:                     ranges[i].begin = lo;
michael@0:                     return;
michael@0:                 }
michael@0:                 ranges.insert(i, CharacterRange(lo, hi));
michael@0:                 return;
michael@0:             }
michael@0:             // Okay, since we didn't hit the last case, the end of the new range is definitely at or after the begining
michael@0:             // If the new range start at or before the end of the last range, then the overlap (if it starts one after the
michael@0:             // end of the last range they concatenate, which is just as good.
michael@0:             if (lo <= (ranges[i].end + 1)) {
michael@0:                 // found an intersect! we'll replace this entry in the array.
michael@0:                 ranges[i].begin = std::min(ranges[i].begin, lo);
michael@0:                 ranges[i].end = std::max(ranges[i].end, hi);
michael@0: 
michael@0:                 // now check if the new range can subsume any subsequent ranges.
michael@0:                 unsigned next = i+1;
michael@0:                 // each iteration of the loop we will either remove something from the list, or break the loop.
michael@0:                 while (next < ranges.size()) {
michael@0:                     if (ranges[next].begin <= (ranges[i].end + 1)) {
michael@0:                         // the next entry now overlaps / concatenates this one.
michael@0:                         ranges[i].end = std::max(ranges[i].end, ranges[next].end);
michael@0:                         ranges.remove(next);
michael@0:                     } else
michael@0:                         break;
michael@0:                 }
michael@0: 
michael@0:                 return;
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         // CharacterRange comes after all existing ranges.
michael@0:         ranges.append(CharacterRange(lo, hi));
michael@0:     }
michael@0: 
michael@0:     bool m_isCaseInsensitive;
michael@0: 
michael@0:     Vector<UChar> m_matches;
michael@0:     Vector<CharacterRange> m_ranges;
michael@0:     Vector<UChar> m_matchesUnicode;
michael@0:     Vector<CharacterRange> m_rangesUnicode;
michael@0: };
michael@0: 
michael@0: class YarrPatternConstructor {
michael@0: public:
michael@0:     YarrPatternConstructor(YarrPattern& pattern)
michael@0:         : m_pattern(pattern)
michael@0:         , m_stackBase(nullptr)
michael@0:         , m_characterClassConstructor(pattern.m_ignoreCase)
michael@0:         , m_invertParentheticalAssertion(false)
michael@0:     {
michael@0:         m_pattern.m_body = newOrCrash<PatternDisjunction>();
michael@0:         m_alternative = m_pattern.m_body->addNewAlternative();
michael@0:         m_pattern.m_disjunctions.append(m_pattern.m_body);
michael@0:     }
michael@0: 
michael@0:     ~YarrPatternConstructor()
michael@0:     {
michael@0:     }
michael@0: 
michael@0:     void reset()
michael@0:     {
michael@0:         m_pattern.reset();
michael@0:         m_characterClassConstructor.reset();
michael@0: 
michael@0:         m_pattern.m_body = newOrCrash<PatternDisjunction>();
michael@0:         m_alternative = m_pattern.m_body->addNewAlternative();
michael@0:         m_pattern.m_disjunctions.append(m_pattern.m_body);
michael@0:     }
michael@0: 
michael@0:     void assertionBOL()
michael@0:     {
michael@0:         if (!m_alternative->m_terms.size() & !m_invertParentheticalAssertion) {
michael@0:             m_alternative->m_startsWithBOL = true;
michael@0:             m_alternative->m_containsBOL = true;
michael@0:             m_pattern.m_containsBOL = true;
michael@0:         }
michael@0:         m_alternative->m_terms.append(PatternTerm::BOL());
michael@0:     }
michael@0:     void assertionEOL()
michael@0:     {
michael@0:         m_alternative->m_terms.append(PatternTerm::EOL());
michael@0:     }
michael@0:     void assertionWordBoundary(bool invert)
michael@0:     {
michael@0:         m_alternative->m_terms.append(PatternTerm::WordBoundary(invert));
michael@0:     }
michael@0: 
michael@0:     void atomPatternCharacter(UChar ch)
michael@0:     {
michael@0:         // We handle case-insensitive checking of unicode characters which do have both
michael@0:         // cases by handling them as if they were defined using a CharacterClass.
