icuSources/common/rbbiscan.cpp

   1
   2 //
   3 //  file:  rbbiscan.cpp
   4 //
   5 //  Copyright (C) 2002-2004, International Business Machines Corporation and others.
   6 //  All Rights Reserved.
   7 //
   8 //  This file contains the Rule Based Break Iterator Rule Builder functions for
   9 //   scanning the rules and assembling a parse tree.  This is the first phase
  10 //   of compiling the rules.
  11 //
  12 //  The overall of the rules is managed by class RBBIRuleBuilder, which will
  13 //  create and use an instance of this class as part of the process.
  14 //
  15
  16 #include "unicode/utypes.h"
  17
  18 #if !UCONFIG_NO_BREAK_ITERATION
  19
  20 #include "unicode/unistr.h"
  21 #include "unicode/uniset.h"
  22 #include "unicode/uchar.h"
  23 #include "unicode/uchriter.h"
  24 #include "unicode/parsepos.h"
  25 #include "unicode/parseerr.h"
  26 #include "uprops.h"
  27 #include "cmemory.h"
  28 #include "cstring.h"
  29
  30 #include "rbbirpt.h"   // Contains state table for the rbbi rules parser.
  31                        //   generated by a Perl script.
  32 #include "rbbirb.h"
  33 #include "rbbinode.h"
  34 #include "rbbiscan.h"
  35 #include "rbbitblb.h"
  36
  37 #include "uassert.h"
  38
  39
  40 //----------------------------------------------------------------------------------------
  41 //
  42 // Unicode Set init strings for each of the character classes needed for parsing a rule file.
  43 //               (Initialized with hex values for portability to EBCDIC based machines.
  44 //                Really ugly, but there's no good way to avoid it.)
  45 //
  46 //              The sets are referred to by name in the rbbirpt.txt, which is the
  47 //              source form of the state transition table for the RBBI rule parser.
  48 //
  49 //----------------------------------------------------------------------------------------
  50 static const UChar gRuleSet_rule_char_pattern[]       = {
  51  //   [    ^      [    \     p     {      Z     }     \     u    0      0    2      0
  52     0x5b, 0x5e, 0x5b, 0x5c, 0x70, 0x7b, 0x5a, 0x7d, 0x5c, 0x75, 0x30, 0x30, 0x32, 0x30,
  53  //   -    \      u    0     0     7      f     ]     -     [    \      p
  54     0x2d, 0x5c, 0x75, 0x30, 0x30, 0x37, 0x66, 0x5d, 0x2d, 0x5b, 0x5c, 0x70,
  55  //   {     L     }    ]     -     [      \     p     {     N    }      ]     ]
  56     0x7b, 0x4c, 0x7d, 0x5d, 0x2d, 0x5b, 0x5c, 0x70, 0x7b, 0x4e, 0x7d, 0x5d, 0x5d, 0};
  57
  58 static const UChar gRuleSet_name_char_pattern[]       = {
  59 //    [    _      \    p     {     L      }     \     p     {    N      }     ]
  60     0x5b, 0x5f, 0x5c, 0x70, 0x7b, 0x4c, 0x7d, 0x5c, 0x70, 0x7b, 0x4e, 0x7d, 0x5d, 0};
  61
  62 static const UChar gRuleSet_digit_char_pattern[] = {
  63 //    [    0      -    9     ]
  64     0x5b, 0x30, 0x2d, 0x39, 0x5d, 0};
  65
  66 static const UChar gRuleSet_name_start_char_pattern[] = {
  67 //    [    _      \    p     {     L      }     ]
  68     0x5b, 0x5f, 0x5c, 0x70, 0x7b, 0x4c, 0x7d, 0x5d, 0 };
  69
  70 static const UChar kAny[] = {0x61, 0x6e, 0x79, 0x00};  // "any"
  71
  72
  73 U_CDECL_BEGIN
  74 static void  U_EXPORT2 U_CALLCONV RBBISetTable_deleter(void *p) {
  75     RBBISetTableEl *px = (RBBISetTableEl *)p;
  76     delete px->key;
  77     // Note:  px->val is owned by the linked list "fSetsListHead" in scanner.
  78     //        Don't delete the value nodes here.
  79     uprv_free(px);
  80 }
  81 U_CDECL_END
  82
  83 U_NAMESPACE_BEGIN
  84
  85 //----------------------------------------------------------------------------------------
  86 //
  87 //  Constructor.
  88 //
  89 //----------------------------------------------------------------------------------------
  90 RBBIRuleScanner::RBBIRuleScanner(RBBIRuleBuilder *rb)
  91 {
  92     fRB                 = rb;
  93     fStackPtr           = 0;
  94     fStack[fStackPtr]   = 0;
  95     fNodeStackPtr       = 0;
  96     fRuleNum            = 0;
  97     fNodeStack[0]       = NULL;
  98
  99     fRuleSets[kRuleSet_rule_char-128]       = NULL;
 100     fRuleSets[kRuleSet_white_space-128]     = NULL;
 101     fRuleSets[kRuleSet_name_char-128]       = NULL;
 102     fRuleSets[kRuleSet_name_start_char-128] = NULL;
 103     fRuleSets[kRuleSet_digit_char-128]      = NULL;
 104     fSymbolTable                            = NULL;
 105     fSetTable                               = NULL;
 106
 107     fScanIndex = 0;
 108     fNextIndex = 0;
 109
 110     fReverseRule        = FALSE;
 111     fLookAheadRule      = FALSE;
 112
 113     fLineNum    = 1;
 114     fCharNum    = 0;
 115     fQuoteMode  = FALSE;
 116
 117     // Do not check status until after all critical fields are sufficiently initialized
 118     //   that the destructor can run cleanly.
 119     if (U_FAILURE(*rb->fStatus)) {
 120         return;
 121     }
 122
 123     //
 124     //  Set up the constant Unicode Sets.
 125     //     Note:  These could be made static, lazily initialized, and shared among
 126     //            all instances of RBBIRuleScanners.  BUT this is quite a bit simpler,
 127     //            and the time to build these few sets should be small compared to a
 128     //            full break iterator build.
