// // file: regexcmp.cpp // // Copyright (C) 2002-2013 International Business Machines Corporation and others. // All Rights Reserved. // // This file contains the ICU regular expression compiler, which is responsible // for processing a regular expression pattern into the compiled form that // is used by the match finding engine. // #include "unicode/utypes.h" #if !UCONFIG_NO_REGULAR_EXPRESSIONS #include "unicode/ustring.h" #include "unicode/unistr.h" #include "unicode/uniset.h" #include "unicode/uchar.h" #include "unicode/uchriter.h" #include "unicode/parsepos.h" #include "unicode/parseerr.h" #include "unicode/regex.h" #include "unicode/utf.h" #include "unicode/utf16.h" #include "patternprops.h" #include "putilimp.h" #include "cmemory.h" #include "cstring.h" #include "uvectr32.h" #include "uvectr64.h" #include "uassert.h" #include "ucln_in.h" #include "uinvchar.h" #include "regeximp.h" #include "regexcst.h" // Contains state table for the regex pattern parser. // generated by a Perl script. #include "regexcmp.h" #include "regexst.h" #include "regextxt.h" U_NAMESPACE_BEGIN //------------------------------------------------------------------------------ // // Constructor. // //------------------------------------------------------------------------------ RegexCompile::RegexCompile(RegexPattern *rxp, UErrorCode &status) : fParenStack(status), fSetStack(status), fSetOpStack(status) { // Lazy init of all shared global sets (needed for init()'s empty text) RegexStaticSets::initGlobals(&status); fStatus = &status; fRXPat = rxp; fScanIndex = 0; fLastChar = -1; fPeekChar = -1; fLineNum = 1; fCharNum = 0; fQuoteMode = FALSE; fInBackslashQuote = FALSE; fModeFlags = fRXPat->fFlags | 0x80000000; fEOLComments = TRUE; fMatchOpenParen = -1; fMatchCloseParen = -1; if (U_SUCCESS(status) && U_FAILURE(rxp->fDeferredStatus)) { status = rxp->fDeferredStatus; } } static const UChar chAmp = 0x26; // '&' static const UChar chDash = 0x2d; // '-' //------------------------------------------------------------------------------ // // Destructor // //------------------------------------------------------------------------------ RegexCompile::~RegexCompile() { } static inline void addCategory(UnicodeSet *set, int32_t value, UErrorCode& ec) { set->addAll(UnicodeSet().applyIntPropertyValue(UCHAR_GENERAL_CATEGORY_MASK, value, ec)); } //------------------------------------------------------------------------------ // // Compile regex pattern. The state machine for rexexp pattern parsing is here. // The state tables are hand-written in the file regexcst.txt, // and converted to the form used here by a perl // script regexcst.pl // //------------------------------------------------------------------------------ void RegexCompile::compile( const UnicodeString &pat, // Source pat to be compiled. UParseError &pp, // Error position info UErrorCode &e) // Error Code { fRXPat->fPatternString = new UnicodeString(pat); UText patternText = UTEXT_INITIALIZER; utext_openConstUnicodeString(&patternText, fRXPat->fPatternString, &e); if (U_SUCCESS(e)) { compile(&patternText, pp, e); utext_close(&patternText); } } // // compile, UText mode // All the work is actually done here. // void RegexCompile::compile( UText *pat, // Source pat to be compiled. UParseError &pp, // Error position info UErrorCode &e) // Error Code { fStatus = &e; fParseErr = &pp; fStackPtr = 0; fStack[fStackPtr] = 0; if (U_FAILURE(*fStatus)) { return; } // There should be no pattern stuff in the RegexPattern object. They can not be reused. U_ASSERT(fRXPat->fPattern == NULL || utext_nativeLength(fRXPat->fPattern) == 0); // Prepare the RegexPattern object to receive the compiled pattern. fRXPat->fPattern = utext_clone(fRXPat->fPattern, pat, FALSE, TRUE, fStatus); fRXPat->fStaticSets = RegexStaticSets::gStaticSets->fPropSets; fRXPat->fStaticSets8 = RegexStaticSets::gStaticSets->fPropSets8; // Initialize the pattern scanning state machine fPatternLength = utext_nativeLength(pat); uint16_t state = 1; const RegexTableEl *tableEl; // UREGEX_LITERAL force entire pattern to be treated as a literal string. if (fModeFlags & UREGEX_LITERAL) { fQuoteMode = TRUE; } nextChar(fC); // Fetch the first char from the pattern string. // // Main loop for the regex pattern parsing state machine. // Runs once per state transition. // Each time through optionally performs, depending on the state table, // - an advance to the the next pattern char // - an action to be performed. // - pushing or popping a state to/from the local state return stack. // file regexcst.txt is the source for the state table. The logic behind // recongizing the pattern syntax is there, not here. // for (;;) { // Bail out if anything has gone wrong. // Regex pattern parsing stops on the first error encountered. if (U_FAILURE(*fStatus)) { break; } U_ASSERT(state != 0); // Find the state table element that matches the input char from the pattern, or the // class of the input character. Start with the first table row for this // state, then linearly scan forward until we find a row that matches the // character. The last row for each state always matches all characters, so // the search will stop there, if not before. // tableEl = &gRuleParseStateTable[state]; REGEX_SCAN_DEBUG_PRINTF(("char, line, col = (\'%c\', %d, %d) state=%s ", fC.fChar, fLineNum, fCharNum, RegexStateNames[state])); for (;;) { // loop through table rows belonging to this state, looking for one // that matches the current input char. REGEX_SCAN_DEBUG_PRINTF((".")); if (tableEl->fCharClass < 127 && fC.fQuoted == FALSE && tableEl->fCharClass == fC.fChar) { // Table row specified an individual character, not a set, and // the input character is not quoted, and // the input character matched it. break; } if (tableEl->fCharClass == 255) { // Table row specified default, match anything character class. break; } if (tableEl->fCharClass == 254 && fC.fQuoted) { // Table row specified "quoted" and the char was quoted. break; } if (tableEl->fCharClass == 253 && fC.fChar == (UChar32)-1) { // Table row specified eof and we hit eof on the input. break; } if (tableEl->fCharClass >= 128 && tableEl->fCharClass < 240 && // Table specs a char class && fC.fQuoted == FALSE && // char is not escaped && fC.fChar != (UChar32)-1) { // char is not EOF U_ASSERT(tableEl->fCharClass <= 137); if (RegexStaticSets::gStaticSets->fRuleSets[tableEl->fCharClass-128].contains(fC.fChar)) { // Table row specified a character class, or set of characters, // and the current char matches it. break; } } // No match on this row, advance to the next row for this state, tableEl++; } REGEX_SCAN_DEBUG_PRINTF(("\n")); // // We've found the row of the state table that matches the current input // character from the rules string. // Perform any action specified by this row in the state table. if (doParseActions(tableEl->fAction) == FALSE) { // Break out of the state machine loop if the // the action signalled some kind of error, or // the action was to exit, occurs on normal end-of-rules-input. break; } if (tableEl->fPushState != 0) { fStackPtr++; if (fStackPtr >= kStackSize) { error(U_REGEX_INTERNAL_ERROR); REGEX_SCAN_DEBUG_PRINTF(("RegexCompile::parse() - state stack overflow.\n")); fStackPtr--; } fStack[fStackPtr] = tableEl->fPushState; } // // NextChar. This is where characters are actually fetched from the pattern. // Happens under control of the 'n' tag in the state table. // if (tableEl->fNextChar) { nextChar(fC); } // Get the next state from the table entry, or from the // state stack if the next state was specified as "pop". if (tableEl->fNextState != 255) { state = tableEl->fNextState; } else { state = fStack[fStackPtr]; fStackPtr--; if (fStackPtr < 0) { // state stack underflow // This will occur if the user pattern has mis-matched parentheses, // with extra close parens. // fStackPtr++; error(U_REGEX_MISMATCHED_PAREN); } } } if (U_FAILURE(*fStatus)) { // Bail out if the pattern had errors. // Set stack cleanup: a successful compile would have left it empty, // but errors can leave temporary sets hanging around. while (!fSetStack.empty()) { delete (UnicodeSet *)fSetStack.pop(); } return; } // // The pattern has now been read and processed, and the compiled code generated. // // // Compute the number of digits requried for the largest capture group number. // fRXPat->fMaxCaptureDigits = 1; int32_t n = 10; int32_t groupCount = fRXPat->fGroupMap->size(); while (n <= groupCount) { fRXPat->fMaxCaptureDigits++; n *= 10; } // // The pattern's fFrameSize so far has accumulated the requirements for // storage for capture parentheses, counters, etc. that are encountered // in the pattern. Add space for the two variables that are always // present in the saved state: the input string position (int64_t) and // the position in the compiled pattern. // fRXPat->fFrameSize+=RESTACKFRAME_HDRCOUNT; // // Optimization pass 1: NOPs, back-references, and case-folding // stripNOPs(); // // Get bounds for the minimum and maximum length of a string that this // pattern can match. Used to avoid looking for matches in strings that // are too short. // fRXPat->fMinMatchLen = minMatchLength(3, fRXPat->fCompiledPat->size()-1); // // Optimization pass 2: match start type // matchStartType(); // // Set up fast latin-1 range sets // int32_t numSets = fRXPat->fSets->size(); fRXPat->fSets8 = new Regex8BitSet[numSets]; // Null pointer check. if (fRXPat->fSets8 == NULL) { e = *fStatus = U_MEMORY_ALLOCATION_ERROR; return; } int32_t i; for (i=0; ifSets->elementAt(i); fRXPat->fSets8[i].init(s); } } //------------------------------------------------------------------------------ // // doParseAction Do some action during regex pattern parsing. // Called by the parse state machine. // // Generation of the match engine PCode happens here, or // in functions called from the parse actions defined here. // // //------------------------------------------------------------------------------ UBool RegexCompile::doParseActions(int32_t action) { UBool returnVal = TRUE; switch ((Regex_PatternParseAction)action) { case doPatStart: // Start of pattern compiles to: //0 SAVE 2 Fall back to position of FAIL //1 jmp 3 //2 FAIL Stop if we ever reach here. //3 NOP Dummy, so start of pattern looks the same as // the start of an ( grouping. //4 NOP Resreved, will be replaced by a save if there are // OR | operators at the top level fRXPat->fCompiledPat->addElement(URX_BUILD(URX_STATE_SAVE, 2), *fStatus); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_JMP, 3), *fStatus); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_FAIL, 0), *fStatus); // Standard open nonCapture paren action emits the two NOPs and // sets up the paren stack frame. doParseActions(doOpenNonCaptureParen); break; case doPatFinish: // We've scanned to the end of the pattern // The end of pattern compiles to: // URX_END // which will stop the runtime match engine. // Encountering end of pattern also behaves like a close paren, // and forces fixups of the State Save at the beginning of the compiled pattern // and of any OR operations at the top level. // handleCloseParen(); if (fParenStack.size() > 0) { // Missing close paren in pattern. error(U_REGEX_MISMATCHED_PAREN); } // add the END operation to the compiled pattern. fRXPat->fCompiledPat->addElement(URX_BUILD(URX_END, 0), *fStatus); // Terminate the pattern compilation state machine. returnVal = FALSE; break; case doOrOperator: // Scanning a '|', as in (A|B) { // Generate code for any pending literals preceding the '|' fixLiterals(FALSE); // Insert a SAVE operation at the start of the pattern section preceding // this OR at this level. This SAVE will branch the match forward // to the right hand side of the OR in the event that the left hand // side fails to match and backtracks. Locate the position for the // save from the location on the top of the parentheses stack. int32_t savePosition = fParenStack.popi(); int32_t op = (int32_t)fRXPat->fCompiledPat->elementAti(savePosition); U_ASSERT(URX_TYPE(op) == URX_NOP); // original contents of reserved location op = URX_BUILD(URX_STATE_SAVE, fRXPat->fCompiledPat->size()+1); fRXPat->fCompiledPat->setElementAt(op, savePosition); // Append an JMP operation into the compiled pattern. The operand for // the JMP will eventually be the location following the ')' for the // group. This will be patched in later, when the ')' is encountered. op = URX_BUILD(URX_JMP, 0); fRXPat->fCompiledPat->addElement(op, *fStatus); // Push the position of the newly added JMP op onto the parentheses stack. // This registers if for fixup when this block's close paren is encountered. fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // Append a NOP to the compiled pattern. This is the slot reserved // for a SAVE in the event that there is yet another '|' following // this one. fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); } break; case doOpenCaptureParen: // Open Paren. // Compile to a // - NOP, which later may be replaced by a save-state if the // parenthesized group gets a * quantifier, followed by // - START_CAPTURE n where n is stack frame offset to the capture group variables. // - NOP, which may later be replaced by a save-state if there // is an '|' alternation within the parens. // // Each capture group gets three slots in the save stack frame: // 0: Capture Group start position (in input string being matched.) // 1: Capture Group end position. // 2: Start of Match-in-progress. // The first two locations are for a completed capture group, and are // referred to by back references and the like. // The third location stores the capture start position when an START_CAPTURE is // encountered. This will be promoted to a completed capture when (and if) the corresponding // END_CAPTURE is encountered. { fixLiterals(); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); int32_t varsLoc = fRXPat->fFrameSize; // Reserve three slots in match stack frame. fRXPat->fFrameSize += 3; int32_t cop = URX_BUILD(URX_START_CAPTURE, varsLoc); fRXPat->fCompiledPat->addElement(cop, *fStatus); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); // On the Parentheses stack, start a new frame and add the postions // of the two NOPs. Depending on what follows in the pattern, the // NOPs may be changed to SAVE_STATE or JMP ops, with a target // address of the end of the parenthesized group. fParenStack.push(fModeFlags, *fStatus); // Match mode state fParenStack.push(capturing, *fStatus); // Frame type. fParenStack.push(fRXPat->fCompiledPat->size()-3, *fStatus); // The first NOP location fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // The second NOP loc // Save the mapping from group number to stack frame variable position. fRXPat->fGroupMap->addElement(varsLoc, *fStatus); } break; case doOpenNonCaptureParen: // Open non-caputuring (grouping only) Paren. // Compile to a // - NOP, which later may be replaced by a save-state if the // parenthesized group gets a * quantifier, followed by // - NOP, which may later be replaced by a save-state if there // is an '|' alternation within the parens. { fixLiterals(); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); // On the Parentheses stack, start a new frame and add the postions // of the two NOPs. fParenStack.push(fModeFlags, *fStatus); // Match mode state fParenStack.push(plain, *fStatus); // Begin a new frame. fParenStack.push(fRXPat->fCompiledPat->size()-2, *fStatus); // The first NOP location fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // The second NOP loc } break; case doOpenAtomicParen: // Open Atomic Paren. (?