michael@0:         if (!m_pattern.m_ignoreCase || isASCII(ch)) {
michael@0:             m_alternative->m_terms.append(PatternTerm(ch));
michael@0:             return;
michael@0:         }
michael@0: 
michael@0:         const UCS2CanonicalizationRange* info = rangeInfoFor(ch);
michael@0:         if (info->type == CanonicalizeUnique) {
michael@0:             m_alternative->m_terms.append(PatternTerm(ch));
michael@0:             return;
michael@0:         }
michael@0: 
michael@0:         m_characterClassConstructor.putUnicodeIgnoreCase(ch, info);
michael@0:         CharacterClass* newCharacterClass = m_characterClassConstructor.charClass();
michael@0:         m_pattern.m_userCharacterClasses.append(newCharacterClass);
michael@0:         m_alternative->m_terms.append(PatternTerm(newCharacterClass, false));
michael@0:     }
michael@0: 
michael@0:     void atomBuiltInCharacterClass(BuiltInCharacterClassID classID, bool invert)
michael@0:     {
michael@0:         switch (classID) {
michael@0:         case DigitClassID:
michael@0:             m_alternative->m_terms.append(PatternTerm(m_pattern.digitsCharacterClass(), invert));
michael@0:             break;
michael@0:         case SpaceClassID:
michael@0:             m_alternative->m_terms.append(PatternTerm(m_pattern.spacesCharacterClass(), invert));
michael@0:             break;
michael@0:         case WordClassID:
michael@0:             m_alternative->m_terms.append(PatternTerm(m_pattern.wordcharCharacterClass(), invert));
michael@0:             break;
michael@0:         case NewlineClassID:
michael@0:             m_alternative->m_terms.append(PatternTerm(m_pattern.newlineCharacterClass(), invert));
michael@0:             break;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     void atomCharacterClassBegin(bool invert = false)
michael@0:     {
michael@0:         m_invertCharacterClass = invert;
michael@0:     }
michael@0: 
michael@0:     void atomCharacterClassAtom(UChar ch)
michael@0:     {
michael@0:         m_characterClassConstructor.putChar(ch);
michael@0:     }
michael@0: 
michael@0:     void atomCharacterClassRange(UChar begin, UChar end)
michael@0:     {
michael@0:         m_characterClassConstructor.putRange(begin, end);
michael@0:     }
michael@0: 
michael@0:     void atomCharacterClassBuiltIn(BuiltInCharacterClassID classID, bool invert)
michael@0:     {
michael@0:         ASSERT(classID != NewlineClassID);
michael@0: 
michael@0:         switch (classID) {
michael@0:         case DigitClassID:
michael@0:             m_characterClassConstructor.append(invert ? m_pattern.nondigitsCharacterClass() : m_pattern.digitsCharacterClass());
michael@0:             break;
michael@0: 
michael@0:         case SpaceClassID:
michael@0:             m_characterClassConstructor.append(invert ? m_pattern.nonspacesCharacterClass() : m_pattern.spacesCharacterClass());
michael@0:             break;
michael@0: 
michael@0:         case WordClassID:
michael@0:             m_characterClassConstructor.append(invert ? m_pattern.nonwordcharCharacterClass() : m_pattern.wordcharCharacterClass());
michael@0:             break;
michael@0: 
michael@0:         default:
michael@0:             ASSERT_NOT_REACHED();
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     void atomCharacterClassEnd()
michael@0:     {
michael@0:         CharacterClass* newCharacterClass = m_characterClassConstructor.charClass();
michael@0:         m_pattern.m_userCharacterClasses.append(newCharacterClass);
michael@0:         m_alternative->m_terms.append(PatternTerm(newCharacterClass, m_invertCharacterClass));
michael@0:     }
michael@0: 
michael@0:     void atomParenthesesSubpatternBegin(bool capture = true)
michael@0:     {
michael@0:         unsigned subpatternId = m_pattern.m_numSubpatterns + 1;
michael@0:         if (capture)
michael@0:             m_pattern.m_numSubpatterns++;
michael@0: 
michael@0:         PatternDisjunction* parenthesesDisjunction = newOrCrash<PatternDisjunction>(m_alternative);
michael@0:         m_pattern.m_disjunctions.append(parenthesesDisjunction);