 129     fRuleSets[kRuleSet_rule_char-128]       = new UnicodeSet(gRuleSet_rule_char_pattern,       *rb->fStatus);
 130     fRuleSets[kRuleSet_white_space-128]     = (UnicodeSet*) uprv_openRuleWhiteSpaceSet(rb->fStatus);
 131     fRuleSets[kRuleSet_name_char-128]       = new UnicodeSet(gRuleSet_name_char_pattern,       *rb->fStatus);
 132     fRuleSets[kRuleSet_name_start_char-128] = new UnicodeSet(gRuleSet_name_start_char_pattern, *rb->fStatus);
 133     fRuleSets[kRuleSet_digit_char-128]      = new UnicodeSet(gRuleSet_digit_char_pattern,      *rb->fStatus);
 134     if (*rb->fStatus == U_ILLEGAL_ARGUMENT_ERROR) {
 135         // This case happens if ICU's data is missing.  UnicodeSet tries to look up property
 136         //   names from the init string, can't find them, and claims an illegal arguement.
 137         //   Change the error so that the actual problem will be clearer to users.
 138         *rb->fStatus = U_BRK_INIT_ERROR;
 139     }
 140     if (U_FAILURE(*rb->fStatus)) {
 141         return;
 142     }
 143
 144     fSymbolTable = new RBBISymbolTable(this, rb->fRules, *rb->fStatus);
 145     fSetTable    = uhash_open(uhash_hashUnicodeString, uhash_compareUnicodeString, rb->fStatus);
 146     uhash_setValueDeleter(fSetTable, RBBISetTable_deleter);
 147 }
 148
 149
 150
 151 //----------------------------------------------------------------------------------------
 152 //
 153 //  Destructor
 154 //
 155 //----------------------------------------------------------------------------------------
 156 RBBIRuleScanner::~RBBIRuleScanner() {
 157     delete fRuleSets[kRuleSet_rule_char-128];
 158     delete fRuleSets[kRuleSet_white_space-128];
 159     delete fRuleSets[kRuleSet_name_char-128];
 160     delete fRuleSets[kRuleSet_name_start_char-128];
 161     delete fRuleSets[kRuleSet_digit_char-128];
 162
 163     delete fSymbolTable;
 164     if (fSetTable != NULL) {
 165          uhash_close(fSetTable);
 166          fSetTable = NULL;
 167
 168     }
 169
 170
 171     // Node Stack.
 172     //   Normally has one entry, which is the entire parse tree for the rules.
 173     //   If errors occured, there may be additional subtrees left on the stack.
 174     while (fNodeStackPtr > 0) {
 175         delete fNodeStack[fNodeStackPtr];
 176         fNodeStackPtr--;
 177     }
 178
 179 }
 180
 181 //----------------------------------------------------------------------------------------
 182 //
 183 //  doParseAction        Do some action during rule parsing.
 184 //                       Called by the parse state machine.
 185 //                       Actions build the parse tree and Unicode Sets,
 186 //                       and maintain the parse stack for nested expressions.
 187 //
 188 //                       TODO:  unify EParseAction and RBBI_RuleParseAction enum types.
 189 //                              They represent exactly the same thing.  They're separate
 190 //                              only to work around enum forward declaration restrictions
 191 //                              in some compilers, while at the same time avoiding multiple
 192 //                              definitions problems.  I'm sure that there's a better way.
 193 //
 194 //----------------------------------------------------------------------------------------
 195 UBool RBBIRuleScanner::doParseActions(EParseAction action)
 196 {
 197     RBBINode *n       = NULL;
 198
 199     UBool   returnVal = TRUE;
 200
 201     switch ((RBBI_RuleParseAction)action) {
 202
 203     case doExprStart:
 204         pushNewNode(RBBINode::opStart);
 205         fRuleNum++;
 206         break;
 207
 208
 209     case doExprOrOperator:
 210         {
 211             fixOpStack(RBBINode::precOpCat);
 212             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 213             RBBINode  *orNode      = pushNewNode(RBBINode::opOr);
 214             orNode->fLeftChild     = operandNode;
 215             operandNode->fParent   = orNode;
 216         }
 217         break;
 218
 219     case doExprCatOperator:
 220         // concatenation operator.
 221         // For the implicit concatenation of adjacent terms in an expression that are
 222         //   not separated by any other operator.  Action is invoked between the
 223         //   actions for the two terms.
 224         {
 225             fixOpStack(RBBINode::precOpCat);
 226             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 227             RBBINode  *catNode     = pushNewNode(RBBINode::opCat);
 228             catNode->fLeftChild    = operandNode;
 229             operandNode->fParent   = catNode;
 230         }
 231         break;
 232
 233     case doLParen:
 234         // Open Paren.
 235         //   The openParen node is a dummy operation type with a low precedence,
 236         //     which has the affect of ensuring that any real binary op that
 237         //     follows within the parens binds more tightly to the operands than
 238         //     stuff outside of the parens.
 239         pushNewNode(RBBINode::opLParen);
 240         break;
 241
 242     case doExprRParen:
 243         fixOpStack(RBBINode::precLParen);
 244         break;
 245
 246     case doNOP:
 247         break;
 248
 249     case doStartAssign:
 250         // We've just scanned "$variable = "
 251         // The top of the node stack has the $variable ref node.
 252
 253         // Save the start position of the RHS text in the StartExpression node
 254         //   that precedes the $variableReference node on the stack.
 255         //   This will eventually be used when saving the full $variable replacement
 256         //   text as a string.
 257         n = fNodeStack[fNodeStackPtr-1];
 258         n->fFirstPos = fNextIndex;              // move past the '='
 259
 260         // Push a new start-of-expression node; needed to keep parse of the
 261         //   RHS expression happy.
 262         pushNewNode(RBBINode::opStart);
 263         break;
 264
 265
 266
 267
 268     case doEndAssign:
 269         {
 270             // We have reached the end of an assignement statement.
 271             //   Current scan char is the ';' that terminates the assignment.