> // Compile to a // - NOP, which later may be replaced if the parenthesized group // has a quantifier, followed by // - STO_SP save state stack position, so it can be restored at the ")" // - NOP, which may later be replaced by a save-state if there // is an '|' alternation within the parens. { fixLiterals(); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); int32_t varLoc = fRXPat->fDataSize; // Reserve a data location for saving the fRXPat->fDataSize += 1; // state stack ptr. int32_t stoOp = URX_BUILD(URX_STO_SP, varLoc); fRXPat->fCompiledPat->addElement(stoOp, *fStatus); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); // On the Parentheses stack, start a new frame and add the postions // of the two NOPs. Depending on what follows in the pattern, the // NOPs may be changed to SAVE_STATE or JMP ops, with a target // address of the end of the parenthesized group. fParenStack.push(fModeFlags, *fStatus); // Match mode state fParenStack.push(atomic, *fStatus); // Frame type. fParenStack.push(fRXPat->fCompiledPat->size()-3, *fStatus); // The first NOP fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // The second NOP } break; case doOpenLookAhead: // Positive Look-ahead (?= stuff ) // // Note: Addition of transparent input regions, with the need to // restore the original regions when failing out of a lookahead // block, complicated this sequence. Some conbined opcodes // might make sense - or might not, lookahead aren't that common. // // Caution: min match length optimization knows about this // sequence; don't change without making updates there too. // // Compiles to // 1 START_LA dataLoc Saves SP, Input Pos // 2. STATE_SAVE 4 on failure of lookahead, goto 4 // 3 JMP 6 continue ... // // 4. LA_END Look Ahead failed. Restore regions. // 5. BACKTRACK and back track again. // // 6. NOP reserved for use by quantifiers on the block. // Look-ahead can't have quantifiers, but paren stack // compile time conventions require the slot anyhow. // 7. NOP may be replaced if there is are '|' ops in the block. // 8. code for parenthesized stuff. // 9. LA_END // // Two data slots are reserved, for saving the stack ptr and the input position. { fixLiterals(); int32_t dataLoc = fRXPat->fDataSize; fRXPat->fDataSize += 2; int32_t op = URX_BUILD(URX_LA_START, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_STATE_SAVE, fRXPat->fCompiledPat->size()+ 2); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_JMP, fRXPat->fCompiledPat->size()+ 3); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_LA_END, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_BACKTRACK, 0); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_NOP, 0); fRXPat->fCompiledPat->addElement(op, *fStatus); fRXPat->fCompiledPat->addElement(op, *fStatus); // On the Parentheses stack, start a new frame and add the postions // of the NOPs. fParenStack.push(fModeFlags, *fStatus); // Match mode state fParenStack.push(lookAhead, *fStatus); // Frame type. fParenStack.push(fRXPat->fCompiledPat->size()-2, *fStatus); // The first NOP location fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // The second NOP location } break; case doOpenLookAheadNeg: // Negated Lookahead. (?! stuff ) // Compiles to // 1. START_LA dataloc // 2. SAVE_STATE 7 // Fail within look-ahead block restores to this state, // // which continues with the match. // 3. NOP // Std. Open Paren sequence, for possible '|' // 4. code for parenthesized stuff. // 5. END_LA // Cut back stack, remove saved state from step 2. // 6. BACKTRACK // code in block succeeded, so neg. lookahead fails. // 7. END_LA // Restore match region, in case look-ahead was using // an alternate (transparent) region. { fixLiterals(); int32_t dataLoc = fRXPat->fDataSize; fRXPat->fDataSize += 2; int32_t op = URX_BUILD(URX_LA_START, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_STATE_SAVE, 0); // dest address will be patched later. fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_NOP, 0); fRXPat->fCompiledPat->addElement(op, *fStatus); // On the Parentheses stack, start a new frame and add the postions // of the StateSave and NOP. fParenStack.push(fModeFlags, *fStatus); // Match mode state fParenStack.push(negLookAhead, *fStatus); // Frame type fParenStack.push(fRXPat->fCompiledPat->size()-2, *fStatus); // The STATE_SAVE location fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // The second NOP location // Instructions #5 - #7 will be added when the ')' is encountered. } break; case doOpenLookBehind: { // Compile a (?<= look-behind open paren. // // Compiles to // 0 URX_LB_START dataLoc // 1 URX_LB_CONT dataLoc // 2 MinMatchLen // 3 MaxMatchLen // 4 URX_NOP Standard '(' boilerplate. // 5 URX_NOP Reserved slot for use with '|' ops within (block). // 6 // 7 URX_LB_END dataLoc # Check match len, restore input len // 8 URX_LA_END dataLoc # Restore stack, input pos // // Allocate a block of matcher data, to contain (when running a match) // 0: Stack ptr on entry // 1: Input Index on entry // 2: Start index of match current match attempt. // 3: Original Input String len. // Generate match code for any pending literals. fixLiterals(); // Allocate data space int32_t dataLoc = fRXPat->fDataSize; fRXPat->fDataSize += 4; // Emit URX_LB_START int32_t op = URX_BUILD(URX_LB_START, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); // Emit URX_LB_CONT op = URX_BUILD(URX_LB_CONT, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); fRXPat->fCompiledPat->addElement(0, *fStatus); // MinMatchLength. To be filled later. fRXPat->fCompiledPat->addElement(0, *fStatus); // MaxMatchLength. To be filled later. // Emit the NOP op = URX_BUILD(URX_NOP, 0); fRXPat->fCompiledPat->addElement(op, *fStatus); fRXPat->fCompiledPat->addElement(op, *fStatus); // On the Parentheses stack, start a new frame and add the postions // of the URX_LB_CONT and the NOP. fParenStack.push(fModeFlags, *fStatus); // Match mode state fParenStack.push(lookBehind, *fStatus); // Frame type fParenStack.push(fRXPat->fCompiledPat->size()-2, *fStatus); // The first NOP location fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // The 2nd NOP location // The final two instructions will be added when the ')' is encountered. } break; case doOpenLookBehindNeg: { // Compile a (? // 8 URX_LBN_END dataLoc # Check match len, cause a FAIL // 9 ... // // Allocate a block of matcher data, to contain (when running a match) // 0: Stack ptr on entry // 1: Input Index on entry // 2: Start index of match current match attempt. // 3: Original Input String len. // Generate match code for any pending literals. fixLiterals(); // Allocate data space int32_t dataLoc = fRXPat->fDataSize; fRXPat->fDataSize += 4; // Emit URX_LB_START int32_t op = URX_BUILD(URX_LB_START, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); // Emit URX_LBN_CONT op = URX_BUILD(URX_LBN_CONT, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); fRXPat->fCompiledPat->addElement(0, *fStatus); // MinMatchLength. To be filled later. fRXPat->fCompiledPat->addElement(0, *fStatus); // MaxMatchLength. To be filled later. fRXPat->fCompiledPat->addElement(0, *fStatus); // Continue Loc. To be filled later. // Emit the NOP op = URX_BUILD(URX_NOP, 0); fRXPat->fCompiledPat->addElement(op, *fStatus); fRXPat->fCompiledPat->addElement(op, *fStatus); // On the Parentheses stack, start a new frame and add the postions // of the URX_LB_CONT and the NOP. fParenStack.push(fModeFlags, *fStatus); // Match mode state fParenStack.push(lookBehindN, *fStatus); // Frame type fParenStack.push(fRXPat->fCompiledPat->size()-2, *fStatus); // The first NOP location fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // The 2nd NOP location // The final two instructions will be added when the ')' is encountered. } break; case doConditionalExpr: // Conditionals such as (?(1)a:b) case doPerlInline: // Perl inline-condtionals. (?{perl code}a|b) We're not perl, no way to do them. error(U_REGEX_UNIMPLEMENTED); break; case doCloseParen: handleCloseParen(); if (fParenStack.size() <= 0) { // Extra close paren, or missing open paren. error(U_REGEX_MISMATCHED_PAREN); } break; case doNOP: break; case doBadOpenParenType: case doRuleError: error(U_REGEX_RULE_SYNTAX); break; case doMismatchedParenErr: error(U_REGEX_MISMATCHED_PAREN); break; case doPlus: // Normal '+' compiles to // 1. stuff to be repeated (already built) // 2. jmp-sav 1 // 3. ... // // Or, if the item to be repeated can match a zero length string, // 1. STO_INP_LOC data-loc // 2. body of stuff to be repeated // 3. JMP_SAV_X 2 // 4. ... // // Or, if the item to be repeated is simple // 1. Item to be repeated. // 2. LOOP_SR_I set number (assuming repeated item is a set ref) // 3. LOOP_C stack location { int32_t topLoc = blockTopLoc(FALSE); // location of item #1 int32_t frameLoc; // Check for simple constructs, which may get special optimized code. if (topLoc == fRXPat->fCompiledPat->size() - 1) { int32_t repeatedOp = (int32_t)fRXPat->fCompiledPat->elementAti(topLoc); if (URX_TYPE(repeatedOp) == URX_SETREF) { // Emit optimized code for [char set]+ int32_t loopOpI = URX_BUILD(URX_LOOP_SR_I, URX_VAL(repeatedOp)); fRXPat->fCompiledPat->addElement(loopOpI, *fStatus); frameLoc = fRXPat->fFrameSize; fRXPat->fFrameSize++; int32_t loopOpC = URX_BUILD(URX_LOOP_C, frameLoc); fRXPat->fCompiledPat->addElement(loopOpC, *fStatus); break; } if (URX_TYPE(repeatedOp) == URX_DOTANY || URX_TYPE(repeatedOp) == URX_DOTANY_ALL || URX_TYPE(repeatedOp) == URX_DOTANY_UNIX) { // Emit Optimized code for .+ operations. int32_t loopOpI = URX_BUILD(URX_LOOP_DOT_I, 0); if (URX_TYPE(repeatedOp) == URX_DOTANY_ALL) { // URX_LOOP_DOT_I operand is a flag indicating ". matches any" mode. loopOpI |= 1; } if (fModeFlags & UREGEX_UNIX_LINES) { loopOpI |= 2; } fRXPat->fCompiledPat->addElement(loopOpI, *fStatus); frameLoc = fRXPat->fFrameSize; fRXPat->fFrameSize++; int32_t loopOpC = URX_BUILD(URX_LOOP_C, frameLoc); fRXPat->fCompiledPat->addElement(loopOpC, *fStatus); break; } } // General case. // Check for minimum match length of zero, which requires // extra loop-breaking code. if (minMatchLength(topLoc, fRXPat->fCompiledPat->size()-1) == 0) { // Zero length match is possible. // Emit the code sequence that can handle it. insertOp(topLoc); frameLoc = fRXPat->fFrameSize; fRXPat->fFrameSize++; int32_t op = URX_BUILD(URX_STO_INP_LOC, frameLoc); fRXPat->fCompiledPat->setElementAt(op, topLoc); op = URX_BUILD(URX_JMP_SAV_X, topLoc+1); fRXPat->fCompiledPat->addElement(op, *fStatus); } else { // Simpler code when the repeated body must match something non-empty int32_t jmpOp = URX_BUILD(URX_JMP_SAV, topLoc); fRXPat->fCompiledPat->addElement(jmpOp, *fStatus); } } break; case doNGPlus: // Non-greedy '+?' compiles to // 1. stuff to be repeated (already built) // 2. state-save 1 // 3. ... { int32_t topLoc = blockTopLoc(FALSE); int32_t saveStateOp = URX_BUILD(URX_STATE_SAVE, topLoc); fRXPat->fCompiledPat->addElement(saveStateOp, *fStatus); } break; case doOpt: // Normal (greedy) ? quantifier. // Compiles to // 1. state save 3 // 2. body of optional block // 3. ... // Insert the state save into the compiled pattern, and we're done. { int32_t saveStateLoc = blockTopLoc(TRUE); int32_t saveStateOp = URX_BUILD(URX_STATE_SAVE, fRXPat->fCompiledPat->size()); fRXPat->fCompiledPat->setElementAt(saveStateOp, saveStateLoc); } break; case doNGOpt: // Non-greedy ?? quantifier // compiles to // 1. jmp 4 // 2. body of optional block // 3 jmp 5 // 4. state save 2 // 5 ... // This code is less than ideal, with two jmps instead of one, because we can only // insert one instruction at the top of the block being iterated. { int32_t jmp1_loc = blockTopLoc(TRUE); int32_t jmp2_loc = fRXPat->fCompiledPat->size(); int32_t jmp1_op = URX_BUILD(URX_JMP, jmp2_loc+1); fRXPat->fCompiledPat->setElementAt(jmp1_op, jmp1_loc); int32_t jmp2_op = URX_BUILD(URX_JMP, jmp2_loc+2); fRXPat->fCompiledPat->addElement(jmp2_op, *fStatus); int32_t save_op = URX_BUILD(URX_STATE_SAVE, jmp1_loc+1); fRXPat->fCompiledPat->addElement(save_op, *fStatus); } break; case doStar: // Normal (greedy) * quantifier. // Compiles to // 1. STATE_SAVE 4 // 2. body of stuff being iterated over // 3. JMP_SAV 2 // 4. ... // // Or, if the body is a simple [Set], // 1. LOOP_SR_I set number // 2. LOOP_C stack location // ... // // Or if this is a .* // 1. LOOP_DOT_I (. matches all mode flag) // 2. LOOP_C stack location // // Or, if the body can match a zero-length string, to inhibit infinite loops, // 1. STATE_SAVE 5 // 2. STO_INP_LOC data-loc // 3. body of stuff // 4. JMP_SAV_X 2 // 5. ... { // location of item #1, the STATE_SAVE int32_t topLoc = blockTopLoc(FALSE); int32_t dataLoc = -1; // Check for simple *, where the construct being repeated // compiled to single opcode, and might be optimizable. if (topLoc == fRXPat->fCompiledPat->size() - 1) { int32_t repeatedOp = (int32_t)fRXPat->fCompiledPat->elementAti(topLoc); if (URX_TYPE(repeatedOp) == URX_SETREF) { // Emit optimized code for a [char set]* int32_t loopOpI = URX_BUILD(URX_LOOP_SR_I, URX_VAL(repeatedOp)); fRXPat->fCompiledPat->setElementAt(loopOpI, topLoc); dataLoc = fRXPat->fFrameSize; fRXPat->fFrameSize++; int32_t loopOpC = URX_BUILD(URX_LOOP_C, dataLoc); fRXPat->fCompiledPat->addElement(loopOpC, *fStatus); break; } if (URX_TYPE(repeatedOp) == URX_DOTANY || URX_TYPE(repeatedOp) == URX_DOTANY_ALL || URX_TYPE(repeatedOp) == URX_DOTANY_UNIX) { // Emit Optimized code for .