michael@0:         m_alternative->m_terms.append(PatternTerm(PatternTerm::TypeParenthesesSubpattern, subpatternId, parenthesesDisjunction, capture, false));
michael@0:         m_alternative = parenthesesDisjunction->addNewAlternative();
michael@0:     }
michael@0: 
michael@0:     void atomParentheticalAssertionBegin(bool invert = false)
michael@0:     {
michael@0:         PatternDisjunction* parenthesesDisjunction = newOrCrash<PatternDisjunction>(m_alternative);
michael@0:         m_pattern.m_disjunctions.append(parenthesesDisjunction);
michael@0:         m_alternative->m_terms.append(PatternTerm(PatternTerm::TypeParentheticalAssertion, m_pattern.m_numSubpatterns + 1, parenthesesDisjunction, false, invert));
michael@0:         m_alternative = parenthesesDisjunction->addNewAlternative();
michael@0:         m_invertParentheticalAssertion = invert;
michael@0:     }
michael@0: 
michael@0:     void atomParenthesesEnd()
michael@0:     {
michael@0:         ASSERT(m_alternative->m_parent);
michael@0:         ASSERT(m_alternative->m_parent->m_parent);
michael@0: 
michael@0:         PatternDisjunction* parenthesesDisjunction = m_alternative->m_parent;
michael@0:         m_alternative = m_alternative->m_parent->m_parent;
michael@0: 
michael@0:         PatternTerm& lastTerm = m_alternative->lastTerm();
michael@0: 
michael@0:         unsigned numParenAlternatives = parenthesesDisjunction->m_alternatives.size();
michael@0:         unsigned numBOLAnchoredAlts = 0;
michael@0: 
michael@0:         for (unsigned i = 0; i < numParenAlternatives; i++) {
michael@0:             // Bubble up BOL flags
michael@0:             if (parenthesesDisjunction->m_alternatives[i]->m_startsWithBOL)
michael@0:                 numBOLAnchoredAlts++;
michael@0:         }
michael@0: 
michael@0:         if (numBOLAnchoredAlts) {
michael@0:             m_alternative->m_containsBOL = true;
michael@0:             // If all the alternatives in parens start with BOL, then so does this one
michael@0:             if (numBOLAnchoredAlts == numParenAlternatives)
michael@0:                 m_alternative->m_startsWithBOL = true;
michael@0:         }
michael@0: 
michael@0:         lastTerm.parentheses.lastSubpatternId = m_pattern.m_numSubpatterns;
michael@0:         m_invertParentheticalAssertion = false;
michael@0:     }
michael@0: 
michael@0:     void atomBackReference(unsigned subpatternId)
michael@0:     {
michael@0:         ASSERT(subpatternId);
michael@0:         m_pattern.m_containsBackreferences = true;
michael@0:         m_pattern.m_maxBackReference = std::max(m_pattern.m_maxBackReference, subpatternId);
michael@0: 
michael@0:         if (subpatternId > m_pattern.m_numSubpatterns) {
michael@0:             m_alternative->m_terms.append(PatternTerm::ForwardReference());
michael@0:             return;
michael@0:         }
michael@0: 
michael@0:         PatternAlternative* currentAlternative = m_alternative;
michael@0:         ASSERT(currentAlternative);
michael@0: 
michael@0:         // Note to self: if we waited until the AST was baked, we could also remove forwards refs
michael@0:         while ((currentAlternative = currentAlternative->m_parent->m_parent)) {
michael@0:             PatternTerm& term = currentAlternative->lastTerm();
michael@0:             ASSERT((term.type == PatternTerm::TypeParenthesesSubpattern) || (term.type == PatternTerm::TypeParentheticalAssertion));
michael@0: 
michael@0:             if ((term.type == PatternTerm::TypeParenthesesSubpattern) && term.capture() && (subpatternId == term.parentheses.subpatternId)) {
michael@0:                 m_alternative->m_terms.append(PatternTerm::ForwardReference());
michael@0:                 return;
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         m_alternative->m_terms.append(PatternTerm(subpatternId));
michael@0:     }
michael@0: 
michael@0:     // deep copy the argument disjunction.  If filterStartsWithBOL is true,
michael@0:     // skip alternatives with m_startsWithBOL set true.