 272
 273             // Terminate expression, leaves expression parse tree rooted in TOS node.
 274             fixOpStack(RBBINode::precStart);
 275
 276             RBBINode *startExprNode  = fNodeStack[fNodeStackPtr-2];
 277             RBBINode *varRefNode     = fNodeStack[fNodeStackPtr-1];
 278             RBBINode *RHSExprNode    = fNodeStack[fNodeStackPtr];
 279
 280             // Save original text of right side of assignment, excluding the terminating ';'
 281             //  in the root of the node for the right-hand-side expression.
 282             RHSExprNode->fFirstPos = startExprNode->fFirstPos;
 283             RHSExprNode->fLastPos  = fScanIndex;
 284             fRB->fRules.extractBetween(RHSExprNode->fFirstPos, RHSExprNode->fLastPos, RHSExprNode->fText);
 285
 286             // Expression parse tree becomes l. child of the $variable reference node.
 287             varRefNode->fLeftChild = RHSExprNode;
 288             RHSExprNode->fParent   = varRefNode;
 289
 290             // Make a symbol table entry for the $variableRef node.
 291             fSymbolTable->addEntry(varRefNode->fText, varRefNode, *fRB->fStatus);
 292
 293             // Clean up the stack.
 294             delete startExprNode;
 295             fNodeStackPtr-=3;
 296             break;
 297         }
 298
 299     case doEndOfRule:
 300         {
 301         fixOpStack(RBBINode::precStart);      // Terminate expression, leaves expression
 302         if (U_FAILURE(*fRB->fStatus)) {       //   parse tree rooted in TOS node.
 303             break;
 304         }
 305         if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "rtree")) {printNodeStack("end of rule");}
 306         U_ASSERT(fNodeStackPtr == 1);
 307
 308         // If this rule includes a look-ahead '/', add a endMark node to the
 309         //   expression tree.
 310         if (fLookAheadRule) {
 311             RBBINode  *thisRule       = fNodeStack[fNodeStackPtr];
 312             RBBINode  *endNode        = pushNewNode(RBBINode::endMark);
 313             RBBINode  *catNode        = pushNewNode(RBBINode::opCat);
 314             fNodeStackPtr -= 2;
 315             catNode->fLeftChild       = thisRule;
 316             catNode->fRightChild      = endNode;
 317             fNodeStack[fNodeStackPtr] = catNode;
 318             endNode->fVal             = fRuleNum;
 319             endNode->fLookAheadEnd    = TRUE;
 320         }
 321
 322         // All rule expressions are ORed together.
 323         // The ';' that terminates an expression really just functions as a '|' with
 324         //   a low operator prededence.
 325         //
 326         // Each of the four sets of rules are collected separately.
 327         //  (forward, reverse, safe_forward, safe_reverse)
 328         //  OR this rule into the appropriate group of them.
 329         //
 330         RBBINode **destRules = (fReverseRule? &fRB->fReverseTree : fRB->fDefaultTree);
 331
 332         if (*destRules != NULL) {
 333             // This is not the first rule encounted.
 334             // OR previous stuff  (from *destRules)
 335             // with the current rule expression (on the Node Stack)
 336             //  with the resulting OR expression going to *destRules
 337             //
 338             RBBINode  *thisRule    = fNodeStack[fNodeStackPtr];
 339             RBBINode  *prevRules   = *destRules;
 340             RBBINode  *orNode      = pushNewNode(RBBINode::opOr);
 341             orNode->fLeftChild     = prevRules;
 342             prevRules->fParent     = orNode;
 343             orNode->fRightChild    = thisRule;
 344             thisRule->fParent      = orNode;
 345             *destRules             = orNode;
 346         }
 347         else
 348         {
 349             // This is the first rule encountered (for this direction).
 350             // Just move its parse tree from the stack to *destRules.
 351             *destRules = fNodeStack[fNodeStackPtr];
 352         }
 353         fReverseRule   = FALSE;   // in preparation for the next rule.
 354         fLookAheadRule = FALSE;
 355         fNodeStackPtr  = 0;
 356         }
 357         break;
 358
 359
 360     case doRuleError:
 361         error(U_BRK_RULE_SYNTAX);
 362         returnVal = FALSE;
 363         break;
 364
 365
 366     case doVariableNameExpectedErr:
 367         error(U_BRK_RULE_SYNTAX);
 368         break;
 369
 370
 371     //
 372     //  Unary operands  + ? *
 373     //    These all appear after the operand to which they apply.
 374     //    When we hit one, the operand (may be a whole sub expression)
 375     //    will be on the top of the stack.
 376     //    Unary Operator becomes TOS, with the old TOS as its one child.
 377     case doUnaryOpPlus:
 378         {
 379             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 380             RBBINode  *plusNode    = pushNewNode(RBBINode::opPlus);
 381             plusNode->fLeftChild   = operandNode;
 382             operandNode->fParent   = plusNode;
 383         }
 384         break;
 385
 386     case doUnaryOpQuestion:
 387         {
 388             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 389             RBBINode  *qNode       = pushNewNode(RBBINode::opQuestion);
 390             qNode->fLeftChild      = operandNode;
 391             operandNode->fParent   = qNode;
 392         }
 393         break;
 394
 395     case doUnaryOpStar:
 396         {
 397             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 398             RBBINode  *starNode    = pushNewNode(RBBINode::opStar);
 399             starNode->fLeftChild   = operandNode;
 400             operandNode->fParent   = starNode;
 401         }
 402         break;
 403
 404     case doRuleChar:
 405         // A "Rule Character" is any single character that is a literal part
 406         // of the regular expression.  Like a, b and c in the expression "(abc*) | [:L:]"
 407         // These are pretty uncommon in break rules; the terms are more commonly
 408         //  sets.  To keep things uniform, treat these characters like as
 409         // sets that just happen to contain only one character.
 410         {
 411             n = pushNewNode(RBBINode::setRef);
 412             findSetFor(fC.fChar, n);
 413             n->fFirstPos = fScanIndex;
 414             n->fLastPos  = fNextIndex;
 415             fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
 416             break;
 417         }
 418
 419     case doDotAny:
 420         // scanned a ".", meaning match any single character.