* operations. int32_t loopOpI = URX_BUILD(URX_LOOP_DOT_I, 0); if (URX_TYPE(repeatedOp) == URX_DOTANY_ALL) { // URX_LOOP_DOT_I operand is a flag indicating . matches any mode. loopOpI |= 1; } if ((fModeFlags & UREGEX_UNIX_LINES) != 0) { loopOpI |= 2; } fRXPat->fCompiledPat->setElementAt(loopOpI, topLoc); dataLoc = fRXPat->fFrameSize; fRXPat->fFrameSize++; int32_t loopOpC = URX_BUILD(URX_LOOP_C, dataLoc); fRXPat->fCompiledPat->addElement(loopOpC, *fStatus); break; } } // Emit general case code for this * // The optimizations did not apply. int32_t saveStateLoc = blockTopLoc(TRUE); int32_t jmpOp = URX_BUILD(URX_JMP_SAV, saveStateLoc+1); // Check for minimum match length of zero, which requires // extra loop-breaking code. if (minMatchLength(saveStateLoc, fRXPat->fCompiledPat->size()-1) == 0) { insertOp(saveStateLoc); dataLoc = fRXPat->fFrameSize; fRXPat->fFrameSize++; int32_t op = URX_BUILD(URX_STO_INP_LOC, dataLoc); fRXPat->fCompiledPat->setElementAt(op, saveStateLoc+1); jmpOp = URX_BUILD(URX_JMP_SAV_X, saveStateLoc+2); } // Locate the position in the compiled pattern where the match will continue // after completing the *. (4 or 5 in the comment above) int32_t continueLoc = fRXPat->fCompiledPat->size()+1; // Put together the save state op store it into the compiled code. int32_t saveStateOp = URX_BUILD(URX_STATE_SAVE, continueLoc); fRXPat->fCompiledPat->setElementAt(saveStateOp, saveStateLoc); // Append the URX_JMP_SAV or URX_JMPX operation to the compiled pattern. fRXPat->fCompiledPat->addElement(jmpOp, *fStatus); } break; case doNGStar: // Non-greedy *? quantifier // compiles to // 1. JMP 3 // 2. body of stuff being iterated over // 3. STATE_SAVE 2 // 4 ... { int32_t jmpLoc = blockTopLoc(TRUE); // loc 1. int32_t saveLoc = fRXPat->fCompiledPat->size(); // loc 3. int32_t jmpOp = URX_BUILD(URX_JMP, saveLoc); int32_t stateSaveOp = URX_BUILD(URX_STATE_SAVE, jmpLoc+1); fRXPat->fCompiledPat->setElementAt(jmpOp, jmpLoc); fRXPat->fCompiledPat->addElement(stateSaveOp, *fStatus); } break; case doIntervalInit: // The '{' opening an interval quantifier was just scanned. // Init the counter varaiables that will accumulate the values as the digits // are scanned. fIntervalLow = 0; fIntervalUpper = -1; break; case doIntevalLowerDigit: // Scanned a digit from the lower value of an {lower,upper} interval { int32_t digitValue = u_charDigitValue(fC.fChar); U_ASSERT(digitValue >= 0); fIntervalLow = fIntervalLow*10 + digitValue; if (fIntervalLow < 0) { error(U_REGEX_NUMBER_TOO_BIG); } } break; case doIntervalUpperDigit: // Scanned a digit from the upper value of an {lower,upper} interval { if (fIntervalUpper < 0) { fIntervalUpper = 0; } int32_t digitValue = u_charDigitValue(fC.fChar); U_ASSERT(digitValue >= 0); fIntervalUpper = fIntervalUpper*10 + digitValue; if (fIntervalUpper < 0) { error(U_REGEX_NUMBER_TOO_BIG); } } break; case doIntervalSame: // Scanned a single value interval like {27}. Upper = Lower. fIntervalUpper = fIntervalLow; break; case doInterval: // Finished scanning a normal {lower,upper} interval. Generate the code for it. if (compileInlineInterval() == FALSE) { compileInterval(URX_CTR_INIT, URX_CTR_LOOP); } break; case doPossessiveInterval: // Finished scanning a Possessive {lower,upper}+ interval. Generate the code for it. { // Remember the loc for the top of the block being looped over. // (Can not reserve a slot in the compiled pattern at this time, because // compileInterval needs to reserve also, and blockTopLoc can only reserve // once per block.) int32_t topLoc = blockTopLoc(FALSE); // Produce normal looping code. compileInterval(URX_CTR_INIT, URX_CTR_LOOP); // Surround the just-emitted normal looping code with a STO_SP ... LD_SP // just as if the loop was inclosed in atomic parentheses. // First the STO_SP before the start of the loop insertOp(topLoc); int32_t varLoc = fRXPat->fDataSize; // Reserve a data location for saving the fRXPat->fDataSize += 1; // state stack ptr. int32_t op = URX_BUILD(URX_STO_SP, varLoc); fRXPat->fCompiledPat->setElementAt(op, topLoc); int32_t loopOp = (int32_t)fRXPat->fCompiledPat->popi(); U_ASSERT(URX_TYPE(loopOp) == URX_CTR_LOOP && URX_VAL(loopOp) == topLoc); loopOp++; // point LoopOp after the just-inserted STO_SP fRXPat->fCompiledPat->push(loopOp, *fStatus); // Then the LD_SP after the end of the loop op = URX_BUILD(URX_LD_SP, varLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); } break; case doNGInterval: // Finished scanning a non-greedy {lower,upper}? interval. Generate the code for it. compileInterval(URX_CTR_INIT_NG, URX_CTR_LOOP_NG); break; case doIntervalError: error(U_REGEX_BAD_INTERVAL); break; case doLiteralChar: // We've just scanned a "normal" character from the pattern, literalChar(fC.fChar); break; case doEscapedLiteralChar: // We've just scanned an backslashed escaped character with no // special meaning. It represents itself. if ((fModeFlags & UREGEX_ERROR_ON_UNKNOWN_ESCAPES) != 0 && ((fC.fChar >= 0x41 && fC.fChar<= 0x5A) || // in [A-Z] (fC.fChar >= 0x61 && fC.fChar <= 0x7a))) { // in [a-z] error(U_REGEX_BAD_ESCAPE_SEQUENCE); } literalChar(fC.fChar); break; case doDotAny: // scanned a ".", match any single character. { fixLiterals(FALSE); int32_t op; if (fModeFlags & UREGEX_DOTALL) { op = URX_BUILD(URX_DOTANY_ALL, 0); } else if (fModeFlags & UREGEX_UNIX_LINES) { op = URX_BUILD(URX_DOTANY_UNIX, 0); } else { op = URX_BUILD(URX_DOTANY, 0); } fRXPat->fCompiledPat->addElement(op, *fStatus); } break; case doCaret: { fixLiterals(FALSE); int32_t op = 0; if ( (fModeFlags & UREGEX_MULTILINE) == 0 && (fModeFlags & UREGEX_UNIX_LINES) == 0) { op = URX_CARET; } else if ((fModeFlags & UREGEX_MULTILINE) != 0 && (fModeFlags & UREGEX_UNIX_LINES) == 0) { op = URX_CARET_M; } else if ((fModeFlags & UREGEX_MULTILINE) == 0 && (fModeFlags & UREGEX_UNIX_LINES) != 0) { op = URX_CARET; // Only testing true start of input. } else if ((fModeFlags & UREGEX_MULTILINE) != 0 && (fModeFlags & UREGEX_UNIX_LINES) != 0) { op = URX_CARET_M_UNIX; } fRXPat->fCompiledPat->addElement(URX_BUILD(op, 0), *fStatus); } break; case doDollar: { fixLiterals(FALSE); int32_t op = 0; if ( (fModeFlags & UREGEX_MULTILINE) == 0 && (fModeFlags & UREGEX_UNIX_LINES) == 0) { op = URX_DOLLAR; } else if ((fModeFlags & UREGEX_MULTILINE) != 0 && (fModeFlags & UREGEX_UNIX_LINES) == 0) { op = URX_DOLLAR_M; } else if ((fModeFlags & UREGEX_MULTILINE) == 0 && (fModeFlags & UREGEX_UNIX_LINES) != 0) { op = URX_DOLLAR_D; } else if ((fModeFlags & UREGEX_MULTILINE) != 0 && (fModeFlags & UREGEX_UNIX_LINES) != 0) { op = URX_DOLLAR_MD; } fRXPat->fCompiledPat->addElement(URX_BUILD(op, 0), *fStatus); } break; case doBackslashA: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_CARET, 0), *fStatus); break; case doBackslashB: { #if UCONFIG_NO_BREAK_ITERATION==1 if (fModeFlags & UREGEX_UWORD) { error(U_UNSUPPORTED_ERROR); } #endif fixLiterals(FALSE); int32_t op = (fModeFlags & UREGEX_UWORD)? URX_BACKSLASH_BU : URX_BACKSLASH_B; fRXPat->fCompiledPat->addElement(URX_BUILD(op, 1), *fStatus); } break; case doBackslashb: { #if UCONFIG_NO_BREAK_ITERATION==1 if (fModeFlags & UREGEX_UWORD) { error(U_UNSUPPORTED_ERROR); } #endif fixLiterals(FALSE); int32_t op = (fModeFlags & UREGEX_UWORD)? URX_BACKSLASH_BU : URX_BACKSLASH_B; fRXPat->fCompiledPat->addElement(URX_BUILD(op, 0), *fStatus); } break; case doBackslashD: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_BACKSLASH_D, 1), *fStatus); break; case doBackslashd: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_BACKSLASH_D, 0), *fStatus); break; case doBackslashG: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_BACKSLASH_G, 0), *fStatus); break; case doBackslashS: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement( URX_BUILD(URX_STAT_SETREF_N, URX_ISSPACE_SET), *fStatus); break; case doBackslashs: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement( URX_BUILD(URX_STATIC_SETREF, URX_ISSPACE_SET), *fStatus); break; case doBackslashW: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement( URX_BUILD(URX_STAT_SETREF_N, URX_ISWORD_SET), *fStatus); break; case doBackslashw: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement( URX_BUILD(URX_STATIC_SETREF, URX_ISWORD_SET), *fStatus); break; case doBackslashX: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_BACKSLASH_X, 0), *fStatus); break; case doBackslashZ: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_DOLLAR, 0), *fStatus); break; case doBackslashz: fixLiterals(FALSE); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_BACKSLASH_Z, 0), *fStatus); break; case doEscapeError: error(U_REGEX_BAD_ESCAPE_SEQUENCE); break; case doExit: fixLiterals(FALSE); returnVal = FALSE; break; case doProperty: { fixLiterals(FALSE); UnicodeSet *theSet = scanProp(); compileSet(theSet); } break; case doNamedChar: { UChar32 c = scanNamedChar(); literalChar(c); } break; case doBackRef: // BackReference. Somewhat unusual in that the front-end can not completely parse // the regular expression, because the number of digits to be consumed // depends on the number of capture groups that have been defined. So // we have to do it here instead. { int32_t numCaptureGroups = fRXPat->fGroupMap->size(); int32_t groupNum = 0; UChar32 c = fC.fChar; for (;;) { // Loop once per digit, for max allowed number of digits in a back reference. int32_t digit = u_charDigitValue(c); groupNum = groupNum * 10 + digit; if (groupNum >= numCaptureGroups) { break; } c = peekCharLL(); if (RegexStaticSets::gStaticSets->fRuleDigitsAlias->contains(c) == FALSE) { break; } nextCharLL(); } // Scan of the back reference in the source regexp is complete. Now generate // the compiled code for it. // Because capture groups can be forward-referenced by back-references, // we fill the operand with the capture group number. At the end // of compilation, it will be changed to the variable's location. U_ASSERT(groupNum > 0); // Shouldn't happen. '\0' begins an octal escape sequence, // and shouldn't enter this code path at all. fixLiterals(FALSE); int32_t op; if (fModeFlags & UREGEX_CASE_INSENSITIVE) { op = URX_BUILD(URX_BACKREF_I, groupNum); } else { op = URX_BUILD(URX_BACKREF, groupNum); } fRXPat->fCompiledPat->addElement(op, *fStatus); } break; case doPossessivePlus: // Possessive ++ quantifier. // Compiles to // 1. STO_SP // 2. body of stuff being iterated over // 3. STATE_SAVE 5 // 4. JMP 2 // 5. LD_SP // 6. ... // // Note: TODO: This is pretty inefficient. A mass of saved state is built up // then unconditionally discarded. Perhaps introduce a new opcode. Ticket 6056 // { // Emit the STO_SP int32_t topLoc = blockTopLoc(TRUE); int32_t stoLoc = fRXPat->fDataSize; fRXPat->fDataSize++; // Reserve the data location for storing save stack ptr. int32_t op = URX_BUILD(URX_STO_SP, stoLoc); fRXPat->fCompiledPat->setElementAt(op, topLoc); // Emit the STATE_SAVE op = URX_BUILD(URX_STATE_SAVE, fRXPat->fCompiledPat->size()+2); fRXPat->fCompiledPat->addElement(op, *fStatus); // Emit the JMP op = URX_BUILD(URX_JMP, topLoc+1); fRXPat->fCompiledPat->addElement(op, *fStatus); // Emit the LD_SP op = URX_BUILD(URX_LD_SP, stoLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); } break; case doPossessiveStar: // Possessive *+ quantifier. // Compiles to // 1. STO_SP loc // 2. STATE_SAVE 5 // 3. body of stuff being iterated over // 4. JMP 2 // 5. LD_SP loc // 6 ... // TODO: do something to cut back the state stack each time through the loop. { // Reserve two slots at the top of the block. int32_t topLoc = blockTopLoc(TRUE); insertOp(topLoc); // emit STO_SP loc int32_t stoLoc = fRXPat->fDataSize; fRXPat->fDataSize++; // Reserve the data location for storing save stack ptr. int32_t op = URX_BUILD(URX_STO_SP, stoLoc); fRXPat->fCompiledPat->setElementAt(op, topLoc); // Emit the SAVE_STATE 5 int32_t L7 = fRXPat->fCompiledPat->size()+1; op = URX_BUILD(URX_STATE_SAVE, L7); fRXPat->fCompiledPat->setElementAt(op, topLoc+1); // Append the JMP operation. op = URX_BUILD(URX_JMP, topLoc+1); fRXPat->fCompiledPat->addElement(op, *fStatus); // Emit the LD_SP loc op = URX_BUILD(URX_LD_SP, stoLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); } break; case doPossessiveOpt: // Possessive ?+ quantifier. // Compiles to // 1. STO_SP loc // 2. SAVE_STATE 5 // 3. body of optional block // 4. LD_SP loc // 5. ... // { // Reserve two slots at the top of the block. int32_t topLoc = blockTopLoc(TRUE); insertOp(topLoc); // Emit the STO_SP int32_t stoLoc = fRXPat->fDataSize; fRXPat->fDataSize++; // Reserve the data location for storing save stack ptr. int32_t op = URX_BUILD(URX_STO_SP, stoLoc); fRXPat->fCompiledPat->setElementAt(op, topLoc); // Emit the SAVE_STATE int32_t continueLoc = fRXPat->fCompiledPat->size()+1; op = URX_BUILD(URX_STATE_SAVE, continueLoc); fRXPat->fCompiledPat->setElementAt(op, topLoc+1); // Emit the LD_SP op = URX_BUILD(URX_LD_SP, stoLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); } break; case doBeginMatchMode: fNewModeFlags = fModeFlags; fSetModeFlag = TRUE; break; case doMatchMode: // (?i) and similar { int32_t bit = 0; switch (fC.fChar) { case 0x69: /* 'i' */ bit = UREGEX_CASE_INSENSITIVE; break; case 0x64: /* 'd' */ bit = UREGEX_UNIX_LINES; break; case 0x6d: /* 'm' */ bit = UREGEX_MULTILINE; break; case 0x73: /* 's' */ bit = UREGEX_DOTALL; break; case 0x75: /* 'u' */ bit = 0; /* Unicode casing */ break; case 0x77: /* 'w' */ bit = UREGEX_UWORD; break; case 0x78: /* 'x' */ bit = UREGEX_COMMENTS; break; case 0x2d: /* '-' */ fSetModeFlag = FALSE; break; default: U_ASSERT(FALSE); // Should never happen. Other chars are filtered out // by the scanner. } if (fSetModeFlag) { fNewModeFlags |= bit; } else { fNewModeFlags &= ~bit; } } break; case doSetMatchMode: // Emit code to match any pending literals, using the not-yet changed match mode. fixLiterals(); // We've got a (?i) or similar. The match mode is being changed, but // the change is not scoped to a parenthesized block. U_ASSERT(fNewModeFlags < 0); fModeFlags = fNewModeFlags; break; case doMatchModeParen: // We've got a (?i: or similar. Begin a parenthesized block, save old // mode flags so they can be restored at the close of the block. // // Compile to a // - NOP, which later may be replaced by a save-state if the // parenthesized group gets a * quantifier, followed by // - NOP, which may later be replaced by a save-state if there // is an '|' alternation within the parens. { fixLiterals(FALSE); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); fRXPat->fCompiledPat->addElement(URX_BUILD(URX_NOP, 0), *fStatus); // On the Parentheses stack, start a new frame and add the postions // of the two NOPs (a normal non-capturing () frame, except for the // saving of the orignal mode flags.) fParenStack.push(fModeFlags, *fStatus); fParenStack.push(flags, *fStatus); // Frame Marker fParenStack.push(fRXPat->fCompiledPat->size()-2, *fStatus); // The first NOP fParenStack.push(fRXPat->fCompiledPat->size()-1, *fStatus); // The second NOP // Set the current mode flags to the new values. U_ASSERT(fNewModeFlags < 0); fModeFlags = fNewModeFlags; } break; case doBadModeFlag: error(U_REGEX_INVALID_FLAG); break; case doSuppressComments: // We have just scanned a '(?'. We now need to prevent the character scanner from // treating a '#' as a to-the-end-of-line comment. // (This Perl compatibility just gets uglier and uglier to do...) fEOLComments = FALSE; break; case doSetAddAmp: { UnicodeSet *set = (UnicodeSet *)fSetStack.peek(); set->add(chAmp); } break; case doSetAddDash: { UnicodeSet *set = (UnicodeSet *)fSetStack.peek(); set->add(chDash); } break; case doSetBackslash_s: { UnicodeSet *set = (UnicodeSet *)fSetStack.peek(); set->addAll(*RegexStaticSets::gStaticSets->fPropSets[URX_ISSPACE_SET]); break; } case doSetBackslash_S: { UnicodeSet *set = (UnicodeSet *)fSetStack.peek(); UnicodeSet SSet(*RegexStaticSets::gStaticSets->fPropSets[URX_ISSPACE_SET]); SSet.complement(); set->addAll(SSet); break; } case doSetBackslash_d: { UnicodeSet *set = (UnicodeSet *)fSetStack.peek(); // TODO - make a static set, ticket 6058. addCategory(set, U_GC_ND_MASK, *fStatus); break; } case doSetBackslash_D: { UnicodeSet *set = (UnicodeSet *)fSetStack.peek(); UnicodeSet digits; // TODO - make a static set, ticket 6058. digits.applyIntPropertyValue(UCHAR_GENERAL_CATEGORY_MASK, U_GC_ND_MASK, *fStatus); digits.complement(); set->addAll(digits); break; } case doSetBackslash_w: { UnicodeSet *set = (UnicodeSet *)fSetStack.peek(); set->addAll(*RegexStaticSets::gStaticSets->fPropSets[URX_ISWORD_SET]); break; } case doSetBackslash_W: { UnicodeSet *set = (UnicodeSet *)fSetStack.peek(); UnicodeSet SSet(*RegexStaticSets::gStaticSets->fPropSets[URX_ISWORD_SET]); SSet.complement(); set->addAll(SSet); break; } case doSetBegin: fixLiterals(FALSE); fSetStack.push(new UnicodeSet(), *fStatus); fSetOpStack.push(setStart, *fStatus); if ((fModeFlags & UREGEX_CASE_INSENSITIVE) != 0) { fSetOpStack.push(setCaseClose, *fStatus); } break; case doSetBeginDifference1: // We have scanned something like [[abc]-[ // Set up a new UnicodeSet for the set beginning with the just-scanned '[' // Push a Difference operator, which will cause the new set to be subtracted from what // went before once it is created. setPushOp(setDifference1); fSetOpStack.push(setStart, *fStatus); if ((fModeFlags & UREGEX_CASE_INSENSITIVE) != 0) { fSetOpStack.push(setCaseClose, *fStatus); } break; case doSetBeginIntersection1: // We have scanned something like [[abc]&[ // Need both the '&' operator and the open '[' operator. setPushOp(setIntersection1); fSetOpStack.push(setStart, *fStatus); if ((fModeFlags & UREGEX_CASE_INSENSITIVE) != 0) { fSetOpStack.push(setCaseClose, *fStatus); } break; case doSetBeginUnion: // We have scanned something like [[abc][ // Need to handle the union operation explicitly [[abc] | [ setPushOp(setUnion); fSetOpStack.push(setStart, *fStatus); if ((fModeFlags & UREGEX_CASE_INSENSITIVE) != 0) { fSetOpStack.push(setCaseClose, *fStatus); } break; case doSetDifference2: // We have scanned something like [abc-- // Consider this to unambiguously be a set difference operator. setPushOp(setDifference2); break; case doSetEnd: // Have encountered the ']' that closes a set. // Force the evaluation of any pending operations within this set, // leave the completed set on the top of the set stack. setEval(setEnd); U_ASSERT(fSetOpStack.peeki()==setStart); fSetOpStack.popi(); break; case doSetFinish: { // Finished a complete set expression, including all nested sets. // The close bracket has already triggered clearing out pending set operators, // the operator stack should be empty and the operand stack should have just // one entry, the result set. U_ASSERT(fSetOpStack.empty()); UnicodeSet *theSet = (UnicodeSet *)fSetStack.pop(); U_ASSERT(fSetStack.empty()); compileSet(theSet); break; } case doSetIntersection2: // Have scanned something like [abc&& setPushOp(setIntersection2); break; case doSetLiteral: // Union the just-scanned literal character into the set being built. // This operation is the highest precedence set operation, so we can always do // it immediately, without waiting to see what follows. It is necessary to perform // any pending '-' or '&' operation first, because these have the same precedence // as union-ing in a literal' { setEval(setUnion); UnicodeSet *s = (UnicodeSet *)fSetStack.peek(); s->add(fC.fChar); fLastSetLiteral = fC.fChar; break; } case doSetLiteralEscaped: // A back-slash escaped literal character was encountered. // Processing is the same as with setLiteral, above, with the addition of // the optional check for errors on escaped ASCII letters. { if ((fModeFlags & UREGEX_ERROR_ON_UNKNOWN_ESCAPES) != 0 && ((fC.fChar >= 0x41 && fC.fChar<= 0x5A) || // in [A-Z] (fC.fChar >= 0x61 && fC.fChar <= 0x7a))) { // in [a-z] error(U_REGEX_BAD_ESCAPE_SEQUENCE); } setEval(setUnion); UnicodeSet *s = (UnicodeSet *)fSetStack.peek(); s->add(fC.fChar); fLastSetLiteral = fC.fChar; break; } case doSetNamedChar: // Scanning a \N{UNICODE CHARACTER NAME} // Aside from the source of the character, the processing is identical to doSetLiteral, // above. { UChar32 c = scanNamedChar(); setEval(setUnion); UnicodeSet *s = (UnicodeSet *)fSetStack.peek(); s->add(c); fLastSetLiteral = c; break; } case doSetNamedRange: // We have scanned literal-\N{CHAR NAME}. Add the range to the set. // The left character is already in the set, and is saved in fLastSetLiteral. // The right side needs to be picked up, the scan is at the 'N'. // Lower Limit > Upper limit being an error matches both Java // and ICU UnicodeSet behavior. { UChar32 c = scanNamedChar(); if (U_SUCCESS(*fStatus) && fLastSetLiteral > c) { error(U_REGEX_INVALID_RANGE); } UnicodeSet *s = (UnicodeSet *)fSetStack.peek(); s->add(fLastSetLiteral, c); fLastSetLiteral = c; break; } case doSetNegate: // Scanned a '^' at the start of a set. // Push the negation operator onto the set op stack. // A twist for case-insensitive matching: // the case closure operation must happen _before_ negation. // But the case closure operation will already be on the stack if it's required. // This requires checking for case closure, and swapping the stack order // if it is present. { int32_t tosOp = fSetOpStack.peeki(); if (tosOp == setCaseClose) { fSetOpStack.popi(); fSetOpStack.push(setNegation, *fStatus); fSetOpStack.push(setCaseClose, *fStatus); } else { fSetOpStack.push(setNegation, *fStatus); } } break; case doSetNoCloseError: error(U_REGEX_MISSING_CLOSE_BRACKET); break; case doSetOpError: error(U_REGEX_RULE_SYNTAX); // -- or && at the end of a set. Illegal. break; case doSetPosixProp: { UnicodeSet *s = scanPosixProp(); if (s != NULL) { UnicodeSet *tos = (UnicodeSet *)fSetStack.peek(); tos->addAll(*s); delete s; } // else error. scanProp() reported the error status already. } break; case doSetProp: // Scanned a \p \P within [brackets]. { UnicodeSet *s = scanProp(); if (s != NULL) { UnicodeSet *tos = (UnicodeSet *)fSetStack.peek(); tos->addAll(*s); delete s; } // else error. scanProp() reported the error status already. } break; case doSetRange: // We have scanned literal-literal. Add the range to the set. // The left character is already in the set, and is saved in fLastSetLiteral. // The right side is the current character. // Lower Limit > Upper limit being an error matches both Java // and ICU UnicodeSet behavior. { if (fLastSetLiteral > fC.fChar) { error(U_REGEX_INVALID_RANGE); } UnicodeSet *s = (UnicodeSet *)fSetStack.peek(); s->add(fLastSetLiteral, fC.fChar); break; } default: U_ASSERT(FALSE); error(U_REGEX_INTERNAL_ERROR); break; } if (U_FAILURE(*fStatus)) { returnVal = FALSE; } return returnVal; } //------------------------------------------------------------------------------ // // literalChar We've encountered a literal character from the pattern, // or an escape sequence that reduces to a character. // Add it to the string containing all literal chars/strings from // the pattern. // //------------------------------------------------------------------------------ void RegexCompile::literalChar(UChar32 c) { fLiteralChars.append(c); } //------------------------------------------------------------------------------ // // fixLiterals When compiling something that can follow a literal // string in a pattern, emit the code to match the // accumulated literal string. // // Optionally, split the last char of the string off into // a single "ONE_CHAR" operation, so that quantifiers can // apply to that char alone. Example: abc* // The * must apply to the 'c' only. // //------------------------------------------------------------------------------ void RegexCompile::fixLiterals(UBool split) { int32_t op = 0; // An op from/for the compiled pattern. // If no literal characters have been scanned but not yet had code generated // for them, nothing needs to be done. if (fLiteralChars.length() == 0) { return; } int32_t indexOfLastCodePoint = fLiteralChars.moveIndex32(fLiteralChars.length(), -1); UChar32 lastCodePoint = fLiteralChars.char32At(indexOfLastCodePoint); // Split: We need to ensure that the last item in the compiled pattern // refers only to the last literal scanned in the pattern, so that // quantifiers (*, +, etc.) affect only it, and not a longer string. // Split before case folding for case insensitive matches. if (split) { fLiteralChars.truncate(indexOfLastCodePoint); fixLiterals(FALSE); // Recursive call, emit code to match the first part of the string. // Note that the truncated literal string may be empty, in which case // nothing will be emitted. literalChar(lastCodePoint); // Re-add the last code point as if it were a new literal. fixLiterals(FALSE); // Second recursive call, code for the final code point. return; } // If we are doing case-insensitive matching, case fold the string. This may expand // the string, e.g. the German sharp-s turns into "ss" if (fModeFlags & UREGEX_CASE_INSENSITIVE) { fLiteralChars.foldCase(); indexOfLastCodePoint = fLiteralChars.moveIndex32(fLiteralChars.length(), -1); lastCodePoint = fLiteralChars.char32At(indexOfLastCodePoint); } if (indexOfLastCodePoint == 0) { // Single character, emit a URX_ONECHAR op to match it. if ((fModeFlags & UREGEX_CASE_INSENSITIVE) && u_hasBinaryProperty(lastCodePoint, UCHAR_CASE_SENSITIVE)) { op = URX_BUILD(URX_ONECHAR_I, lastCodePoint); } else { op = URX_BUILD(URX_ONECHAR, lastCodePoint); } fRXPat->fCompiledPat->addElement(op, *fStatus); } else { // Two or more chars, emit a URX_STRING to match them. if (fModeFlags & UREGEX_CASE_INSENSITIVE) { op = URX_BUILD(URX_STRING_I, fRXPat->fLiteralText.length()); } else { // TODO here: add optimization to split case sensitive strings of length two // into two single char ops, for efficiency. op = URX_BUILD(URX_STRING, fRXPat->fLiteralText.length()); } fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_STRING_LEN, fLiteralChars.length()); fRXPat->fCompiledPat->addElement(op, *fStatus); // Add this string into the accumulated strings of the compiled pattern. fRXPat->fLiteralText.append(fLiteralChars); } fLiteralChars.remove(); } //------------------------------------------------------------------------------ // // insertOp() Insert a slot for a new opcode into the already // compiled pattern code. // // Fill the slot with a NOP. Our caller will replace it // with what they really wanted. // //------------------------------------------------------------------------------ void RegexCompile::insertOp(int32_t where) { UVector64 *code = fRXPat->fCompiledPat; U_ASSERT(where>0 && where < code->size()); int32_t nop = URX_BUILD(URX_NOP, 0); code->insertElementAt(nop, where, *fStatus); // Walk through the pattern, looking for any ops with targets that // were moved down by the insert. Fix them. int32_t loc; for (loc=0; locsize(); loc++) { int32_t op = (int32_t)code->elementAti(loc); int32_t opType = URX_TYPE(op); int32_t opValue = URX_VAL(op); if ((opType == URX_JMP || opType == URX_JMPX || opType == URX_STATE_SAVE || opType == URX_CTR_LOOP || opType == URX_CTR_LOOP_NG || opType == URX_JMP_SAV || opType == URX_JMP_SAV_X || opType == URX_RELOC_OPRND) && opValue > where) { // Target location for this opcode is after the insertion point and // needs to be incremented to adjust for the insertion. opValue++; op = URX_BUILD(opType, opValue); code->setElementAt(op, loc); } } // Now fix up the parentheses stack. All positive values in it are locations in // the compiled pattern. (Negative values are frame boundaries, and don't need fixing.) for (loc=0; locsize()); if (x>where) { x++; fParenStack.setElementAt(x, loc); } } if (fMatchCloseParen > where) { fMatchCloseParen++; } if (fMatchOpenParen > where) { fMatchOpenParen++; } } //------------------------------------------------------------------------------ // // blockTopLoc() Find or create a location in the compiled pattern // at the start of the operation or block that has // just been compiled. Needed when a quantifier (* or // whatever) appears, and we need to add an operation // at the start of the thing being quantified. // // (Parenthesized Blocks) have a slot with a NOP that // is reserved for this purpose. .* or similar don't // and a slot needs to be added. // // parameter reserveLoc : TRUE - ensure that there is space to add an opcode // at the returned location. // FALSE - just return the address, // do not reserve a location there. // //------------------------------------------------------------------------------ int32_t RegexCompile::blockTopLoc(UBool reserveLoc) { int32_t theLoc; fixLiterals(TRUE); // Emit code for any pending literals. // If last item was a string, emit separate op for the its last char. if (fRXPat->fCompiledPat->size() == fMatchCloseParen) { // The item just processed is a parenthesized block. theLoc = fMatchOpenParen; // A slot is already reserved for us. U_ASSERT(theLoc > 0); U_ASSERT(URX_TYPE(((uint32_t)fRXPat->fCompiledPat->elementAti(theLoc))) == URX_NOP); } else { // Item just compiled is a single thing, a ".", or a single char, a string or a set reference. // No slot for STATE_SAVE was pre-reserved in the compiled code. // We need to make space now. theLoc = fRXPat->fCompiledPat->size()-1; int32_t opAtTheLoc = (int32_t)fRXPat->fCompiledPat->elementAti(theLoc); if (URX_TYPE(opAtTheLoc) == URX_STRING_LEN) { // Strings take two opcode, we want the position of the first one. // We can have a string at this point if a single character case-folded to two. theLoc--; } if (reserveLoc) { int32_t nop = URX_BUILD(URX_NOP, 0); fRXPat->fCompiledPat->insertElementAt(nop, theLoc, *fStatus); } } return theLoc; } //------------------------------------------------------------------------------ // // handleCloseParen When compiling a close paren, we need to go back // and fix up any JMP or SAVE operations within the // parenthesized block that need to target the end // of the block. The locations of these are kept on // the paretheses stack. // // This function is called both when encountering a // real ) and at the end of the pattern. // //------------------------------------------------------------------------------ void RegexCompile::handleCloseParen() { int32_t patIdx; int32_t patOp; if (fParenStack.size() <= 0) { error(U_REGEX_MISMATCHED_PAREN); return; } // Emit code for any pending literals. fixLiterals(FALSE); // Fixup any operations within the just-closed parenthesized group // that need to reference the end of the (block). // (The first one popped from the stack is an unused slot for // alternation (OR) state save, but applying the fixup to it does no harm.) for (;;) { patIdx = fParenStack.popi(); if (patIdx < 0) { // value < 0 flags the start of the frame on the paren stack. break; } U_ASSERT(patIdx>0 && patIdx <= fRXPat->fCompiledPat->size()); patOp = (int32_t)fRXPat->fCompiledPat->elementAti(patIdx); U_ASSERT(URX_VAL(patOp) == 0); // Branch target for JMP should not be set. patOp |= fRXPat->fCompiledPat->size(); // Set it now. fRXPat->fCompiledPat->setElementAt(patOp, patIdx); fMatchOpenParen = patIdx; } // At the close of any parenthesized block, restore the match mode flags to // the value they had at the open paren. Saved value is // at the top of the paren stack. fModeFlags = fParenStack.popi(); U_ASSERT(fModeFlags < 0); // DO any additional fixups, depending on the specific kind of // parentesized grouping this is switch (patIdx) { case plain: case flags: // No additional fixups required. // (Grouping-only parentheses) break; case capturing: // Capturing Parentheses. // Insert a End Capture op into the pattern. // The frame offset of the variables for this cg is obtained from the // start capture op and put it into the end-capture op. { int32_t captureOp = (int32_t)fRXPat->fCompiledPat->elementAti(fMatchOpenParen+1); U_ASSERT(URX_TYPE(captureOp) == URX_START_CAPTURE); int32_t frameVarLocation = URX_VAL(captureOp); int32_t endCaptureOp = URX_BUILD(URX_END_CAPTURE, frameVarLocation); fRXPat->fCompiledPat->addElement(endCaptureOp, *fStatus); } break; case atomic: // Atomic Parenthesis. // Insert a LD_SP operation to restore the state stack to the position // it was when the atomic parens were entered. { int32_t stoOp = (int32_t)fRXPat->fCompiledPat->elementAti(fMatchOpenParen+1); U_ASSERT(URX_TYPE(stoOp) == URX_STO_SP); int32_t stoLoc = URX_VAL(stoOp); int32_t ldOp = URX_BUILD(URX_LD_SP, stoLoc); fRXPat->fCompiledPat->addElement(ldOp, *fStatus); } break; case lookAhead: { int32_t startOp = (int32_t)fRXPat->fCompiledPat->elementAti(fMatchOpenParen-5); U_ASSERT(URX_TYPE(startOp) == URX_LA_START); int32_t dataLoc = URX_VAL(startOp); int32_t op = URX_BUILD(URX_LA_END, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); } break; case negLookAhead: { // See comment at doOpenLookAheadNeg int32_t startOp = (int32_t)fRXPat->fCompiledPat->elementAti(fMatchOpenParen-1); U_ASSERT(URX_TYPE(startOp) == URX_LA_START); int32_t dataLoc = URX_VAL(startOp); int32_t op = URX_BUILD(URX_LA_END, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_BACKTRACK, 0); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_LA_END, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); // Patch the URX_SAVE near the top of the block. // The destination of the SAVE is the final LA_END that was just added. int32_t saveOp = (int32_t)fRXPat->fCompiledPat->elementAti(fMatchOpenParen); U_ASSERT(URX_TYPE(saveOp) == URX_STATE_SAVE); int32_t dest = fRXPat->fCompiledPat->size()-1; saveOp = URX_BUILD(URX_STATE_SAVE, dest); fRXPat->fCompiledPat->setElementAt(saveOp, fMatchOpenParen); } break; case lookBehind: { // See comment at doOpenLookBehind. // Append the URX_LB_END and URX_LA_END to the compiled pattern. int32_t startOp = (int32_t)fRXPat->fCompiledPat->elementAti(fMatchOpenParen-4); U_ASSERT(URX_TYPE(startOp) == URX_LB_START); int32_t dataLoc = URX_VAL(startOp); int32_t op = URX_BUILD(URX_LB_END, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); op = URX_BUILD(URX_LA_END, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); // Determine the min and max bounds for the length of the // string that the pattern can match. // An unbounded upper limit is an error. int32_t patEnd = fRXPat->fCompiledPat->size() - 1; int32_t minML = minMatchLength(fMatchOpenParen, patEnd); int32_t maxML = maxMatchLength(fMatchOpenParen, patEnd); if (maxML == INT32_MAX) { error(U_REGEX_LOOK_BEHIND_LIMIT); break; } U_ASSERT(minML <= maxML); // Insert the min and max match len bounds into the URX_LB_CONT op that // appears at the top of the look-behind block, at location fMatchOpenParen+1 fRXPat->fCompiledPat->setElementAt(minML, fMatchOpenParen-2); fRXPat->fCompiledPat->setElementAt(maxML, fMatchOpenParen-1); } break; case lookBehindN: { // See comment at doOpenLookBehindNeg. // Append the URX_LBN_END to the compiled pattern. int32_t startOp = (int32_t)fRXPat->fCompiledPat->elementAti(fMatchOpenParen-5); U_ASSERT(URX_TYPE(startOp) == URX_LB_START); int32_t dataLoc = URX_VAL(startOp); int32_t op = URX_BUILD(URX_LBN_END, dataLoc); fRXPat->fCompiledPat->addElement(op, *fStatus); // Determine the min and max bounds for the length of the // string that the pattern can match. // An unbounded upper limit is an error. int32_t patEnd = fRXPat->fCompiledPat->size() - 1; int32_t minML = minMatchLength(fMatchOpenParen, patEnd); int32_t maxML = maxMatchLength(fMatchOpenParen, patEnd); if (maxML == INT32_MAX) { error(U_REGEX_LOOK_BEHIND_LIMIT); break; } U_ASSERT(minML <= maxML); // Insert the min and max match len bounds into the URX_LB_CONT op that // appears at the top of the look-behind block, at location fMatchOpenParen+1 fRXPat->fCompiledPat->setElementAt(minML, fMatchOpenParen-3); fRXPat->fCompiledPat->setElementAt(maxML, fMatchOpenParen-2); // Insert the pattern location to continue at after a successful match // as the last operand of the URX_LBN_CONT op = URX_BUILD(URX_RELOC_OPRND, fRXPat->fCompiledPat->size()); fRXPat->fCompiledPat->setElementAt(op, fMatchOpenParen-1); } break; default: U_ASSERT(FALSE); } // remember the next location in the compiled pattern. // The compilation of Quantifiers will look at this to see whether its looping // over a parenthesized block or a single item fMatchCloseParen = fRXPat->fCompiledPat->size(); } //------------------------------------------------------------------------------ // // compileSet Compile the pattern operations for a reference to a // UnicodeSet. // //------------------------------------------------------------------------------ void RegexCompile::compileSet(UnicodeSet *theSet) { if (theSet == NULL) { return; } // Remove any strings from the set. // There shoudn't be any, but just in case. // (Case Closure can add them; if we had a simple case closure avaialble that // ignored strings, that would be better.) theSet->removeAllStrings(); int32_t setSize = theSet->size(); switch (setSize) { case 0: { // Set of no elements. Always fails to match. fRXPat->fCompiledPat->addElement(URX_BUILD(URX_BACKTRACK, 0), *fStatus); delete theSet; } break; case 1: { // The set contains only a single code point. Put it into // the compiled pattern as a single char operation rather // than a set, and discard the set itself. literalChar(theSet->charAt(0)); delete theSet; } break; default: { // The set contains two or more chars. (the normal case) // Put it into the compiled pattern as a set. int32_t setNumber = fRXPat->fSets->size(); fRXPat->fSets->addElement(theSet, *fStatus); int32_t setOp = URX_BUILD(URX_SETREF, setNumber); fRXPat->fCompiledPat->addElement(setOp, *fStatus); } } } //------------------------------------------------------------------------------ // // compileInterval Generate the code for a {min, max} style interval quantifier. // Except for the specific opcodes used, the code is the same // for all three types (greedy, non-greedy, possessive) of // intervals. The opcodes are supplied as parameters. // // The code for interval loops has this form: // 0 CTR_INIT counter loc (in stack frame) // 1 5 patt address of CTR_LOOP at bottom of block // 2 min count // 3 max count (-1 for unbounded) // 4 ... block to be iterated over // 5 CTR_LOOP // // In //------------------------------------------------------------------------------ void RegexCompile::compileInterval(int32_t InitOp, int32_t LoopOp) { // The CTR_INIT op at the top of the block with the {n,m} quantifier takes // four slots in the compiled code. Reserve them. int32_t topOfBlock = blockTopLoc(TRUE); insertOp(topOfBlock); insertOp(topOfBlock); insertOp(topOfBlock); // The operands for the CTR_INIT opcode include the index in the matcher data // of the counter. Allocate it now. int32_t counterLoc = fRXPat->fFrameSize; fRXPat->fFrameSize++; int32_t op = URX_BUILD(InitOp, counterLoc); fRXPat->fCompiledPat->setElementAt(op, topOfBlock); // The second operand of CTR_INIT is the location following the end of the loop. // Must put in as a URX_RELOC_OPRND so that the value will be adjusted if the // compilation of something later on causes the code to grow and the target // position to move. int32_t loopEnd = fRXPat->fCompiledPat->size(); op = URX_BUILD(URX_RELOC_OPRND, loopEnd); fRXPat->fCompiledPat->setElementAt(op, topOfBlock+1); // Followed by the min and max counts. fRXPat->fCompiledPat->setElementAt(fIntervalLow, topOfBlock+2); fRXPat->fCompiledPat->setElementAt(fIntervalUpper, topOfBlock+3); // Apend the CTR_LOOP op. The operand is the location of the CTR_INIT op. // Goes at end of the block being looped over, so just append to the code so far. op = URX_BUILD(LoopOp, topOfBlock); fRXPat->fCompiledPat->addElement(op, *fStatus); if ((fIntervalLow & 0xff000000) != 0 || (fIntervalUpper > 0 && (fIntervalUpper & 0xff000000) != 0)) { error(U_REGEX_NUMBER_TOO_BIG); } if (fIntervalLow > fIntervalUpper && fIntervalUpper != -1) { error(U_REGEX_MAX_LT_MIN); } } UBool RegexCompile::compileInlineInterval() { if (fIntervalUpper > 10 || fIntervalUpper < fIntervalLow) { // Too big to inline. Fail, which will cause looping code to be generated. // (Upper < Lower picks up unbounded upper and errors, both.) return FALSE; } int32_t topOfBlock = blockTopLoc(FALSE); if (fIntervalUpper == 0) { // Pathological case. Attempt no matches, as if the block doesn't exist. fRXPat->fCompiledPat->setSize(topOfBlock); return TRUE; } if (topOfBlock != fRXPat->fCompiledPat->size()-1 && fIntervalUpper != 1) { // The thing being repeated is not a single op, but some // more complex block. Do it as a loop, not inlines. // Note that things "repeated" a max of once are handled as inline, because // the one copy of the code already generated is just fine. return FALSE; } // Pick up the opcode that is to be repeated // int32_t op = (int32_t)fRXPat->fCompiledPat->elementAti(topOfBlock); // Compute the pattern location where the inline sequence // will end, and set up the state save op that will be needed. // int32_t endOfSequenceLoc = fRXPat->fCompiledPat->size()-1 + fIntervalUpper + (fIntervalUpper-fIntervalLow); int32_t saveOp = URX_BUILD(URX_STATE_SAVE, endOfSequenceLoc); if (fIntervalLow == 0) { insertOp(topOfBlock); fRXPat->fCompiledPat->setElementAt(saveOp, topOfBlock); } // Loop, emitting the op for the thing being repeated each time. // Loop starts at 1 because one instance of the op already exists in the pattern, // it was put there when it was originally encountered. int32_t i; for (i=1; ifCompiledPat->addElement(saveOp, *fStatus); } if (i > fIntervalLow) { fRXPat->fCompiledPat->addElement(saveOp, *fStatus); } fRXPat->fCompiledPat->addElement(op, *fStatus); } return TRUE; } //------------------------------------------------------------------------------ // // matchStartType Determine how a match can start. // Used to optimize find() operations. // // Operation is very similar to minMatchLength(). Walk the compiled // pattern, keeping an on-going minimum-match-length. For any // op where the min match coming in is zero, add that ops possible // starting matches to the possible starts for the overall pattern. // //------------------------------------------------------------------------------ void RegexCompile::matchStartType() { if (U_FAILURE(*fStatus)) { return; } int32_t loc; // Location in the pattern of the current op being processed. int32_t op; // The op being processed int32_t opType; // The opcode type of the op int32_t currentLen = 0; // Minimum length of a match to this point (loc) in the pattern int32_t numInitialStrings = 0; // Number of strings encountered that could match at start. UBool atStart = TRUE; // True if no part of the pattern yet encountered // could have advanced the position in a match. // (Maximum match length so far == 0) // forwardedLength is a vector holding minimum-match-length values that // are propagated forward in the pattern by JMP or STATE_SAVE operations. // It must be one longer than the pattern being checked because some ops // will jmp to a end-of-block+1 location from within a block, and we must // count those when checking the block. int32_t end = fRXPat->fCompiledPat->size(); UVector32 forwardedLength(end+1, *fStatus); forwardedLength.setSize(end+1); for (loc=3; locfCompiledPat->elementAti(loc); opType = URX_TYPE(op); // The loop is advancing linearly through the pattern. // If the op we are now at was the destination of a branch in the pattern, // and that path has a shorter minimum length than the current accumulated value, // replace the current accumulated value. if (forwardedLength.elementAti(loc) < currentLen) { currentLen = forwardedLength.elementAti(loc); U_ASSERT(currentLen>=0 && currentLen < INT32_MAX); } switch (opType) { // Ops that don't change the total length matched case URX_RESERVED_OP: case URX_END: case URX_FAIL: case URX_STRING_LEN: case URX_NOP: case URX_START_CAPTURE: case URX_END_CAPTURE: case URX_BACKSLASH_B: case URX_BACKSLASH_BU: case URX_BACKSLASH_G: case URX_BACKSLASH_Z: case URX_DOLLAR: case URX_DOLLAR_M: case URX_DOLLAR_D: case URX_DOLLAR_MD: case URX_RELOC_OPRND: case URX_STO_INP_LOC: case URX_BACKREF: // BackRef. Must assume that it might be a zero length match case URX_BACKREF_I: case URX_STO_SP: // Setup for atomic or possessive blocks. Doesn't change what can match. case URX_LD_SP: break; case URX_CARET: if (atStart) { fRXPat->fStartType = START_START; } break; case URX_CARET_M: case URX_CARET_M_UNIX: if (atStart) { fRXPat->fStartType = START_LINE; } break; case URX_ONECHAR: if (currentLen == 0) { // This character could appear at the start of a match. // Add it to the set of possible starting characters. fRXPat->fInitialChars->add(URX_VAL(op)); numInitialStrings += 2; } currentLen++; atStart = FALSE; break; case URX_SETREF: if (currentLen == 0) { int32_t sn = URX_VAL(op); U_ASSERT(sn > 0 && sn < fRXPat->fSets->size()); const UnicodeSet *s = (UnicodeSet *)fRXPat->fSets->elementAt(sn); fRXPat->fInitialChars->addAll(*s); numInitialStrings += 2; } currentLen++; atStart = FALSE; break; case URX_LOOP_SR_I: // [Set]*, like a SETREF, above, in what it can match, // but may not match at all, so currentLen is not incremented. if (currentLen == 0) { int32_t sn = URX_VAL(op); U_ASSERT(sn > 0 && sn < fRXPat->fSets->size()); const UnicodeSet *s = (UnicodeSet *)fRXPat->fSets->elementAt(sn); fRXPat->fInitialChars->addAll(*s); numInitialStrings += 2; } atStart = FALSE; break; case URX_LOOP_DOT_I: if (currentLen == 0) { // .* at the start of a pattern. // Any character can begin the match. fRXPat->fInitialChars->clear(); fRXPat->fInitialChars->complement(); numInitialStrings += 2; } atStart = FALSE; break; case URX_STATIC_SETREF: if (currentLen == 0) { int32_t sn = URX_VAL(op); U_ASSERT(sn>0 && snfStaticSets[sn]; fRXPat->fInitialChars->addAll(*s); numInitialStrings += 2; } currentLen++; atStart = FALSE; break; case URX_STAT_SETREF_N: if (currentLen == 0) { int32_t sn = URX_VAL(op); const UnicodeSet *s = fRXPat->fStaticSets[sn]; UnicodeSet sc(*s); sc.complement(); fRXPat->fInitialChars->addAll(sc); numInitialStrings += 2; } currentLen++; atStart = FALSE; break; case URX_BACKSLASH_D: // Digit Char if (currentLen == 0) { UnicodeSet s; s.applyIntPropertyValue(UCHAR_GENERAL_CATEGORY_MASK, U_GC_ND_MASK, *fStatus); if (URX_VAL(op) != 0) { s.complement(); } fRXPat->fInitialChars->addAll(s); numInitialStrings += 2; } currentLen++; atStart = FALSE; break; case URX_ONECHAR_I: // Case Insensitive Single Character. if (currentLen == 0) { UChar32 c = URX_VAL(op); if (u_hasBinaryProperty(c, UCHAR_CASE_SENSITIVE)) { // Disable optimizations on first char of match. // TODO: Compute the set of chars that case fold to this char, or to // a string that begins with this char. // For simple case folding, this code worked: // UnicodeSet s(c, c); // s.closeOver(USET_CASE_INSENSITIVE); // fRXPat->fInitialChars->addAll(s); fRXPat->fInitialChars->clear(); fRXPat->fInitialChars->complement(); } else { // Char has no case variants. Just add it as-is to the // set of possible starting chars. fRXPat->fInitialChars->add(c); } numInitialStrings += 2; } currentLen++; atStart = FALSE; break; case URX_BACKSLASH_X: // Grahpeme Cluster. Minimum is 1, max unbounded. case URX_DOTANY_ALL: // . matches one or two. case URX_DOTANY: case URX_DOTANY_UNIX: if (currentLen == 0) { // These constructs are all bad news when they appear at the start // of a match. Any character can begin the match. fRXPat->fInitialChars->clear(); fRXPat->fInitialChars->complement(); numInitialStrings += 2; } currentLen++; atStart = FALSE; break; case URX_JMPX: loc++; // Except for extra operand on URX_JMPX, same as URX_JMP. case URX_JMP: { int32_t jmpDest = URX_VAL(op); if (jmpDest < loc) { // Loop of some kind. Can safely ignore, the worst that will happen // is that we understate the true minimum length currentLen = forwardedLength.elementAti(loc+1); } else { // Forward jump. Propagate the current min length to the target loc of the jump. U_ASSERT(jmpDest <= end+1); if (forwardedLength.elementAti(jmpDest) > currentLen) { forwardedLength.setElementAt(currentLen, jmpDest); } } } atStart = FALSE; break; case URX_JMP_SAV: case URX_JMP_SAV_X: // Combo of state save to the next loc, + jmp backwards. // Net effect on min. length computation is nothing. atStart = FALSE; break; case URX_BACKTRACK: // Fails are kind of like a branch, except that the min length was // propagated already, by the state save. currentLen = forwardedLength.elementAti(loc+1); atStart = FALSE; break; case URX_STATE_SAVE: { // State Save, for forward jumps, propagate the current minimum. // of the state save. int32_t jmpDest = URX_VAL(op); if (jmpDest > loc) { if (currentLen < forwardedLength.elementAti(jmpDest)) { forwardedLength.setElementAt(currentLen, jmpDest); } } } atStart = FALSE; break; case URX_STRING: { loc++; int32_t stringLenOp = (int32_t)fRXPat->fCompiledPat->elementAti(loc); int32_t stringLen = URX_VAL(stringLenOp); U_ASSERT(URX_TYPE(stringLenOp) == URX_STRING_LEN); U_ASSERT(stringLenOp >= 2); if (currentLen == 0) { // Add the starting character of this string to the set of possible starting // characters for this pattern. int32_t stringStartIdx = URX_VAL(op); UChar32 c = fRXPat->fLiteralText.char32At(stringStartIdx); fRXPat->fInitialChars->add(c); // Remember this string. After the entire pattern has been checked, // if nothing else is identified that can start a match, we'll use it. numInitialStrings++; fRXPat->fInitialStringIdx = stringStartIdx; fRXPat->fInitialStringLen = stringLen; } currentLen += stringLen; atStart = FALSE; } break; case URX_STRING_I: { // Case-insensitive string. Unlike exact-match strings, we won't // attempt a string search for possible match positions. But we // do update the set of possible starting characters. loc++; int32_t stringLenOp = (int32_t)fRXPat->fCompiledPat->elementAti(loc); int32_t stringLen = URX_VAL(stringLenOp); U_ASSERT(URX_TYPE(stringLenOp) == URX_STRING_LEN); U_ASSERT(stringLenOp >= 2); if (currentLen == 0) { // Add the starting character of this string to the set of possible starting // characters for this pattern. int32_t stringStartIdx = URX_VAL(op); UChar32 c = fRXPat->fLiteralText.char32At(stringStartIdx); UnicodeSet s(c, c); // TODO: compute correct set of starting chars for full case folding. // For the moment, say any char can start. // s.closeOver(USET_CASE_INSENSITIVE); s.clear(); s.complement(); fRXPat->fInitialChars->addAll(s); numInitialStrings += 2; // Matching on an initial string not possible. } currentLen += stringLen; atStart = FALSE; } break; case URX_CTR_INIT: case URX_CTR_INIT_NG: { // Loop Init Ops. These don't change the min length, but they are 4 word ops // so location must be updated accordingly. // Loop Init Ops. // If the min loop count == 0 // move loc forwards to the end of the loop, skipping over the body. // If the min count is > 0, // continue normal processing of the body of the loop. int32_t loopEndLoc = (int32_t)fRXPat->fCompiledPat->elementAti(loc+1); loopEndLoc = URX_VAL(loopEndLoc); int32_t minLoopCount = (int32_t)fRXPat->fCompiledPat->elementAti(loc+2); if (minLoopCount == 0) { // Min Loop Count of 0, treat like a forward branch and // move the current minimum length up to the target // (end of loop) location. U_ASSERT(loopEndLoc <= end+1); if (forwardedLength.elementAti(loopEndLoc) > currentLen) { forwardedLength.setElementAt(currentLen, loopEndLoc); } } loc+=3; // Skips over operands of CTR_INIT } atStart = FALSE; break; case URX_CTR_LOOP: case URX_CTR_LOOP_NG: // Loop ops. // The jump is conditional, backwards only. atStart = FALSE; break; case URX_LOOP_C: // More loop ops. These state-save to themselves. // don't change the minimum match atStart = FALSE; break; case URX_LA_START: case URX_LB_START: { // Look-around. Scan forward until the matching look-ahead end, // without processing the look-around block. This is overly pessimistic. // Keep track of the nesting depth of look-around blocks. Boilerplate code for // lookahead contains two LA_END instructions, so count goes up by two // for each LA_START. int32_t depth = (opType == URX_LA_START? 2: 1); for (;;) { loc++; op = (int32_t)fRXPat->fCompiledPat->elementAti(loc); if (URX_TYPE(op) == URX_LA_START) { depth+=2; } if (URX_TYPE(op) == URX_LB_START) { depth++; } if (URX_TYPE(op) == URX_LA_END || URX_TYPE(op)==URX_LBN_END) { depth--; if (depth == 0) { break; } } if (URX_TYPE(op) == URX_STATE_SAVE) { // Need this because neg lookahead blocks will FAIL to outside // of the block. int32_t jmpDest = URX_VAL(op); if (jmpDest > loc) { if (currentLen < forwardedLength.elementAti(jmpDest)) { forwardedLength.setElementAt(currentLen, jmpDest); } } } U_ASSERT(loc <= end); } } break; case URX_LA_END: case URX_LB_CONT: case URX_LB_END: case URX_LBN_CONT: case URX_LBN_END: U_ASSERT(FALSE); // Shouldn't get here. These ops should be // consumed by the scan in URX_LA_START and LB_START break; default: U_ASSERT(FALSE); } } // We have finished walking through the ops. Check whether some forward jump // propagated a shorter length to location end+1. if (forwardedLength.elementAti(end+1) < currentLen) { currentLen = forwardedLength.elementAti(end+1); } fRXPat->fInitialChars8->init(fRXPat->fInitialChars); // Sort out what we should check for when looking for candidate match start positions. // In order of preference, // 1. Start of input text buffer. // 2. A literal string. // 3. Start of line in multi-line mode. // 4. A single literal character. // 5. A character from a set of characters. // if (fRXPat->fStartType == START_START) { // Match only at the start of an input text string. // start type is already set. We're done. } else if (numInitialStrings == 1 && fRXPat->fMinMatchLen > 0) { // Match beginning only with a literal string. UChar32 c = fRXPat->fLiteralText.char32At(fRXPat->fInitialStringIdx); U_ASSERT(fRXPat->fInitialChars->contains(c)); fRXPat->fStartType = START_STRING; fRXPat->fInitialChar = c; } else if (fRXPat->fStartType == START_LINE) { // Match at start of line in Multi-Line mode. // Nothing to do here; everything is already set. } else if (fRXPat->fMinMatchLen == 0) { // Zero length match possible. We could start anywhere. fRXPat->fStartType = START_NO_INFO; } else if (fRXPat->fInitialChars->size() == 1) { // All matches begin with the same char. fRXPat->fStartType = START_CHAR; fRXPat->fInitialChar = fRXPat->fInitialChars->charAt(0); U_ASSERT(fRXPat->fInitialChar != (UChar32)-1); } else if (fRXPat->fInitialChars->contains((UChar32)0, (UChar32)0x10ffff) == FALSE && fRXPat->fMinMatchLen > 0) { // Matches start with a set of character smaller than the set of all chars. fRXPat->fStartType = START_SET; } else { // Matches can start with anything fRXPat->fStartType = START_NO_INFO; } return; } //------------------------------------------------------------------------------ // // minMatchLength Calculate the length of the shortest string that could // match the specified pattern. // Length is in 16 bit code units, not code points. // // The calculated length may not be exact. The returned // value may be shorter than the actual minimum; it must // never be longer. // // start and end are the range of p-code operations to be // examined. The endpoints are included in the range. // //------------------------------------------------------------------------------ int32_t RegexCompile::minMatchLength(int32_t start, int32_t end) { if (U_FAILURE(*fStatus)) { return 0; } U_ASSERT(start <= end); U_ASSERT(end < fRXPat->fCompiledPat->size()); int32_t loc; int32_t op; int32_t opType; int32_t currentLen = 0; // forwardedLength is a vector holding minimum-match-length values that // are propagated forward in the pattern by JMP or STATE_SAVE operations. // It must be one longer than the pattern being checked because some ops // will jmp to a end-of-block+1 location from within a block, and we must // count those when checking the block. UVector32 forwardedLength(end+2, *fStatus); forwardedLength.setSize(end+2); for (loc=start; loc<=end+1; loc++) { forwardedLength.setElementAt(INT32_MAX, loc); } for (loc = start; loc<=end; loc++) { op = (int32_t)fRXPat->fCompiledPat->elementAti(loc); opType = URX_TYPE(op); // The loop is advancing linearly through the pattern. // If the op we are now at was the destination of a branch in the pattern, // and that path has a shorter minimum length than the current accumulated value, // replace the current accumulated value. // U_ASSERT(currentLen>=0 && currentLen < INT32_MAX); // MinLength == INT32_MAX for some // no-match-possible cases. if (forwardedLength.elementAti(loc) < currentLen) { currentLen = forwardedLength.elementAti(loc); U_ASSERT(currentLen>=0 && currentLen < INT32_MAX); } switch (opType) { // Ops that don't change the total length matched case URX_RESERVED_OP: case URX_END: case URX_STRING_LEN: case URX_NOP: case URX_START_CAPTURE: case URX_END_CAPTURE: case URX_BACKSLASH_B: case URX_BACKSLASH_BU: case URX_BACKSLASH_G: case URX_BACKSLASH_Z: case URX_CARET: case URX_DOLLAR: case URX_DOLLAR_M: case URX_DOLLAR_D: case URX_DOLLAR_MD: case URX_RELOC_OPRND: case URX_STO_INP_LOC: case URX_CARET_M: case URX_CARET_M_UNIX: case URX_BACKREF: // BackRef. Must assume that it might be a zero length match case URX_BACKREF_I: case URX_STO_SP: // Setup for atomic or possessive blocks. Doesn't change what can match. case URX_LD_SP: case URX_JMP_SAV: case URX_JMP_SAV_X: break; // Ops that match a minimum of one character (one or two 16 bit code units.) // case URX_ONECHAR: case URX_STATIC_SETREF: case URX_STAT_SETREF_N: case URX_SETREF: case URX_BACKSLASH_D: case URX_ONECHAR_I: case URX_BACKSLASH_X: // Grahpeme Cluster. Minimum is 1, max unbounded. case URX_DOTANY_ALL: // . matches one or two. case URX_DOTANY: case URX_DOTANY_UNIX: currentLen++; break; case URX_JMPX: loc++; // URX_JMPX has an extra operand, ignored here, // otherwise processed identically to URX_JMP. case URX_JMP: { int32_t jmpDest = URX_VAL(op); if (jmpDest < loc) { // Loop of some kind. Can safely ignore, the worst that will happen // is that we understate the true minimum length currentLen = forwardedLength.elementAti(loc+1); } else { // Forward jump. Propagate the current min length to the target loc of the jump. U_ASSERT(jmpDest <= end+1); if (forwardedLength.elementAti(jmpDest) > currentLen) { forwardedLength.setElementAt(currentLen, jmpDest); } } } break; case URX_BACKTRACK: { // Back-tracks are kind of like a branch, except that the min length was // propagated already, by the state save. currentLen = forwardedLength.elementAti(loc+1); } break; case URX_STATE_SAVE: { // State Save, for forward jumps, propagate the current minimum. // of the state save. int32_t jmpDest = URX_VAL(op); if (jmpDest > loc) { if (currentLen < forwardedLength.elementAti(jmpDest)) { forwardedLength.setElementAt(currentLen, jmpDest); } } } break; case URX_STRING: { loc++; int32_t stringLenOp = (int32_t)fRXPat->fCompiledPat->elementAti(loc); currentLen += URX_VAL(stringLenOp); } break; case URX_STRING_I: { loc++; // TODO: with full case folding, matching input text may be shorter than // the string we have here. More smarts could put some bounds on it. // Assume a min length of one for now. A min length of zero causes // optimization failures for a pattern like "string"+ // currentLen += URX_VAL(stringLenOp); currentLen += 1; } break; case URX_CTR_INIT: case URX_CTR_INIT_NG: { // Loop Init Ops. // If the min loop count == 0 // move loc forwards to the end of the loop, skipping over the body. // If the min count is > 0, // continue normal processing of the body of the loop. int32_t loopEndLoc = (int32_t)fRXPat->fCompiledPat->elementAti(loc+1); loopEndLoc = URX_VAL(loopEndLoc); int32_t minLoopCount = (int32_t)fRXPat->fCompiledPat->elementAti(loc+2); if (minLoopCount == 0) { loc = loopEndLoc; } else { loc+=3; // Skips over operands of CTR_INIT } } break; case URX_CTR_LOOP: case URX_CTR_LOOP_NG: // Loop ops. // The jump is conditional, backwards only. break; case URX_LOOP_SR_I: case URX_LOOP_DOT_I: case URX_LOOP_C: // More loop ops. These state-save to themselves. // don't change the minimum match - could match nothing at all. break; case URX_LA_START: case URX_LB_START: { // Look-around. Scan forward until the matching look-ahead end, // without processing the look-around block. This is overly pessimistic for look-ahead, // it assumes that the look-ahead match might be zero-length. // TODO: Positive lookahead could recursively do the block, then continue // with the longer of the block or the value coming in. Ticket 6060 int32_t depth = (opType == URX_LA_START? 2: 1);; for (;;) { loc++; op = (int32_t)fRXPat->fCompiledPat->elementAti(loc); if (URX_TYPE(op) == URX_LA_START) { // The boilerplate for look-ahead includes two LA_END insturctions, // Depth will be decremented by each one when it is seen. depth += 2; } if (URX_TYPE(op) == URX_LB_START) { depth++; } if (URX_TYPE(op) == URX_LA_END) { depth--; if (depth == 0) { break; } } if (URX_TYPE(op)==URX_LBN_END) { depth--; if (depth == 0) { break; } } if (URX_TYPE(op) == URX_STATE_SAVE) { // Need this because neg lookahead blocks will FAIL to outside // of the block. int32_t jmpDest = URX_VAL(op); if (jmpDest > loc) { if (currentLen < forwardedLength.elementAti(jmpDest)) { forwardedLength.setElementAt(currentLen, jmpDest); } } } U_ASSERT(loc <= end); } } break; case URX_LA_END: case URX_LB_CONT: case URX_LB_END: case URX_LBN_CONT: case URX_LBN_END: // Only come here if the matching URX_LA_START or URX_LB_START was not in the // range being sized, which happens when measuring size of look-behind blocks. break; default: U_ASSERT(FALSE); } } // We have finished walking through the ops. Check whether some forward jump // propagated a shorter length to location end+1. if (forwardedLength.elementAti(end+1) < currentLen) { currentLen = forwardedLength.elementAti(end+1); U_ASSERT(currentLen>=0 && currentLen < INT32_MAX); } return currentLen; } // Increment with overflow check. // val and delta will both be positive. static int32_t safeIncrement(int32_t val, int32_t delta) { if (INT32_MAX - val > delta) { return val + delta; } else { return INT32_MAX; } } //------------------------------------------------------------------------------ // // maxMatchLength Calculate the length of the longest string that could // match the specified pattern. // Length is in 16 bit code units, not code points. // // The calculated length may not be exact. The returned // value may be longer than the actual maximum; it must // never be shorter. // //------------------------------------------------------------------------------ int32_t RegexCompile::maxMatchLength(int32_t start, int32_t end) { if (U_FAILURE(*fStatus)) { return 0; } U_ASSERT(start <= end); U_ASSERT(end < fRXPat->fCompiledPat->size()); int32_t loc; int32_t op; int32_t opType; int32_t currentLen = 0; UVector32 forwardedLength(end+1, *fStatus); forwardedLength.setSize(end+1); for (loc=start; loc<=end; loc++) { forwardedLength.setElementAt(0, loc); } for (loc = start; loc<=end; loc++) { op = (int32_t)fRXPat->fCompiledPat->elementAti(loc); opType = URX_TYPE(op); // The loop is advancing linearly through the pattern. // If the op we are now at was the destination of a branch in the pattern, // and that path has a longer maximum length than the current accumulated value, // replace the current accumulated value. if (forwardedLength.elementAti(loc) > currentLen) { currentLen = forwardedLength.elementAti(loc); } switch (opType) { // Ops that don't change the total length matched case URX_RESERVED_OP: case URX_END: case URX_STRING_LEN: case URX_NOP: case URX_START_CAPTURE: case URX_END_CAPTURE: case URX_BACKSLASH_B: case URX_BACKSLASH_BU: case URX_BACKSLASH_G: case URX_BACKSLASH_Z: case URX_CARET: case URX_DOLLAR: case URX_DOLLAR_M: case URX_DOLLAR_D: case URX_DOLLAR_MD: case URX_RELOC_OPRND: case URX_STO_INP_LOC: case URX_CARET_M: case URX_CARET_M_UNIX: case URX_STO_SP: // Setup for atomic or possessive blocks. Doesn't change what can match. case URX_LD_SP: case URX_LB_END: case URX_LB_CONT: case URX_LBN_CONT: case URX_LBN_END: break; // Ops that increase that cause an unbounded increase in the length // of a matched