michael@0:     PatternDisjunction* copyDisjunction(PatternDisjunction* disjunction, bool filterStartsWithBOL = false)
michael@0:     {
michael@0:         PatternDisjunction* newDisjunction = 0;
michael@0:         for (unsigned alt = 0; alt < disjunction->m_alternatives.size(); ++alt) {
michael@0:             PatternAlternative* alternative = disjunction->m_alternatives[alt];
michael@0:             if (!filterStartsWithBOL || !alternative->m_startsWithBOL) {
michael@0:                 if (!newDisjunction) {
michael@0:                     newDisjunction = newOrCrash<PatternDisjunction>();
michael@0:                     newDisjunction->m_parent = disjunction->m_parent;
michael@0:                 }
michael@0:                 PatternAlternative* newAlternative = newDisjunction->addNewAlternative();
michael@0:                 newAlternative->m_terms.reserve(alternative->m_terms.size());
michael@0:                 for (unsigned i = 0; i < alternative->m_terms.size(); ++i)
michael@0:                     newAlternative->m_terms.append(copyTerm(alternative->m_terms[i], filterStartsWithBOL));
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         if (newDisjunction)
michael@0:             m_pattern.m_disjunctions.append(newDisjunction);
michael@0:         return newDisjunction;
michael@0:     }
michael@0: 
michael@0:     PatternTerm copyTerm(PatternTerm& term, bool filterStartsWithBOL = false)
michael@0:     {
michael@0:         if ((term.type != PatternTerm::TypeParenthesesSubpattern) && (term.type != PatternTerm::TypeParentheticalAssertion))
michael@0:             return PatternTerm(term);
michael@0: 
michael@0:         PatternTerm termCopy = term;
michael@0:         termCopy.parentheses.disjunction = copyDisjunction(termCopy.parentheses.disjunction, filterStartsWithBOL);
michael@0:         return termCopy;
michael@0:     }
michael@0: 
michael@0:     void quantifyAtom(unsigned min, unsigned max, bool greedy)
michael@0:     {
michael@0:         ASSERT(min <= max);
michael@0:         ASSERT(m_alternative->m_terms.size());
michael@0: 
michael@0:         if (!max) {
michael@0:             m_alternative->removeLastTerm();
michael@0:             return;
michael@0:         }
michael@0: 
michael@0:         PatternTerm& term = m_alternative->lastTerm();
michael@0:         ASSERT(term.type > PatternTerm::TypeAssertionWordBoundary);
michael@0:         ASSERT((term.quantityCount == 1) && (term.quantityType == QuantifierFixedCount));
michael@0: 
michael@0:         if (term.type == PatternTerm::TypeParentheticalAssertion) {
michael@0:             // If an assertion is quantified with a minimum count of zero, it can simply be removed.
michael@0:             // This arises from the RepeatMatcher behaviour in the spec. Matching an assertion never
michael@0:             // results in any input being consumed, however the continuation passed to the assertion
michael@0:             // (called in steps, 8c and 9 of the RepeatMatcher definition, ES5.1 15.10.2.5) will
michael@0:             // reject all zero length matches (see step 2.1). A match from the continuation of the
michael@0:             // expression will still be accepted regardless (via steps 8a and 11) - the upshot of all
michael@0:             // this is that matches from the assertion are not required, and won't be accepted anyway,
michael@0:             // so no need to ever run it.
michael@0:             if (!min)
michael@0:                 m_alternative->removeLastTerm();
michael@0:             // We never need to run an assertion more than once. Subsequent interations will be run
michael@0:             // with the same start index (since assertions are non-capturing) and the same captures
michael@0:             // (per step 4 of RepeatMatcher in ES5.1 15.10.2.5), and as such will always produce the
michael@0:             // same result and captures. If the first match succeeds then the subsequent (min - 1)
michael@0:             // matches will too. Any additional optional matches will fail (on the same basis as the
michael@0:             // minimum zero quantified assertions, above), but this will still result in a match.
michael@0:             return;
michael@0:         }
michael@0: 
michael@0:         if (min == 0)
michael@0:             term.quantify(max, greedy   ? QuantifierGreedy : QuantifierNonGreedy);
michael@0:         else if (min == max)
michael@0:             term.quantify(min, QuantifierFixedCount);
michael@0:         else {
michael@0:             term.quantify(min, QuantifierFixedCount);
michael@0:             m_alternative->m_terms.append(copyTerm(term));
michael@0:             // NOTE: this term is interesting from an analysis perspective, in that it can be ignored.....
michael@0:             m_alternative->lastTerm().quantify((max == quantifyInfinite) ? max : max - min, greedy ? QuantifierGreedy : QuantifierNonGreedy);
michael@0:             if (m_alternative->lastTerm().type == PatternTerm::TypeParenthesesSubpattern)
michael@0:                 m_alternative->lastTerm().parentheses.isCopy = true;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     void disjunction()
michael@0:     {
michael@0:         m_alternative = m_alternative->m_parent->addNewAlternative();
michael@0:     }
michael@0: 
michael@0:     ErrorCode setupAlternativeOffsets(PatternAlternative* alternative, unsigned currentCallFrameSize, unsigned initialInputPosition,
michael@0:                                       unsigned *callFrameSizeOut)
michael@0:     {
michael@0:         /*
michael@0:          * Attempt detection of over-recursion:
michael@0:          * "1MB should be enough stack for anyone."