 421         {
 422             n = pushNewNode(RBBINode::setRef);
 423             findSetFor(kAny, n);
 424             n->fFirstPos = fScanIndex;
 425             n->fLastPos  = fNextIndex;
 426             fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
 427             break;
 428         }
 429         break;
 430
 431     case doSlash:
 432         // Scanned a '/', which identifies a look-ahead break position in a rule.
 433         n = pushNewNode(RBBINode::lookAhead);
 434         n->fVal      = fRuleNum;
 435         n->fFirstPos = fScanIndex;
 436         n->fLastPos  = fNextIndex;
 437         fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
 438         fLookAheadRule = TRUE;
 439         break;
 440
 441
 442     case doStartTagValue:
 443         // Scanned a '{', the opening delimiter for a tag value within a rule.
 444         n = pushNewNode(RBBINode::tag);
 445         n->fVal      = 0;
 446         n->fFirstPos = fScanIndex;
 447         n->fLastPos  = fNextIndex;
 448         break;
 449
 450     case doTagDigit:
 451         // Just scanned a decimal digit that's part of a tag value
 452         {
 453             n = fNodeStack[fNodeStackPtr];
 454             uint32_t v = u_charDigitValue(fC.fChar);
 455             U_ASSERT(v < 10);
 456             n->fVal = n->fVal*10 + v;
 457             break;
 458         }
 459
 460     case doTagValue:
 461         n = fNodeStack[fNodeStackPtr];
 462         n->fLastPos = fNextIndex;
 463         fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
 464         break;
 465
 466     case doTagExpectedError:
 467         error(U_BRK_MALFORMED_RULE_TAG);
 468         returnVal = FALSE;
 469         break;
 470
 471     case doOptionStart:
 472         // Scanning a !!option.   At the start of string.
 473         fOptionStart = fScanIndex;
 474         break;
 475
 476     case doOptionEnd:
 477         {
 478             UnicodeString opt(fRB->fRules, fOptionStart, fScanIndex-fOptionStart);
 479             if (opt == UNICODE_STRING("chain", 5)) {
 480                 fRB->fChainRules = TRUE;
 481             } else if (opt == UNICODE_STRING("LBCMNoChain", 11)) {
 482                 fRB->fLBCMNoChain = TRUE;
 483             } else if (opt == UNICODE_STRING("forward", 7)) {
 484                 fRB->fDefaultTree   = &fRB->fForwardTree;
 485             } else if (opt == UNICODE_STRING("reverse", 7)) {
 486                 fRB->fDefaultTree   = &fRB->fReverseTree;
 487             } else if (opt == UNICODE_STRING("safe_forward", 12)) {
 488                 fRB->fDefaultTree   = &fRB->fSafeFwdTree;
 489             } else if (opt == UNICODE_STRING("safe_reverse", 12)) {
 490                 fRB->fDefaultTree   = &fRB->fSafeRevTree;
 491             } else if (opt == UNICODE_STRING("lookAheadHardBreak", 18)) {
 492                 fRB->fLookAheadHardBreak = TRUE;
 493             } else {
 494                 error(U_BRK_UNRECOGNIZED_OPTION);
 495             }
 496         }
 497         break;
 498
 499     case doReverseDir:
 500         fReverseRule = TRUE;
 501         break;
 502
 503     case doStartVariableName:
 504         n = pushNewNode(RBBINode::varRef);
 505         if (U_FAILURE(*fRB->fStatus)) {
 506             break;
 507         }
 508         n->fFirstPos = fScanIndex;
 509         break;
 510
 511     case doEndVariableName:
 512         n = fNodeStack[fNodeStackPtr];
 513         if (n==NULL || n->fType != RBBINode::varRef) {
 514             error(U_BRK_INTERNAL_ERROR);
 515             break;
 516         }
 517         n->fLastPos = fScanIndex;
 518         fRB->fRules.extractBetween(n->fFirstPos+1, n->fLastPos, n->fText);
 519         // Look the newly scanned name up in the symbol table
 520         //   If there's an entry, set the l. child of the var ref to the replacement expression.
 521         //   (We also pass through here when scanning assignments, but no harm is done, other
 522         //    than a slight wasted effort that seems hard to avoid.  Lookup will be null)
 523         n->fLeftChild = fSymbolTable->lookupNode(n->fText);
 524         break;
 525
 526     case doCheckVarDef:
 527         n = fNodeStack[fNodeStackPtr];
 528         if (n->fLeftChild == NULL) {
 529             error(U_BRK_UNDEFINED_VARIABLE);
 530             returnVal = FALSE;
 531         }
 532         break;
 533
 534     case doExprFinished:
 535         break;
 536
 537     case doRuleErrorAssignExpr:
 538         error(U_BRK_ASSIGN_ERROR);
 539         returnVal = FALSE;
 540         break;
 541
 542     case doExit:
 543         returnVal = FALSE;
 544         break;
 545
 546     case doScanUnicodeSet:
 547         scanSet();
 548         break;
 549
 550     default:
 551         error(U_BRK_INTERNAL_ERROR);
 552         returnVal = FALSE;
 553         break;
 554     }
 555     return returnVal;
 556 }
 557
 558
 559
 560
 561 //----------------------------------------------------------------------------------------
 562 //
 563 //  Error         Report a rule parse error.
 564 //                Only report it if no previous error has been recorded.
 565 //
 566 //----------------------------------------------------------------------------------------
 567 void RBBIRuleScanner::error(UErrorCode e) {
 568     if (U_SUCCESS(*fRB->fStatus)) {
 569         *fRB->fStatus = e;
 570         fRB->fParseError->line  = fLineNum;
 571         fRB->fParseError->offset = fCharNum;
 572         fRB->fParseError->preContext[0] = 0;
 573         fRB->fParseError->preContext[0] = 0;
 574     }
 575 }
 576
 577
 578
 579
 580 //----------------------------------------------------------------------------------------
 581 //
 582 //  fixOpStack   The parse stack holds partially assembled chunks of the parse tree.