string, or that increase it a hard to characterize way. // Call the max length unbounded, and stop further checking. case URX_BACKREF: // BackRef. Must assume that it might be a zero length match case URX_BACKREF_I: case URX_BACKSLASH_X: // Grahpeme Cluster. Minimum is 1, max unbounded. currentLen = INT32_MAX; break; // Ops that match a max of one character (possibly two 16 bit code units.) // case URX_STATIC_SETREF: case URX_STAT_SETREF_N: case URX_SETREF: case URX_BACKSLASH_D: case URX_ONECHAR_I: case URX_DOTANY_ALL: case URX_DOTANY: case URX_DOTANY_UNIX: currentLen = safeIncrement(currentLen, 2); break; // Single literal character. Increase current max length by one or two, // depending on whether the char is in the supplementary range. case URX_ONECHAR: currentLen = safeIncrement(currentLen, 1); if (URX_VAL(op) > 0x10000) { currentLen = safeIncrement(currentLen, 1); } break; // Jumps. // case URX_JMP: case URX_JMPX: case URX_JMP_SAV: case URX_JMP_SAV_X: { int32_t jmpDest = URX_VAL(op); if (jmpDest < loc) { // Loop of some kind. Max match length is unbounded. currentLen = INT32_MAX; } else { // Forward jump. Propagate the current min length to the target loc of the jump. if (forwardedLength.elementAti(jmpDest) < currentLen) { forwardedLength.setElementAt(currentLen, jmpDest); } currentLen = 0; } } break; case URX_BACKTRACK: // back-tracks are kind of like a branch, except that the max length was // propagated already, by the state save. currentLen = forwardedLength.elementAti(loc+1); break; case URX_STATE_SAVE: { // State Save, for forward jumps, propagate the current minimum. // of the state save. // For backwards jumps, they create a loop, maximum // match length is unbounded. int32_t jmpDest = URX_VAL(op); if (jmpDest > loc) { if (currentLen > forwardedLength.elementAti(jmpDest)) { forwardedLength.setElementAt(currentLen, jmpDest); } } else { currentLen = INT32_MAX; } } break; case URX_STRING: { loc++; int32_t stringLenOp = (int32_t)fRXPat->fCompiledPat->elementAti(loc); currentLen = safeIncrement(currentLen, URX_VAL(stringLenOp)); break; } case URX_STRING_I: // TODO: This code assumes that any user string that matches will be no longer // than our compiled string, with case insensitive matching. // Our compiled string has been case-folded already. // // Any matching user string will have no more code points than our // compiled (folded) string. Folding may add code points, but // not remove them. // // There is a potential problem if a supplemental code point // case-folds to a BMP code point. In this case our compiled string // could be shorter (in code units) than a matching user string. // // At this time (Unicode 6.1) there are no such characters, and this case // is not being handled. A test, intltest regex/Bug9283, will fail if // any problematic characters are added to Unicode. // // If this happens, we can make a set of the BMP chars that the // troublesome supplementals fold to, scan our string, and bump the // currentLen one extra for each that is found. // { loc++; int32_t stringLenOp = (int32_t)fRXPat->fCompiledPat->elementAti(loc); currentLen = safeIncrement(currentLen, URX_VAL(stringLenOp)); } break; case URX_CTR_INIT: case URX_CTR_INIT_NG: // For Loops, recursively call this function on the pattern for the loop body, // then multiply the result by the maximum loop count. { int32_t loopEndLoc = URX_VAL(fRXPat->fCompiledPat->elementAti(loc+1)); if (loopEndLoc == loc+4) { // Loop has an empty body. No affect on max match length. // Continue processing with code after the loop end. loc = loopEndLoc; break; } int32_t maxLoopCount = fRXPat->fCompiledPat->elementAti(loc+3); if (maxLoopCount == -1) { // Unbounded Loop. No upper bound on match length. currentLen = INT32_MAX; break; } U_ASSERT(loopEndLoc >= loc+4); int32_t blockLen = maxMatchLength(loc+4, loopEndLoc-1); // Recursive call. if (blockLen == INT32_MAX) { currentLen = blockLen; break; } currentLen += blockLen * maxLoopCount; loc = loopEndLoc; break; } case URX_CTR_LOOP: case URX_CTR_LOOP_NG: // These opcodes will be skipped over by code for URX_CRT_INIT. // We shouldn't encounter them here. U_ASSERT(FALSE); break; case URX_LOOP_SR_I: case URX_LOOP_DOT_I: case URX_LOOP_C: // For anything to do with loops, make the match length unbounded. currentLen = INT32_MAX; break; case URX_LA_START: case URX_LA_END: // Look-ahead. Just ignore, treat the look-ahead block as if // it were normal pattern. Gives a too-long match length, // but good enough for now. break; // End of look-ahead ops should always be consumed by the processing at // the URX_LA_START op. // U_ASSERT(FALSE); // break; case URX_LB_START: { // Look-behind. Scan forward until the matching look-around end, // without processing the look-behind block. int32_t depth = 0; for (;;) { loc++; op = (int32_t)fRXPat->fCompiledPat->elementAti(loc); if (URX_TYPE(op) == URX_LA_START || URX_TYPE(op) == URX_LB_START) { depth++; } if (URX_TYPE(op) == URX_LA_END || URX_TYPE(op)==URX_LBN_END) { if (depth == 0) { break; } depth--; } U_ASSERT(loc < end); } } break; default: U_ASSERT(FALSE); } if (currentLen == INT32_MAX) { // The maximum length is unbounded. // Stop further processing of the pattern. break; } } return currentLen; } //------------------------------------------------------------------------------ // // stripNOPs Remove any NOP operations from the compiled pattern code. // Extra NOPs are inserted for some constructs during the initial // code generation to provide locations that may be patched later. // Many end up unneeded, and are removed by this function. // // In order to minimize the number of passes through the pattern, // back-reference fixup is also performed here (adjusting // back-reference operands to point to the correct frame offsets). // //------------------------------------------------------------------------------ void RegexCompile::stripNOPs() { if (U_FAILURE(*fStatus)) { return; } int32_t end = fRXPat->fCompiledPat->size(); UVector32 deltas(end, *fStatus); // Make a first pass over the code, computing the amount that things // will be offset at each location in the original code. int32_t loc; int32_t d = 0; for (loc=0; locfCompiledPat->elementAti(loc); if (URX_TYPE(op) == URX_NOP) { d++; } } UnicodeString caseStringBuffer; // Make a second pass over the code, removing the NOPs by moving following // code up, and patching operands that refer to code locations that // are being moved. The array of offsets from the first step is used // to compute the new operand values. int32_t src; int32_t dst = 0; for (src=0; srcfCompiledPat->elementAti(src); int32_t opType = URX_TYPE(op); switch (opType) { case URX_NOP: break; case URX_STATE_SAVE: case URX_JMP: case URX_CTR_LOOP: case URX_CTR_LOOP_NG: case URX_RELOC_OPRND: case URX_JMPX: case URX_JMP_SAV: case URX_JMP_SAV_X: // These are instructions with operands that refer to code locations. { int32_t operandAddress = URX_VAL(op); U_ASSERT(operandAddress>=0 && operandAddressfCompiledPat->setElementAt(op, dst); dst++; break; } case URX_BACKREF: case URX_BACKREF_I: { int32_t where = URX_VAL(op); if (where > fRXPat->fGroupMap->size()) { error(U_REGEX_INVALID_BACK_REF); break; } where = fRXPat->fGroupMap->elementAti(where-1); op = URX_BUILD(opType, where); fRXPat->fCompiledPat->setElementAt(op, dst); dst++; fRXPat->fNeedsAltInput = TRUE; break; } case URX_RESERVED_OP: case URX_RESERVED_OP_N: case URX_BACKTRACK: case URX_END: case URX_ONECHAR: case URX_STRING: case URX_STRING_LEN: case URX_START_CAPTURE: case URX_END_CAPTURE: case URX_STATIC_SETREF: case URX_STAT_SETREF_N: case URX_SETREF: case URX_DOTANY: case URX_FAIL: case URX_BACKSLASH_B: case URX_BACKSLASH_BU: case URX_BACKSLASH_G: case URX_BACKSLASH_X: case URX_BACKSLASH_Z: case URX_DOTANY_ALL: case URX_BACKSLASH_D: case URX_CARET: case URX_DOLLAR: case URX_CTR_INIT: case URX_CTR_INIT_NG: case URX_DOTANY_UNIX: case URX_STO_SP: case URX_LD_SP: case URX_STO_INP_LOC: case URX_LA_START: case URX_LA_END: case URX_ONECHAR_I: case URX_STRING_I: case URX_DOLLAR_M: case URX_CARET_M: case URX_CARET_M_UNIX: case URX_LB_START: case URX_LB_CONT: case URX_LB_END: case URX_LBN_CONT: case URX_LBN_END: case URX_LOOP_SR_I: case URX_LOOP_DOT_I: case URX_LOOP_C: case URX_DOLLAR_D: case URX_DOLLAR_MD: // These instructions are unaltered by the relocation. fRXPat->fCompiledPat->setElementAt(op, dst); dst++; break; default: // Some op is unaccounted for. U_ASSERT(FALSE); error(U_REGEX_INTERNAL_ERROR); } } fRXPat->fCompiledPat->setSize(dst); } //------------------------------------------------------------------------------ // // Error Report a rule parse error. // Only report it if no previous error has been recorded. // //------------------------------------------------------------------------------ void RegexCompile::error(UErrorCode e) { if (U_SUCCESS(*fStatus)) { *fStatus = e; // Hmm. fParseErr (UParseError) line & offset fields are int32_t in public // API (see common/unicode/parseerr.h), while fLineNum and fCharNum are // int64_t. If the values of the latter are out of range for the former, // set them to the appropriate "field not supported" values. if (fLineNum > 0x7FFFFFFF) { fParseErr->line = 0; fParseErr->offset = -1; } else if (fCharNum > 0x7FFFFFFF) { fParseErr->line = (int32_t)fLineNum; fParseErr->offset = -1; } else { fParseErr->line = (int32_t)fLineNum; fParseErr->offset = (int32_t)fCharNum; } UErrorCode status = U_ZERO_ERROR; // throwaway status for extracting context // Fill in the context. // Note: extractBetween() pins supplied indicies to the string bounds. uprv_memset(fParseErr->preContext, 0, sizeof(fParseErr->preContext)); uprv_memset(fParseErr->postContext, 0, sizeof(fParseErr->postContext)); utext_extract(fRXPat->fPattern, fScanIndex-U_PARSE_CONTEXT_LEN+1, fScanIndex, fParseErr->preContext, U_PARSE_CONTEXT_LEN, &status); utext_extract(fRXPat->fPattern, fScanIndex, fScanIndex+U_PARSE_CONTEXT_LEN-1, fParseErr->postContext, U_PARSE_CONTEXT_LEN, &status); } } // // Assorted Unicode character constants. // Numeric because there is no portable way to enter them as literals. // (Think EBCDIC). // static const UChar chCR = 0x0d; // New lines, for terminating comments. static const UChar chLF = 0x0a; // Line Feed static const UChar chPound = 0x23; // '#', introduces a comment. static const UChar chDigit0 = 0x30; // '0' static const UChar chDigit7 = 0x37; // '9' static const UChar chColon = 0x3A; // ':' static const UChar chE = 0x45; // 'E' static const UChar chQ = 0x51; // 'Q' //static const UChar chN = 0x4E; // 'N' static const UChar chP = 0x50; // 'P' static const UChar chBackSlash = 0x5c; // '\' introduces a char escape //static const UChar chLBracket = 0x5b; // '[' static const UChar chRBracket = 0x5d; // ']' static const UChar chUp = 0x5e; // '^' static const UChar chLowerP = 0x70; static const UChar chLBrace = 0x7b; // '{' static const UChar chRBrace = 0x7d; // '}' static const UChar chNEL = 0x85; // NEL newline variant static const UChar chLS = 0x2028; // Unicode Line Separator //------------------------------------------------------------------------------ // // nextCharLL Low Level Next Char from the regex pattern. // Get a char from the string, keep track of input position // for error reporting. // //------------------------------------------------------------------------------ UChar32 RegexCompile::nextCharLL() { UChar32 ch; if (fPeekChar != -1) { ch = fPeekChar; fPeekChar = -1; return ch; } // assume we're already in the right place ch = UTEXT_NEXT32(fRXPat->fPattern); if (ch == U_SENTINEL) { return ch; } if (ch == chCR || ch == chNEL || ch == chLS || (ch == chLF && fLastChar != chCR)) { // Character is starting a new line. Bump up the line number, and // reset the column to 0. fLineNum++; fCharNum=0; } else { // Character is not starting a new line. Except in the case of a // LF following a CR, increment the column position. if (ch != chLF) { fCharNum++; } } fLastChar = ch; return ch; } //------------------------------------------------------------------------------ // // peekCharLL Low Level Character Scanning, sneak a peek at the next // character without actually getting it. // //------------------------------------------------------------------------------ UChar32 RegexCompile::peekCharLL() { if (fPeekChar == -1) { fPeekChar = nextCharLL(); } return fPeekChar; } //------------------------------------------------------------------------------ // // nextChar for pattern scanning. At this level, we handle stripping // out comments and processing some backslash character escapes. // The rest of the pattern grammar is handled at the next level up. // //------------------------------------------------------------------------------ void RegexCompile::nextChar(RegexPatternChar &c) { fScanIndex = UTEXT_GETNATIVEINDEX(fRXPat->fPattern); c.fChar = nextCharLL(); c.fQuoted = FALSE; if (fQuoteMode) { c.fQuoted = TRUE; if ((c.fChar==chBackSlash && peekCharLL()==chE && ((fModeFlags & UREGEX_LITERAL) == 0)) || c.fChar == (UChar32)-1) { fQuoteMode = FALSE; // Exit quote mode, nextCharLL(); // discard the E nextChar(c); // recurse to get the real next char } } else if (fInBackslashQuote) { // The current character immediately follows a '\' // Don't check for any further escapes, just return it as-is. // Don't set c.fQuoted, because that would prevent the state machine from // dispatching on the character. fInBackslashQuote = FALSE; } else { // We are not in a \Q quoted region \E of the source. // if (fModeFlags & UREGEX_COMMENTS) { // // We are in free-spacing and comments mode. // Scan through any white space and comments, until we // reach a significant character or the end of inut. for (;;) { if (c.fChar == (UChar32)-1) { break; // End of Input } if (c.fChar == chPound && fEOLComments == TRUE) { // Start of a comment. Consume the rest of it, until EOF or a new line for (;;) { c.fChar = nextCharLL(); if (c.fChar == (UChar32)-1 || // EOF c.fChar == chCR || c.fChar == chLF || c.fChar == chNEL || c.fChar == chLS) { break; } } } // TODO: check what Java & Perl do with non-ASCII white spaces. Ticket 6061. if (PatternProps::isWhiteSpace(c.fChar) == FALSE) { break; } c.fChar = nextCharLL(); } } // // check for backslash escaped characters. // if (c.fChar == chBackSlash) { int64_t pos = UTEXT_GETNATIVEINDEX(fRXPat->fPattern); if (RegexStaticSets::gStaticSets->fUnescapeCharSet.contains(peekCharLL())) { // // A '\' sequence that is handled by ICU's standard unescapeAt function. // Includes \uxxxx, \n, \r, many others. // Return the single equivalent character. // nextCharLL(); // get & discard the peeked char. c.fQuoted = TRUE; if (UTEXT_FULL_TEXT_IN_CHUNK(fRXPat->fPattern, fPatternLength)) { int32_t endIndex = (int32_t)pos; c.fChar = u_unescapeAt(uregex_ucstr_unescape_charAt, &endIndex, (int32_t)fPatternLength, (void *)fRXPat->fPattern->chunkContents); if (endIndex == pos) { error(U_REGEX_BAD_ESCAPE_SEQUENCE); } fCharNum += endIndex - pos; UTEXT_SETNATIVEINDEX(fRXPat->fPattern, endIndex); } else { int32_t offset = 0; struct URegexUTextUnescapeCharContext context = U_REGEX_UTEXT_UNESCAPE_CONTEXT(fRXPat->fPattern); UTEXT_SETNATIVEINDEX(fRXPat->fPattern, pos); c.fChar = u_unescapeAt(uregex_utext_unescape_charAt, &offset, INT32_MAX, &context); if (offset == 0) { error(U_REGEX_BAD_ESCAPE_SEQUENCE); } else if (context.lastOffset == offset) { UTEXT_PREVIOUS32(fRXPat->fPattern); } else if (context.lastOffset != offset-1) { utext_moveIndex32(fRXPat->fPattern, offset - context.lastOffset - 1); } fCharNum += offset; } } else if (peekCharLL() == chDigit0) { // Octal Escape, using Java Regexp Conventions // which are \0 followed by 1-3 octal digits. // Different from ICU Unescape handling of Octal, which does not // require the leading 0. // Java also has the convention of only consuming 2 octal digits if // the three digit number would be > 0xff // c.fChar = 0; nextCharLL(); // Consume the initial 0. int index; for (index=0; index<3; index++) { int32_t ch = peekCharLL(); if (chchDigit7) { if (index==0) { // \0 is not followed by any octal digits. error(U_REGEX_BAD_ESCAPE_SEQUENCE); } break; } c.fChar <<= 3; c.fChar += ch&7; if (c.fChar <= 255) { nextCharLL(); } else { // The last digit made the number too big. Forget we saw it. c.fChar >>= 3; } } c.fQuoted = TRUE; } else if (peekCharLL() == chQ) { // "\Q" enter quote mode, which will continue until "\E" fQuoteMode = TRUE; nextCharLL(); // discard the 'Q'. nextChar(c); // recurse to get the real next char. } else { // We are in a '\' escape that will be handled by the state table scanner. // Just return the backslash, but remember that the following char is to // be taken literally. fInBackslashQuote = TRUE; } } } // re-enable # to end-of-line comments, in case they were disabled. // They are disabled by the parser upon seeing '(?', but this lasts for // the fetching of the next character only. fEOLComments = TRUE; // putc(c.fChar, stdout); } //------------------------------------------------------------------------------ // // scanNamedChar // Get a UChar32 from a \N{UNICODE CHARACTER NAME} in the pattern. // // The scan position will be at the 'N'. On return // the scan position should be just after the '}' // // Return the UChar32 // //------------------------------------------------------------------------------ UChar32 RegexCompile::scanNamedChar() { if (U_FAILURE(*fStatus)) { return 0; } nextChar(fC); if (fC.fChar != chLBrace) { error(U_REGEX_PROPERTY_SYNTAX); return 0; } UnicodeString charName; for (;;) { nextChar(fC); if (fC.fChar == chRBrace) { break; } if (fC.fChar == -1) { error(U_REGEX_PROPERTY_SYNTAX); return 0; } charName.append(fC.fChar); } char name[100]; if (!uprv_isInvariantUString(charName.getBuffer(), charName.length()) || (uint32_t)charName.length()>=sizeof(name)) { // All Unicode character names have only invariant characters. // The API to get a character, given a name, accepts only char *, forcing us to convert, // which requires this error check error(U_REGEX_PROPERTY_SYNTAX); return 0; } charName.extract(0, charName.length(), name, sizeof(name), US_INV); UChar32 theChar = u_charFromName(U_UNICODE_CHAR_NAME, name, fStatus); if (U_FAILURE(*fStatus)) { error(U_REGEX_PROPERTY_SYNTAX); } nextChar(fC); // Continue overall regex pattern processing with char after the '}' return theChar; } //------------------------------------------------------------------------------ // // scanProp Construct a UnicodeSet from the text at the current scan // position, which will be of the form \p{whaterver} // // The scan position will be at the 'p' or 'P'. On return // the scan position should be just after the '}' // // Return a UnicodeSet, constructed from the \P pattern, // or NULL if the pattern is invalid. // //------------------------------------------------------------------------------ UnicodeSet *RegexCompile::scanProp() { UnicodeSet *uset = NULL; if (U_FAILURE(*fStatus)) { return NULL; } U_ASSERT(fC.fChar == chLowerP || fC.fChar == chP); UBool negated = (fC.fChar == chP); UnicodeString propertyName; nextChar(fC); if (fC.fChar != chLBrace) { error(U_REGEX_PROPERTY_SYNTAX); return NULL; } for (;;) { nextChar(fC); if (fC.fChar == chRBrace) { break; } if (fC.fChar == -1) { // Hit the end of the input string without finding the closing '}' error(U_REGEX_PROPERTY_SYNTAX); return NULL; } propertyName.append(fC.fChar); } uset = createSetForProperty(propertyName, negated); nextChar(fC); // Move input scan to position following the closing '}' return uset; } //------------------------------------------------------------------------------ // // scanPosixProp Construct a UnicodeSet from the text at the current scan // position, which is expected be of the form [:property expression:] // // The scan position will be at the opening ':'. On return // the scan position must be on the closing ']' // // Return a UnicodeSet constructed from the pattern, // or NULL if this is not a valid POSIX-style set expression. // If not a property expression, restore the initial scan position // (to the opening ':') // // Note: the opening '[:' is not sufficient to guarantee that // this is a [:property:] expression. // [:'+=,] is a perfectly good ordinary set expression that // happens to include ':' as one of its characters. // //------------------------------------------------------------------------------ UnicodeSet *RegexCompile::scanPosixProp() { UnicodeSet *uset = NULL; if (U_FAILURE(*fStatus)) { return NULL; } U_ASSERT(fC.fChar == chColon); // Save the scanner state. // TODO: move this into the scanner, with the state encapsulated in some way. Ticket 6062 int64_t savedScanIndex = fScanIndex; int64_t savedNextIndex = UTEXT_GETNATIVEINDEX(fRXPat->fPattern); UBool savedQuoteMode = fQuoteMode; UBool savedInBackslashQuote = fInBackslashQuote; UBool savedEOLComments = fEOLComments; int64_t savedLineNum = fLineNum; int64_t savedCharNum = fCharNum; UChar32 savedLastChar = fLastChar; UChar32 savedPeekChar = fPeekChar; RegexPatternChar savedfC = fC; // Scan for a closing ]. A little tricky because there are some perverse // edge cases possible. "[:abc\Qdef:] \E]" is a valid non-property expression, // ending on the second closing ]. UnicodeString propName; UBool negated = FALSE; // Check for and consume the '^' in a negated POSIX property, e.g. [:^Letter:] nextChar(fC); if (fC.fChar == chUp) { negated = TRUE; nextChar(fC); } // Scan for the closing ":]", collecting the property name along the way. UBool sawPropSetTerminator = FALSE; for (;;) { propName.append(fC.fChar); nextChar(fC); if (fC.fQuoted || fC.fChar == -1) { // Escaped characters or end of input - either says this isn't a [:Property:] break; } if (fC.fChar == chColon) { nextChar(fC); if (fC.fChar == chRBracket) { sawPropSetTerminator = TRUE; } break; } } if (sawPropSetTerminator) { uset = createSetForProperty(propName, negated); } else { // No closing ":]". // Restore the original scan position. // The main scanner will retry the input as a normal set expression, // not a [:Property:] expression. fScanIndex = savedScanIndex; fQuoteMode = savedQuoteMode; fInBackslashQuote = savedInBackslashQuote; fEOLComments = savedEOLComments; fLineNum = savedLineNum; fCharNum = savedCharNum; fLastChar = savedLastChar; fPeekChar = savedPeekChar; fC = savedfC; UTEXT_SETNATIVEINDEX(fRXPat->fPattern, savedNextIndex); } return uset; } static inline void addIdentifierIgnorable(UnicodeSet *set, UErrorCode& ec) { set->add(0, 8).add(0x0e, 0x1b).add(0x7f, 0x9f); addCategory(set, U_GC_CF_MASK, ec); } // // Create a Unicode Set from a Unicode Property expression. // This is common code underlying both \p{...} ane [:...:] expressions. // Includes trying the Java "properties" that aren't supported as // normal ICU UnicodeSet properties // static const UChar posSetPrefix[] = {0x5b, 0x5c, 0x70, 0x7b, 0}; // "[\p{" static const UChar negSetPrefix[] = {0x5b, 0x5c, 0x50, 0x7b, 0}; // "[\P{" UnicodeSet *RegexCompile::createSetForProperty(const UnicodeString &propName, UBool negated) { UnicodeString setExpr; UnicodeSet *set; uint32_t usetFlags = 0; if (U_FAILURE(*fStatus)) { return NULL; } // // First try the property as we received it // if (negated) { setExpr.append(negSetPrefix, -1); } else { setExpr.append(posSetPrefix, -1); } setExpr.append(propName); setExpr.append(chRBrace); setExpr.append(chRBracket); if (fModeFlags & UREGEX_CASE_INSENSITIVE) { usetFlags |= USET_CASE_INSENSITIVE; } set = new UnicodeSet(setExpr, usetFlags, NULL, *fStatus); if (U_SUCCESS(*fStatus)) { return set; } delete set; set = NULL; // // The property as it was didn't work. // Do [:word:]. It is not recognized as a property by UnicodeSet. "word" not standard POSIX // or standard Java, but many other regular expression packages do recognize it. if (propName.caseCompare(UNICODE_STRING_SIMPLE("word"), 0) == 0) { *fStatus = U_ZERO_ERROR; set = new UnicodeSet(*(fRXPat->fStaticSets[URX_ISWORD_SET])); if (set == NULL) { *fStatus = U_MEMORY_ALLOCATION_ERROR; return set; } if (negated) { set->complement(); } return set; } // Do Java fixes - // InGreek -> InGreek or Coptic, that being the official Unicode name for that block. // InCombiningMarksforSymbols -> InCombiningDiacriticalMarksforSymbols. // // Note on Spaces: either "InCombiningMarksForSymbols" or "InCombining Marks for Symbols" // is accepted by Java. The property part of the name is compared // case-insenstively. The spaces must be exactly as shown, either // all there, or all omitted, with exactly one at each position // if they are present. From checking against JDK 1.6 // // This code should be removed when ICU properties support the Java compatibility names // (ICU 4.0?) // UnicodeString mPropName = propName; if (mPropName.caseCompare(UNICODE_STRING_SIMPLE("InGreek"), 0) == 0) { mPropName = UNICODE_STRING_SIMPLE("InGreek and Coptic"); } if (mPropName.caseCompare(UNICODE_STRING_SIMPLE("InCombining Marks for Symbols"), 0) == 0 || mPropName.caseCompare(UNICODE_STRING_SIMPLE("InCombiningMarksforSymbols"), 0) == 0) { mPropName = UNICODE_STRING_SIMPLE("InCombining Diacritical Marks for Symbols"); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("all")) == 0) { mPropName = UNICODE_STRING_SIMPLE("javaValidCodePoint"); } // See if the property looks like a Java "InBlockName", which // we will recast as "Block=BlockName" // static const UChar IN[] = {0x49, 0x6E, 0}; // "In" static const UChar BLOCK[] = {0x42, 0x6C, 0x6f, 0x63, 0x6b, 0x3d, 00}; // "Block=" if (mPropName.startsWith(IN, 2) && propName.length()>=3) { setExpr.truncate(4); // Leaves "[\p{", or "[\P{" setExpr.append(BLOCK, -1); setExpr.append(UnicodeString(mPropName, 2)); // Property with the leading "In" removed. setExpr.append(chRBrace); setExpr.append(chRBracket); *fStatus = U_ZERO_ERROR; set = new UnicodeSet(setExpr, usetFlags, NULL, *fStatus); if (U_SUCCESS(*fStatus)) { return set; } delete set; set = NULL; } if (propName.startsWith(UNICODE_STRING_SIMPLE("java")) || propName.compare(UNICODE_STRING_SIMPLE("all")) == 0) { UErrorCode localStatus = U_ZERO_ERROR; //setExpr.remove(); set = new UnicodeSet(); // // Try the various Java specific properties. // These all begin with "java" // if (mPropName.compare(UNICODE_STRING_SIMPLE("javaDefined")) == 0) { addCategory(set, U_GC_CN_MASK, localStatus); set->complement(); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaDigit")) == 0) { addCategory(set, U_GC_ND_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaIdentifierIgnorable")) == 0) { addIdentifierIgnorable(set, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaISOControl")) == 0) { set->add(0, 0x1F).add(0x7F, 0x9F); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaJavaIdentifierPart")) == 0) { addCategory(set, U_GC_L_MASK, localStatus); addCategory(set, U_GC_SC_MASK, localStatus); addCategory(set, U_GC_PC_MASK, localStatus); addCategory(set, U_GC_ND_MASK, localStatus); addCategory(set, U_GC_NL_MASK, localStatus); addCategory(set, U_GC_MC_MASK, localStatus); addCategory(set, U_GC_MN_MASK, localStatus); addIdentifierIgnorable(set, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaJavaIdentifierStart")) == 0) { addCategory(set, U_GC_L_MASK, localStatus); addCategory(set, U_GC_NL_MASK, localStatus); addCategory(set, U_GC_SC_MASK, localStatus); addCategory(set, U_GC_PC_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaLetter")) == 0) { addCategory(set, U_GC_L_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaLetterOrDigit")) == 0) { addCategory(set, U_GC_L_MASK, localStatus); addCategory(set, U_GC_ND_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaLowerCase")) == 0) { addCategory(set, U_GC_LL_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaMirrored")) == 0) { set->applyIntPropertyValue(UCHAR_BIDI_MIRRORED, 1, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaSpaceChar")) == 0) { addCategory(set, U_GC_Z_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaSupplementaryCodePoint")) == 0) { set->add(0x10000, UnicodeSet::MAX_VALUE); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaTitleCase")) == 0) { addCategory(set, U_GC_LT_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaUnicodeIdentifierStart")) == 0) { addCategory(set, U_GC_L_MASK, localStatus); addCategory(set, U_GC_NL_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaUnicodeIdentifierPart")) == 0) { addCategory(set, U_GC_L_MASK, localStatus); addCategory(set, U_GC_PC_MASK, localStatus); addCategory(set, U_GC_ND_MASK, localStatus); addCategory(set, U_GC_NL_MASK, localStatus); addCategory(set, U_GC_MC_MASK, localStatus); addCategory(set, U_GC_MN_MASK, localStatus); addIdentifierIgnorable(set, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaUpperCase")) == 0) { addCategory(set, U_GC_LU_MASK, localStatus); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaValidCodePoint")) == 0) { set->add(0, UnicodeSet::MAX_VALUE); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("javaWhitespace")) == 0) { addCategory(set, U_GC_Z_MASK, localStatus); set->removeAll(UnicodeSet().add(0xa0).add(0x2007).add(0x202f)); set->add(9, 0x0d).add(0x1c, 0x1f); } else if (mPropName.compare(UNICODE_STRING_SIMPLE("all")) == 0) { set->add(0, UnicodeSet::MAX_VALUE); } if (U_SUCCESS(localStatus) && !set->isEmpty()) { *fStatus = U_ZERO_ERROR; if (usetFlags & USET_CASE_INSENSITIVE) { set->closeOver(USET_CASE_INSENSITIVE); } if (negated) { set->complement(); } return set; } delete set; set = NULL; } error(*fStatus); return NULL; } // // SetEval Part of the evaluation of [set expressions]. // Perform any pending (stacked) operations with precedence // equal or greater to that of the next operator encountered // in the expression. // void RegexCompile::setEval(int32_t nextOp) { UnicodeSet *rightOperand = NULL; UnicodeSet *leftOperand = NULL; for (;;) { U_ASSERT(fSetOpStack.empty()==FALSE); int32_t pendingSetOperation = fSetOpStack.peeki(); if ((pendingSetOperation&0xffff0000) < (nextOp&0xffff0000)) { break; } fSetOpStack.popi(); U_ASSERT(fSetStack.empty() == FALSE); rightOperand = (UnicodeSet *)fSetStack.peek(); switch (pendingSetOperation) { case setNegation: rightOperand->complement(); break; case setCaseClose: // TODO: need a simple close function. Ticket 6065 rightOperand->closeOver(USET_CASE_INSENSITIVE); rightOperand->removeAllStrings(); break; case setDifference1: case setDifference2: fSetStack.pop(); leftOperand = (UnicodeSet *)fSetStack.peek(); leftOperand->removeAll(*rightOperand); delete rightOperand; break; case setIntersection1: case setIntersection2: fSetStack.pop(); leftOperand = (UnicodeSet *)fSetStack.peek(); leftOperand->retainAll(*rightOperand); delete rightOperand; break; case setUnion: fSetStack.pop(); leftOperand = (UnicodeSet *)fSetStack.peek(); leftOperand->addAll(*rightOperand); delete rightOperand; break; default: U_ASSERT(FALSE); break; } } } void RegexCompile::setPushOp(int32_t op) { setEval(op); fSetOpStack.push(op, *fStatus); fSetStack.push(new UnicodeSet(), *fStatus); } U_NAMESPACE_END #endif // !UCONFIG_NO_REGULAR_EXPRESSIONS