michael@0:          */
michael@0:         uint8_t stackDummy_;
michael@0:         if (m_stackBase - &stackDummy_ > 1024*1024)
michael@0:             return PatternTooLarge;
michael@0: 
michael@0:         alternative->m_hasFixedSize = true;
michael@0:         Checked<unsigned> currentInputPosition = initialInputPosition;
michael@0: 
michael@0:         for (unsigned i = 0; i < alternative->m_terms.size(); ++i) {
michael@0:             PatternTerm& term = alternative->m_terms[i];
michael@0: 
michael@0:             switch (term.type) {
michael@0:             case PatternTerm::TypeAssertionBOL:
michael@0:             case PatternTerm::TypeAssertionEOL:
michael@0:             case PatternTerm::TypeAssertionWordBoundary:
michael@0:                 if (Checked<int>(currentInputPosition).safeGet(term.inputPosition))
michael@0:                     return RuntimeError;
michael@0:                 break;
michael@0: 
michael@0:             case PatternTerm::TypeBackReference:
michael@0:                 if (Checked<int>(currentInputPosition).safeGet(term.inputPosition))
michael@0:                     return RuntimeError;
michael@0:                 term.frameLocation = currentCallFrameSize;
michael@0:                 currentCallFrameSize += YarrStackSpaceForBackTrackInfoBackReference;
michael@0:                 alternative->m_hasFixedSize = false;
michael@0:                 break;
michael@0: 
michael@0:             case PatternTerm::TypeForwardReference:
michael@0:                 break;
michael@0: 
michael@0:             case PatternTerm::TypePatternCharacter:
michael@0:                 if (Checked<int>(currentInputPosition).safeGet(term.inputPosition))
michael@0:                     return RuntimeError;
michael@0:                 if (term.quantityType != QuantifierFixedCount) {
michael@0:                     term.frameLocation = currentCallFrameSize;
michael@0:                     currentCallFrameSize += YarrStackSpaceForBackTrackInfoPatternCharacter;
michael@0:                     alternative->m_hasFixedSize = false;
michael@0:                 } else
michael@0:                     currentInputPosition += term.quantityCount;
michael@0:                 break;
michael@0: 
michael@0:             case PatternTerm::TypeCharacterClass:
michael@0:                 if (Checked<int>(currentInputPosition).safeGet(term.inputPosition))
michael@0:                     return RuntimeError;
michael@0:                 if (term.quantityType != QuantifierFixedCount) {
michael@0:                     term.frameLocation = currentCallFrameSize;
michael@0:                     currentCallFrameSize += YarrStackSpaceForBackTrackInfoCharacterClass;
michael@0:                     alternative->m_hasFixedSize = false;
michael@0:                 } else
michael@0:                     currentInputPosition += term.quantityCount;
michael@0:                 break;
michael@0: 
michael@0:             case PatternTerm::TypeParenthesesSubpattern:
michael@0:                 // Note: for fixed once parentheses we will ensure at least the minimum is available; others are on their own.
michael@0:                 term.frameLocation = currentCallFrameSize;
michael@0:                 unsigned position;
michael@0:                 if (currentInputPosition.safeGet(position))
michael@0:                     return RuntimeError;
michael@0:                 if (term.quantityCount == 1 && !term.parentheses.isCopy) {
michael@0:                     if (term.quantityType != QuantifierFixedCount)
michael@0:                         currentCallFrameSize += YarrStackSpaceForBackTrackInfoParenthesesOnce;
michael@0:                     if (ErrorCode error = setupDisjunctionOffsets(term.parentheses.disjunction, currentCallFrameSize, position, &currentCallFrameSize))
michael@0:                         return error;
michael@0:                     // If quantity is fixed, then pre-check its minimum size.
michael@0:                     if (term.quantityType == QuantifierFixedCount)
michael@0:                         currentInputPosition += term.parentheses.disjunction->m_minimumSize;
michael@0:                     if (Checked<int>(currentInputPosition).safeGet(term.inputPosition))
michael@0:                         return RuntimeError;
michael@0:                 } else if (term.parentheses.isTerminal) {
michael@0:                     currentCallFrameSize += YarrStackSpaceForBackTrackInfoParenthesesTerminal;
michael@0:                     if (ErrorCode error = setupDisjunctionOffsets(term.parentheses.disjunction, currentCallFrameSize, position, &currentCallFrameSize))
michael@0:                         return error;
michael@0:                     if (Checked<int>(currentInputPosition).safeGet(term.inputPosition))
michael@0:                         return RuntimeError;
michael@0:                 } else {
michael@0:                     if (Checked<int>(currentInputPosition).safeGet(term.inputPosition))
michael@0:                         return RuntimeError;
michael@0:                     unsigned dummy;
michael@0:                     if (ErrorCode error = setupDisjunctionOffsets(term.parentheses.disjunction, BASE_FRAME_SIZE, position, &dummy))
michael@0:                         return error;
michael@0:                     currentCallFrameSize += YarrStackSpaceForBackTrackInfoParentheses;
michael@0:                 }
michael@0:                 // Fixed count of 1 could be accepted, if they have a fixed size *AND* if all alternatives are of the same length.