 583 //               An entry on the stack may be as small as a single setRef node,
 584 //               or as large as the parse tree
 585 //               for an entire expression (this will be the one item left on the stack
 586 //               when the parsing of an RBBI rule completes.
 587 //
 588 //               This function is called when a binary operator is encountered.
 589 //               It looks back up the stack for operators that are not yet associated
 590 //               with a right operand, and if the precedence of the stacked operator >=
 591 //               the precedence of the current operator, binds the operand left,
 592 //               to the previously encountered operator.
 593 //
 594 //----------------------------------------------------------------------------------------
 595 void RBBIRuleScanner::fixOpStack(RBBINode::OpPrecedence p) {
 596     RBBINode *n;
 597     // printNodeStack("entering fixOpStack()");
 598     for (;;) {
 599         n = fNodeStack[fNodeStackPtr-1];   // an operator node
 600         if (n->fPrecedence == 0) {
 601             RBBIDebugPuts("RBBIRuleScanner::fixOpStack, bad operator node");
 602             error(U_BRK_INTERNAL_ERROR);
 603             return;
 604         }
 605
 606         if (n->fPrecedence < p || n->fPrecedence <= RBBINode::precLParen) {
 607             // The most recent operand goes with the current operator,
 608             //   not with the previously stacked one.
 609             break;
 610         }
 611             // Stack operator is a binary op  ( '|' or concatenation)
 612             //   TOS operand becomes right child of this operator.
 613             //   Resulting subexpression becomes the TOS operand.
 614             n->fRightChild = fNodeStack[fNodeStackPtr];
 615             fNodeStack[fNodeStackPtr]->fParent = n;
 616             fNodeStackPtr--;
 617         // printNodeStack("looping in fixOpStack()   ");
 618     }
 619
 620     if (p <= RBBINode::precLParen) {
 621         // Scan is at a right paren or end of expression.
 622         //  The scanned item must match the stack, or else there was an error.
 623         //  Discard the left paren (or start expr) node from the stack,
 624             //  leaving the completed (sub)expression as TOS.
 625             if (n->fPrecedence != p) {
 626                 // Right paren encountered matched start of expression node, or
 627                 // end of expression matched with a left paren node.
 628                 error(U_BRK_MISMATCHED_PAREN);
 629             }
 630             fNodeStack[fNodeStackPtr-1] = fNodeStack[fNodeStackPtr];
 631             fNodeStackPtr--;
 632             // Delete the now-discarded LParen or Start node.
 633             delete n;
 634     }
 635     // printNodeStack("leaving fixOpStack()");
 636 }
 637
 638
 639
 640
 641 //----------------------------------------------------------------------------------------
 642 //
 643 //   findSetFor    given a UnicodeString,
 644 //                  - find the corresponding Unicode Set  (uset node)
 645 //                         (create one if necessary)
 646 //                  - Set fLeftChild of the caller's node (should be a setRef node)
 647 //                         to the uset node
 648 //                 Maintain a hash table of uset nodes, so the same one is always used
 649 //                    for the same string.
 650 //                 If a "to adopt" set is provided and we haven't seen this key before,
 651 //                    add the provided set to the hash table.
 652 //                 If the string is one (32 bit) char in length, the set contains
 653 //                    just one element which is the char in question.
 654 //                 If the string is "any", return a set containing all chars.
 655 //
 656 //----------------------------------------------------------------------------------------
 657 void RBBIRuleScanner::findSetFor(const UnicodeString &s, RBBINode *node, UnicodeSet *setToAdopt) {
 658
 659     RBBISetTableEl   *el;
 660
 661     // First check whether we've already cached a set for this string.
 662     // If so, just use the cached set in the new node.
 663     //   delete any set provided by the caller, since we own it.
 664     el = (RBBISetTableEl *)uhash_get(fSetTable, &s);
 665     if (el != NULL) {
 666         delete setToAdopt;
 667         node->fLeftChild = el->val;
 668         U_ASSERT(node->fLeftChild->fType == RBBINode::uset);
 669         return;
 670     }
 671
 672     // Haven't seen this set before.
 673     // If the caller didn't provide us with a prebuilt set,
 674     //   create a new UnicodeSet now.
 675     if (setToAdopt == NULL) {
 676         if (s.compare(kAny, -1) == 0) {
 677             setToAdopt = new UnicodeSet(0x000000, 0x10ffff);
 678         } else {
 679             UChar32 c;
 680             c = s.char32At(0);
 681             setToAdopt = new UnicodeSet(c, c);
 682         }
 683     }
 684
 685     //
 686     // Make a new uset node to refer to this UnicodeSet
 687     // This new uset node becomes the child of the caller's setReference node.
 688     //
 689     RBBINode *usetNode    = new RBBINode(RBBINode::uset);
 690     usetNode->fInputSet   = setToAdopt;
 691     usetNode->fParent     = node;
 692     node->fLeftChild      = usetNode;
 693     usetNode->fText = s;
 694
 695
 696     //
 697     // Add the new uset node to the list of all uset nodes.
 698     //
 699     fRB->fUSetNodes->addElement(usetNode, *fRB->fStatus);
 700
 701
 702     //
 703     // Add the new set to the set hash table.
 704     //
 705     el      = (RBBISetTableEl *)uprv_malloc(sizeof(RBBISetTableEl));
 706     UnicodeString *tkey = new UnicodeString(s);
 707     if (tkey == NULL || el == NULL || setToAdopt == NULL) {
 708         error(U_MEMORY_ALLOCATION_ERROR);
 709         return;
 710     }
 711     el->key = tkey;
 712     el->val = usetNode;
 713     uhash_put(fSetTable, el->key, el, fRB->fStatus);
 714
 715     return;
 716 }
 717
 718
 719
 720 //
 721 //  Assorted Unicode character constants.
 722 //     Numeric because there is no portable way to enter them as literals.
 723 //     (Think EBCDIC).
 724 //
 725 static const UChar      chCR        = 0x0d;      // New lines, for terminating comments.