michael@0:                 alternative->m_hasFixedSize = false;
michael@0:                 break;
michael@0: 
michael@0:             case PatternTerm::TypeParentheticalAssertion:
michael@0:                 if (Checked<int>(currentInputPosition).safeGet(term.inputPosition))
michael@0:                     return RuntimeError;
michael@0:                 term.frameLocation = currentCallFrameSize;
michael@0:                 if (ErrorCode error = setupDisjunctionOffsets(term.parentheses.disjunction, currentCallFrameSize + YarrStackSpaceForBackTrackInfoParentheticalAssertion, term.inputPosition, &currentCallFrameSize))
michael@0:                     return error;
michael@0:                 break;
michael@0: 
michael@0:             case PatternTerm::TypeDotStarEnclosure:
michael@0:                 alternative->m_hasFixedSize = false;
michael@0:                 term.inputPosition = initialInputPosition;
michael@0:                 break;
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         if ((currentInputPosition - initialInputPosition).safeGet(alternative->m_minimumSize))
michael@0:             return RuntimeError;
michael@0:         *callFrameSizeOut = currentCallFrameSize;
michael@0:         return NoError;
michael@0:     }
michael@0: 
michael@0:     ErrorCode setupDisjunctionOffsets(PatternDisjunction* disjunction, unsigned initialCallFrameSize, unsigned initialInputPosition, unsigned *maximumCallFrameSizeOut)
michael@0:     {
michael@0:         if ((disjunction != m_pattern.m_body) && (disjunction->m_alternatives.size() > 1))
michael@0:             initialCallFrameSize += YarrStackSpaceForBackTrackInfoAlternative;
michael@0: 
michael@0:         unsigned minimumInputSize = UINT_MAX;
michael@0:         unsigned maximumCallFrameSize = 0;
michael@0:         bool hasFixedSize = true;
michael@0: 
michael@0:         for (unsigned alt = 0; alt < disjunction->m_alternatives.size(); ++alt) {
michael@0:             PatternAlternative* alternative = disjunction->m_alternatives[alt];
michael@0:             unsigned currentAlternativeCallFrameSize;
michael@0:             if (ErrorCode error = setupAlternativeOffsets(alternative, initialCallFrameSize, initialInputPosition, &currentAlternativeCallFrameSize))
michael@0:                 return error;
michael@0:             minimumInputSize = std::min(minimumInputSize, alternative->m_minimumSize);
michael@0:             maximumCallFrameSize = std::max(maximumCallFrameSize, currentAlternativeCallFrameSize);
michael@0:             hasFixedSize &= alternative->m_hasFixedSize;
michael@0:         }
michael@0: 
michael@0:         ASSERT(minimumInputSize != UINT_MAX);
michael@0:         if (minimumInputSize == UINT_MAX)
michael@0:             return PatternTooLarge;
michael@0: 
michael@0:         ASSERT(minimumInputSize != UINT_MAX);
michael@0:         ASSERT(maximumCallFrameSize >= initialCallFrameSize);
michael@0: 
michael@0:         disjunction->m_hasFixedSize = hasFixedSize;
michael@0:         disjunction->m_minimumSize = minimumInputSize;
michael@0:         disjunction->m_callFrameSize = maximumCallFrameSize;
michael@0:         *maximumCallFrameSizeOut = maximumCallFrameSize;
michael@0:         return NoError;
michael@0:     }
michael@0: 
michael@0:     ErrorCode setupOffsets()
michael@0:     {
michael@0:         unsigned dummy;
michael@0:         return setupDisjunctionOffsets(m_pattern.m_body, BASE_FRAME_SIZE, 0, &dummy);
michael@0:     }
michael@0: 
michael@0:     // This optimization identifies sets of parentheses that we will never need to backtrack.
michael@0:     // In these cases we do not need to store state from prior iterations.
michael@0:     // We can presently avoid backtracking for:
michael@0:     //   * where the parens are at the end of the regular expression (last term in any of the
michael@0:     //     alternatives of the main body disjunction).
michael@0:     //   * where the parens are non-capturing, and quantified unbounded greedy (*).
michael@0:     //   * where the parens do not contain any capturing subpatterns.
michael@0:     void checkForTerminalParentheses()
michael@0:     {
michael@0:         // This check is much too crude; should be just checking whether the candidate
michael@0:         // node contains nested capturing subpatterns, not the whole expression!
michael@0:         if (m_pattern.m_numSubpatterns)
michael@0:             return;
michael@0: 
michael@0:         Vector<PatternAlternative*>& alternatives = m_pattern.m_body->m_alternatives;
michael@0:         for (size_t i = 0; i < alternatives.size(); ++i) {
michael@0:             Vector<PatternTerm>& terms = alternatives[i]->m_terms;
michael@0:             if (terms.size()) {
michael@0:                 PatternTerm& term = terms.last();
michael@0:                 if (term.type == PatternTerm::TypeParenthesesSubpattern
michael@0:                     && term.quantityType == QuantifierGreedy
michael@0:                     && term.quantityCount == quantifyInfinite
michael@0:                     && !term.capture())
michael@0:                     term.parentheses.isTerminal = true;
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     void optimizeBOL()
michael@0:     {
michael@0:         // Look for expressions containing beginning of line (^) anchoring and unroll them.