 726 static const UChar      chLF        = 0x0a;
 727 static const UChar      chNEL       = 0x85;      //    NEL newline variant
 728 static const UChar      chLS        = 0x2028;    //    Unicode Line Separator
 729 static const UChar      chApos      = 0x27;      //  single quote, for quoted chars.
 730 static const UChar      chPound     = 0x23;      // '#', introduces a comment.
 731 static const UChar      chBackSlash = 0x5c;      // '\'  introduces a char escape
 732 static const UChar      chLParen    = 0x28;
 733 static const UChar      chRParen    = 0x29;
 734
 735
 736 //----------------------------------------------------------------------------------------
 737 //
 738 //  stripRules    Return a rules string without unnecessary
 739 //                characters.
 740 //
 741 //----------------------------------------------------------------------------------------
 742 UnicodeString RBBIRuleScanner::stripRules(const UnicodeString &rules) {
 743     UnicodeString strippedRules;
 744     int rulesLength = rules.length();
 745     for (int idx = 0; idx < rulesLength; ) {
 746         UChar ch = rules[idx++];
 747         if (ch == chPound) {
 748             while (idx < rulesLength
 749                 && ch != chCR && ch != chLF && ch != chNEL)
 750             {
 751                 ch = rules[idx++];
 752             }
 753         }
 754         if (!u_isISOControl(ch)) {
 755             strippedRules.append(ch);
 756         }
 757     }
 758     // strippedRules = strippedRules.unescape();
 759     return strippedRules;
 760 }
 761
 762
 763 //----------------------------------------------------------------------------------------
 764 //
 765 //  nextCharLL    Low Level Next Char from rule input source.
 766 //                Get a char from the input character iterator,
 767 //                keep track of input position for error reporting.
 768 //
 769 //----------------------------------------------------------------------------------------
 770 UChar32  RBBIRuleScanner::nextCharLL() {
 771     UChar32  ch;
 772
 773     if (fNextIndex >= fRB->fRules.length()) {
 774         return (UChar32)-1;
 775     }
 776     ch         = fRB->fRules.char32At(fNextIndex);
 777     fNextIndex = fRB->fRules.moveIndex32(fNextIndex, 1);
 778
 779     if (ch == chCR ||
 780         ch == chNEL ||
 781         ch == chLS   ||
 782         ch == chLF && fLastChar != chCR) {
 783         // Character is starting a new line.  Bump up the line number, and
 784         //  reset the column to 0.
 785         fLineNum++;
 786         fCharNum=0;
 787         if (fQuoteMode) {
 788             error(U_BRK_NEW_LINE_IN_QUOTED_STRING);
 789             fQuoteMode = FALSE;
 790         }
 791     }
 792     else {
 793         // Character is not starting a new line.  Except in the case of a
 794         //   LF following a CR, increment the column position.
 795         if (ch != chLF) {
 796             fCharNum++;
 797         }
 798     }
 799     fLastChar = ch;
 800     return ch;
 801 }
 802
 803
 804 //---------------------------------------------------------------------------------
 805 //
 806 //   nextChar     for rules scanning.  At this level, we handle stripping
 807 //                out comments and processing backslash character escapes.
 808 //                The rest of the rules grammar is handled at the next level up.
 809 //
 810 //---------------------------------------------------------------------------------
 811 void RBBIRuleScanner::nextChar(RBBIRuleChar &c) {
 812
 813     // Unicode Character constants needed for the processing done by nextChar(),
 814     //   in hex because literals wont work on EBCDIC machines.
 815
 816     fScanIndex = fNextIndex;
 817     c.fChar    = nextCharLL();
 818     c.fEscaped = FALSE;
 819
 820     //
 821     //  check for '' sequence.
 822     //  These are recognized in all contexts, whether in quoted text or not.
 823     //
 824     if (c.fChar == chApos) {
 825         if (fRB->fRules.char32At(fNextIndex) == chApos) {
 826             c.fChar    = nextCharLL();        // get nextChar officially so character counts
 827             c.fEscaped = TRUE;                //   stay correct.
 828         }
 829         else
 830         {
 831             // Single quote, by itself.
 832             //   Toggle quoting mode.
 833             //   Return either '('  or ')', because quotes cause a grouping of the quoted text.
 834             fQuoteMode = !fQuoteMode;
 835             if (fQuoteMode == TRUE) {
 836                 c.fChar = chLParen;
 837             } else {
 838                 c.fChar = chRParen;
 839             }
 840             c.fEscaped = FALSE;      // The paren that we return is not escaped.
 841             return;
 842         }
 843     }
 844
 845     if (fQuoteMode) {
 846         c.fEscaped = TRUE;
 847     }
 848     else
 849     {
 850         // We are not in a 'quoted region' of the source.
 851         //
 852         if (c.fChar == chPound) {
 853             // Start of a comment.  Consume the rest of it.
 854             //  The new-line char that terminates the comment is always returned.
 855             //  It will be treated as white-space, and serves to break up anything
 856             //    that might otherwise incorrectly clump together with a comment in
 857             //    the middle (a variable name, for example.)
 858             for (;;) {
 859                 c.fChar = nextCharLL();
 860                 if (c.fChar == (UChar32)-1 ||  // EOF
 861                     c.fChar == chCR     ||
 862                     c.fChar == chLF     ||
 863                     c.fChar == chNEL    ||
 864                     c.fChar == chLS)       {break;}
 865             }
 866         }
 867         if (c.fChar == (UChar32)-1) {
 868             return;
 869         }
 870
 871         //
 872         //  check for backslash escaped characters.
 873         //  Use UnicodeString::unescapeAt() to handle them.
 874         //
 875         if (c.fChar == chBackSlash) {
 876             c.fEscaped = TRUE;
 877             int32_t startX = fNextIndex;
 878             c.fChar = fRB->fRules.unescapeAt(fNextIndex);
 879             if (fNextIndex == startX) {
 880                 error(U_BRK_HEX_DIGITS_EXPECTED);
 881             }
 882             fCharNum += fNextIndex-startX;
 883         }
 884     }
 885     // putc(c.fChar, stdout);
 886 }
 887
 888 //---------------------------------------------------------------------------------
 889 //
 890 //  Parse RBBI rules.   The state machine for rules parsing is here.
 891 //                      The state tables are hand-written in the file rbbirpt.txt,
 892 //                      and converted to the form used here by a perl
 893 //                      script rbbicst.pl
 894 //
 895 //---------------------------------------------------------------------------------
 896 void RBBIRuleScanner::parse() {
 897     uint16_t                state;
 898     const RBBIRuleTableEl  *tableEl;
 899
 900     if (U_FAILURE(*fRB->fStatus)) {
 901         return;
 902     }
 903
 904     state = 1;
 905     nextChar(fC);
 906     //
 907     // Main loop for the rule parsing state machine.
 908     //   Runs once per state transition.
 909     //   Each time through optionally performs, depending on the state table,
 910     //      - an advance to the the next input char
 911     //      - an action to be performed.
 912     //      - pushing or popping a state to/from the local state return stack.
 913     //
 914     for (;;) {
 915         //  Bail out if anything has gone wrong.
 916         //  RBBI rule file parsing stops on the first error encountered.
 917         if (U_FAILURE(*fRB->fStatus)) {
 918             break;
 919         }
 920
 921         // Quit if state == 0.  This is the normal way to exit the state machine.
 922         //
 923         if (state == 0) {
 924             break;
 925         }
 926
 927         // Find the state table element that matches the input char from the rule, or the
 928         //    class of the input character.  Start with the first table row for this
 929         //    state, then linearly scan forward until we find a row that matches the
 930         //    character.  The last row for each state always matches all characters, so
 931         //    the search will stop there, if not before.
 932         //
 933         tableEl = &gRuleParseStateTable[state];
 934         #ifdef RBBI_DEBUG
 935             if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "scan")) {
 936                 RBBIDebugPrintf("char, line, col = (\'%c\', %d, %d)    state=%s ",
 937                     fC.fChar, fLineNum, fCharNum, RBBIRuleStateNames[state]);
 938             }
 939         #endif
 940
 941         for (;;) {
 942             if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "scan")) { RBBIDebugPrintf(".");}
 943             if (tableEl->fCharClass < 127 && fC.fEscaped == FALSE &&   tableEl->fCharClass == fC.fChar) {
 944                 // Table row specified an individual character, not a set, and
 945                 //   the input character is not escaped, and
 946                 //   the input character matched it.
 947                 break;
 948             }
 949             if (tableEl->fCharClass == 255) {
 950                 // Table row specified default, match anything character class.
 951                 break;
 952             }
 953             if (tableEl->fCharClass == 254 && fC.fEscaped)  {
 954                 // Table row specified "escaped" and the char was escaped.
 955                 break;
 956             }
 957             if (tableEl->fCharClass == 253 && fC.fEscaped &&
 958                 (fC.fChar == 0x50 || fC.fChar == 0x70 ))  {
 959                 // Table row specified "escaped P" and the char is either 'p' or 'P'.
 960                 break;
 961             }
 962             if (tableEl->fCharClass == 252 && fC.fChar == (UChar32)-1)  {
 963                 // Table row specified eof and we hit eof on the input.
 964                 break;
 965             }
 966
 967             if (tableEl->fCharClass >= 128 && tableEl->fCharClass < 240 &&   // Table specs a char class &&
 968                 fC.fEscaped == FALSE &&                                      //   char is not escaped &&
 969                 fC.fChar != (UChar32)-1) {                                   //   char is not EOF
 970                 UnicodeSet *uniset = fRuleSets[tableEl->fCharClass-128];
 971                 if (uniset->contains(fC.fChar)) {
 972                     // Table row specified a character class, or set of characters,
 973                     //   and the current char matches it.
 974                     break;
 975                 }
 976             }
 977
 978             // No match on this row, advance to the next  row for this state,
 979             tableEl++;
 980         }
 981         if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "scan")) { RBBIDebugPrintf("\n");}
 982
 983         //
 984         // We've found the row of the state table that matches the current input
 985         //   character from the rules string.
 986         // Perform any action specified  by this row in the state table.
 987         if (doParseActions((EParseAction)tableEl->fAction) == FALSE) {
 988             // Break out of the state machine loop if the
 989             //   the action signalled some kind of error, or
 990             //   the action was to exit, occurs on normal end-of-rules-input.
 991             break;
 992         }
 993
 994         if (tableEl->fPushState != 0) {
 995             fStackPtr++;
 996             if (fStackPtr >= kStackSize) {
 997                 error(U_BRK_INTERNAL_ERROR);
 998                 RBBIDebugPuts("RBBIRuleScanner::parse() - state stack overflow.");
 999                 fStackPtr--;
1000             }
1001             fStack[fStackPtr] = tableEl->fPushState;
1002         }
1003
1004         if (tableEl->fNextChar) {
1005             nextChar(fC);
1006         }
1007
1008         // Get the next state from the table entry, or from the
1009         //   state stack if the next state was specified as "pop".
1010         if (tableEl->fNextState != 255) {
1011             state = tableEl->fNextState;
1012         } else {
1013             state = fStack[fStackPtr];
1014             fStackPtr--;
1015             if (fStackPtr < 0) {
1016                 error(U_BRK_INTERNAL_ERROR);
1017                 RBBIDebugPuts("RBBIRuleScanner::parse() - state stack underflow.");
1018                 fStackPtr++;
1019             }
1020         }
1021
1022     }
1023
1024     //
1025     // If there were NO user specified reverse rules, set up the equivalent of ".*;"
1026     //
1027     if (fRB->fReverseTree == NULL) {
1028         fRB->fReverseTree  = pushNewNode(RBBINode::opStar);
1029         RBBINode  *operand = pushNewNode(RBBINode::setRef);
1030         findSetFor(kAny, operand);
1031         fRB->fReverseTree->fLeftChild = operand;
1032         operand->fParent              = fRB->fReverseTree;
1033         fNodeStackPtr -= 2;
1034     }
1035
1036
1037     //
1038     // Parsing of the input RBBI rules is complete.