michael@0:         // e.g. /^a|^b|c/ becomes /^a|^b|c/ which is executed once followed by /c/ which loops
michael@0:         // This code relies on the parsing code tagging alternatives with m_containsBOL and
michael@0:         // m_startsWithBOL and rolling those up to containing alternatives.
michael@0:         // At this point, this is only valid for non-multiline expressions.
michael@0:         PatternDisjunction* disjunction = m_pattern.m_body;
michael@0: 
michael@0:         if (!m_pattern.m_containsBOL || m_pattern.m_multiline)
michael@0:             return;
michael@0: 
michael@0:         PatternDisjunction* loopDisjunction = copyDisjunction(disjunction, true);
michael@0: 
michael@0:         // Set alternatives in disjunction to "onceThrough"
michael@0:         for (unsigned alt = 0; alt < disjunction->m_alternatives.size(); ++alt)
michael@0:             disjunction->m_alternatives[alt]->setOnceThrough();
michael@0: 
michael@0:         if (loopDisjunction) {
michael@0:             // Move alternatives from loopDisjunction to disjunction
michael@0:             for (unsigned alt = 0; alt < loopDisjunction->m_alternatives.size(); ++alt)
michael@0:                 disjunction->m_alternatives.append(loopDisjunction->m_alternatives[alt]);
michael@0: 
michael@0:             loopDisjunction->m_alternatives.clear();
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     bool containsCapturingTerms(PatternAlternative* alternative, size_t firstTermIndex, size_t lastTermIndex)
michael@0:     {
michael@0:         Vector<PatternTerm>& terms = alternative->m_terms;
michael@0: 
michael@0:         for (size_t termIndex = firstTermIndex; termIndex <= lastTermIndex; ++termIndex) {
michael@0:             PatternTerm& term = terms[termIndex];
michael@0: 
michael@0:             if (term.m_capture)
michael@0:                 return true;
michael@0: 
michael@0:             if (term.type == PatternTerm::TypeParenthesesSubpattern) {
michael@0:                 PatternDisjunction* nestedDisjunction = term.parentheses.disjunction;
michael@0:                 for (unsigned alt = 0; alt < nestedDisjunction->m_alternatives.size(); ++alt) {
michael@0:                     PatternAlternative *pattern = nestedDisjunction->m_alternatives[alt];
michael@0:                     if (pattern->m_terms.size() == 0)
michael@0:                         continue;
michael@0:                     if (containsCapturingTerms(pattern, 0, pattern->m_terms.size() - 1))
michael@0:                         return true;
michael@0:                 }
michael@0:             }
michael@0:         }
michael@0: 
michael@0:         return false;
michael@0:     }
michael@0: 
michael@0:     // This optimization identifies alternatives in the form of
michael@0:     // [^].*[?]<expression>.*[$] for expressions that don't have any
michael@0:     // capturing terms. The alternative is changed to <expression>
michael@0:     // followed by processing of the dot stars to find and adjust the
michael@0:     // beginning and the end of the match.
michael@0:     void optimizeDotStarWrappedExpressions()
michael@0:     {
michael@0:         Vector<PatternAlternative*>& alternatives = m_pattern.m_body->m_alternatives;
michael@0:         if (alternatives.size() != 1)
michael@0:             return;
michael@0: 
michael@0:         PatternAlternative* alternative = alternatives[0];
michael@0:         Vector<PatternTerm>& terms = alternative->m_terms;
michael@0:         if (terms.size() >= 3) {
michael@0:             bool startsWithBOL = false;
michael@0:             bool endsWithEOL = false;
michael@0:             size_t termIndex, firstExpressionTerm, lastExpressionTerm;
michael@0: 
michael@0:             termIndex = 0;
michael@0:             if (terms[termIndex].type == PatternTerm::TypeAssertionBOL) {
michael@0:                 startsWithBOL = true;
michael@0:                 ++termIndex;
michael@0:             }
michael@0: 
michael@0:             PatternTerm& firstNonAnchorTerm = terms[termIndex];
michael@0:             if ((firstNonAnchorTerm.type != PatternTerm::TypeCharacterClass) || (firstNonAnchorTerm.characterClass != m_pattern.newlineCharacterClass()) || !((firstNonAnchorTerm.quantityType == QuantifierGreedy) || (firstNonAnchorTerm.quantityType == QuantifierNonGreedy)))
michael@0:                 return;
michael@0: 
michael@0:             firstExpressionTerm = termIndex + 1;
michael@0: 
michael@0:             termIndex = terms.size() - 1;
michael@0:             if (terms[termIndex].type == PatternTerm::TypeAssertionEOL) {
michael@0:                 endsWithEOL = true;
michael@0:                 --termIndex;
michael@0:             }
michael@0: 
michael@0:             PatternTerm& lastNonAnchorTerm = terms[termIndex];
michael@0:             if ((lastNonAnchorTerm.type != PatternTerm::TypeCharacterClass) || (lastNonAnchorTerm.characterClass != m_pattern.newlineCharacterClass()) || (lastNonAnchorTerm.quantityType != QuantifierGreedy))
michael@0:                 return;
michael@0: 
michael@0:             lastExpressionTerm = termIndex - 1;
michael@0: 
michael@0:             if (firstExpressionTerm > lastExpressionTerm)
michael@0:                 return;
michael@0: 
michael@0:             if (!containsCapturingTerms(alternative, firstExpressionTerm, lastExpressionTerm)) {
michael@0:                 for (termIndex = terms.size() - 1; termIndex > lastExpressionTerm; --termIndex)
michael@0:                     terms.remove(termIndex);
michael@0: 
michael@0:                 for (termIndex = firstExpressionTerm; termIndex > 0; --termIndex)
michael@0:                     terms.remove(termIndex - 1);
michael@0: 
michael@0:                 terms.append(PatternTerm(startsWithBOL, endsWithEOL));
michael@0: 
michael@0:                 m_pattern.m_containsBOL = false;
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     void setStackBase(uint8_t *stackBase) {
michael@0:         m_stackBase = stackBase;
michael@0:     }
michael@0: 
michael@0: private:
michael@0:     YarrPattern& m_pattern;
michael@0:     uint8_t * m_stackBase;
michael@0:     PatternAlternative* m_alternative;
michael@0:     CharacterClassConstructor m_characterClassConstructor;
michael@0:     bool m_invertCharacterClass;
michael@0:     bool m_invertParentheticalAssertion;
michael@0: };
michael@0: 
michael@0: ErrorCode YarrPattern::compile(const String& patternString)
michael@0: {
michael@0:     YarrPatternConstructor constructor(*this);
michael@0: 
michael@0:     if (ErrorCode error = parse(constructor, patternString))
michael@0:         return error;
michael@0: 
michael@0:     // If the pattern contains illegal backreferences reset & reparse.
michael@0:     // Quoting Netscape's "What's new in JavaScript 1.2",
michael@0:     //      "Note: if the number of left parentheses is less than the number specified
michael@0:     //       in \#, the \# is taken as an octal escape as described in the next row."
michael@0:     if (containsIllegalBackReference()) {
michael@0:         unsigned numSubpatterns = m_numSubpatterns;
michael@0: 
michael@0:         constructor.reset();
michael@0: #if !ASSERT_DISABLED
michael@0:         ErrorCode error =
michael@0: #endif
michael@0:             parse(constructor, patternString, numSubpatterns);
michael@0: 
michael@0:         ASSERT(!error);
michael@0:         ASSERT(numSubpatterns == m_numSubpatterns);
michael@0:     }
michael@0: 
michael@0:     uint8_t stackDummy_;
michael@0:     constructor.setStackBase(&stackDummy_);
michael@0: 
michael@0:     constructor.checkForTerminalParentheses();
michael@0:     constructor.optimizeDotStarWrappedExpressions();
michael@0:     constructor.optimizeBOL();
michael@0: 
michael@0:     if (ErrorCode error = constructor.setupOffsets())
michael@0:         return error;
michael@0: 
michael@0:     return NoError;
michael@0: }
michael@0: 
michael@0: YarrPattern::YarrPattern(const String& pattern, bool ignoreCase, bool multiline, ErrorCode* error)
michael@0:     : m_ignoreCase(ignoreCase)
michael@0:     , m_multiline(multiline)
michael@0:     , m_containsBackreferences(false)
michael@0:     , m_containsBOL(false)
michael@0:     , m_numSubpatterns(0)
michael@0:     , m_maxBackReference(0)
michael@0:     , newlineCached(0)
michael@0:     , digitsCached(0)
michael@0:     , spacesCached(0)
michael@0:     , wordcharCached(0)
michael@0:     , nondigitsCached(0)
michael@0:     , nonspacesCached(0)
michael@0:     , nonwordcharCached(0)
michael@0: {
michael@0:     *error = compile(pattern);
michael@0: }
michael@0: 
michael@0: } }