1039     // We now have a parse tree for the rule expressions
1040     // and a list of all UnicodeSets that are referenced.
1041     //
1042 #ifdef RBBI_DEBUG
1043     if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "symbols")) {fSymbolTable->rbbiSymtablePrint();}
1044     if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "ptree"))
1045     {
1046         RBBIDebugPrintf("Completed Forward Rules Parse Tree...\n");
1047         fRB->fForwardTree->printTree(TRUE);
1048         RBBIDebugPrintf("\nCompleted Reverse Rules Parse Tree...\n");
1049         fRB->fReverseTree->printTree(TRUE);
1050         RBBIDebugPrintf("\nCompleted Safe Point Forward Rules Parse Tree...\n");
1051         fRB->fSafeFwdTree->printTree(TRUE);
1052         RBBIDebugPrintf("\nCompleted Safe Point Reverse Rules Parse Tree...\n");
1053         fRB->fSafeRevTree->printTree(TRUE);
1054     }
1055 #endif
1056 }
1057
1058
1059 //---------------------------------------------------------------------------------
1060 //
1061 //  printNodeStack     for debugging...
1062 //
1063 //---------------------------------------------------------------------------------
1064 #ifdef RBBI_DEBUG
1065 void RBBIRuleScanner::printNodeStack(const char *) {}
1066 #else
1067 void RBBIRuleScanner::printNodeStack(const char *title) {
1068     int i;
1069     RBBIDebugPrintf("%s.  Dumping node stack...\n", title);
1070     for (i=fNodeStackPtr; i>0; i--) {fNodeStack[i]->printTree(TRUE);}
1071 }
1072 #endif
1073
1074
1075
1076
1077 //---------------------------------------------------------------------------------
1078 //
1079 //  pushNewNode   create a new RBBINode of the specified type and push it
1080 //                onto the stack of nodes.
1081 //
1082 //---------------------------------------------------------------------------------
1083 RBBINode  *RBBIRuleScanner::pushNewNode(RBBINode::NodeType  t) {
1084     fNodeStackPtr++;
1085     if (fNodeStackPtr >= kStackSize) {
1086         error(U_BRK_INTERNAL_ERROR);
1087         RBBIDebugPuts("RBBIRuleScanner::pushNewNode - stack overflow.");
1088         *fRB->fStatus = U_BRK_INTERNAL_ERROR;
1089         return NULL;
1090     }
1091     fNodeStack[fNodeStackPtr] = new RBBINode(t);
1092     if (fNodeStack[fNodeStackPtr] == NULL) {
1093         *fRB->fStatus = U_MEMORY_ALLOCATION_ERROR;
1094     }
1095     return fNodeStack[fNodeStackPtr];
1096 }
1097
1098
1099
1100 //---------------------------------------------------------------------------------
1101 //
1102 //  scanSet    Construct a UnicodeSet from the text at the current scan
1103 //             position.  Advance the scan position to the first character
1104 //             after the set.
1105 //
1106 //             A new RBBI setref node referring to the set is pushed onto the node
1107 //             stack.
1108 //
1109 //             The scan position is normally under the control of the state machine
1110 //             that controls rule parsing.  UnicodeSets, however, are parsed by
1111 //             the UnicodeSet constructor, not by the RBBI rule parser.
1112 //
1113 //---------------------------------------------------------------------------------
1114 void RBBIRuleScanner::scanSet() {
1115     UnicodeSet    *uset;
1116     ParsePosition  pos;
1117     int            startPos;
1118     int            i;
1119
1120     if (U_FAILURE(*fRB->fStatus)) {
1121         return;
1122     }
1123
1124     pos.setIndex(fScanIndex);
1125     startPos = fScanIndex;
1126     UErrorCode localStatus = U_ZERO_ERROR;
1127     uset = new UnicodeSet(fRB->fRules, pos, USET_IGNORE_SPACE,
1128                          fSymbolTable,
1129                          localStatus);
1130     if (U_FAILURE(localStatus)) {
1131         //  TODO:  Get more accurate position of the error from UnicodeSet's return info.
1132         //         UnicodeSet appears to not be reporting correctly at this time.
1133         RBBIDebugPrintf("UnicodeSet parse postion.ErrorIndex = %d\n", pos.getIndex());
1134         error(localStatus);
1135         delete uset;
1136         return;
1137     }
1138
1139     // Verify that the set contains at least one code point.
1140     //
1141     if (uset->charAt(0) == -1) {
1142         // This set is empty.
1143         //  Make it an error, because it almost certainly is not what the user wanted.
1144         //  Also, avoids having to think about corner cases in the tree manipulation code
1145         //   that occurs later on.
1146         error(U_BRK_RULE_EMPTY_SET);
1147         delete uset;
1148         return;
1149     }
1150
1151
1152     // Advance the RBBI parse postion over the UnicodeSet pattern.
1153     //   Don't just set fScanIndex because the line/char positions maintained
1154     //   for error reporting would be thrown off.
1155     i = pos.getIndex();
1156     for (;;) {
1157         if (fNextIndex >= i) {
1158             break;
1159         }
1160         nextCharLL();
1161     }
1162
1163     if (U_SUCCESS(*fRB->fStatus)) {
1164         RBBINode         *n;
1165
1166         n = pushNewNode(RBBINode::setRef);
1167         n->fFirstPos = startPos;
1168         n->fLastPos  = fNextIndex;
1169         fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
1170         //  findSetFor() serves several purposes here:
1171         //     - Adopts storage for the UnicodeSet, will be responsible for deleting.
1172         //     - Mantains collection of all sets in use, needed later for establishing
1173         //          character categories for run time engine.
1174         //     - Eliminates mulitiple instances of the same set.
1175         //     - Creates a new uset node if necessary (if this isn't a duplicate.)
1176         findSetFor(n->fText, n, uset);
1177     }
1178
1179 }
1180
1181 U_NAMESPACE_END
1182
1183 #endif /* #if !UCONFIG_NO_BREAK_ITERATION */