icuSources/common/rbbiscan57.cpp

   1 // © 2016 and later: Unicode, Inc. and others.
   2 // License & terms of use: http://www.unicode.org/copyright.html
   3 //
   4 //  Copyright (C) 2002-2016, International Business Machines Corporation and others.
   5 //  All Rights Reserved.
   6 //
   7 //  This file contains the Rule Based Break Iterator Rule Builder functions for
   8 //   scanning the rules and assembling a parse tree.  This is the first phase
   9 //   of compiling the rules.
  10 //
  11 //  The overall of the rules is managed by class RBBIRuleBuilder, which will
  12 //  create and use an instance of this class as part of the process.
  13 //
  14
  15 #include "unicode/utypes.h"
  16
  17 #if !UCONFIG_NO_BREAK_ITERATION
  18
  19 #include "unicode/unistr.h"
  20 #include "unicode/uniset.h"
  21 #include "unicode/uchar.h"
  22 #include "unicode/uchriter.h"
  23 #include "unicode/parsepos.h"
  24 #include "unicode/parseerr.h"
  25 #include "cmemory.h"
  26 #include "cstring.h"
  27
  28 #include "rbbirpt.h"   // Contains state table for the rbbi rules parser.
  29                        //   generated by a Perl script.
  30 #include "rbbirb57.h"
  31 #include "rbbinode.h"
  32 #include "rbbiscan57.h"
  33 #include "rbbitblb57.h"
  34
  35 #include "uassert.h"
  36
  37 //------------------------------------------------------------------------------
  38 //
  39 // Unicode Set init strings for each of the character classes needed for parsing a rule file.
  40 //               (Initialized with hex values for portability to EBCDIC based machines.
  41 //                Really ugly, but there's no good way to avoid it.)
  42 //
  43 //              The sets are referred to by name in the rbbirpt.txt, which is the
  44 //              source form of the state transition table for the RBBI rule parser.
  45 //
  46 //------------------------------------------------------------------------------
  47 static const UChar gRuleSet_rule_char_pattern[]       = {
  48  //   [    ^      [    \     p     {      Z     }     \     u    0      0    2      0
  49     0x5b, 0x5e, 0x5b, 0x5c, 0x70, 0x7b, 0x5a, 0x7d, 0x5c, 0x75, 0x30, 0x30, 0x32, 0x30,
  50  //   -    \      u    0     0     7      f     ]     -     [    \      p
  51     0x2d, 0x5c, 0x75, 0x30, 0x30, 0x37, 0x66, 0x5d, 0x2d, 0x5b, 0x5c, 0x70,
  52  //   {     L     }    ]     -     [      \     p     {     N    }      ]     ]
  53     0x7b, 0x4c, 0x7d, 0x5d, 0x2d, 0x5b, 0x5c, 0x70, 0x7b, 0x4e, 0x7d, 0x5d, 0x5d, 0};
  54
  55 static const UChar gRuleSet_name_char_pattern[]       = {
  56 //    [    _      \    p     {     L      }     \     p     {    N      }     ]
  57     0x5b, 0x5f, 0x5c, 0x70, 0x7b, 0x4c, 0x7d, 0x5c, 0x70, 0x7b, 0x4e, 0x7d, 0x5d, 0};
  58
  59 static const UChar gRuleSet_digit_char_pattern[] = {
  60 //    [    0      -    9     ]
  61     0x5b, 0x30, 0x2d, 0x39, 0x5d, 0};
  62
  63 static const UChar gRuleSet_name_start_char_pattern[] = {
  64 //    [    _      \    p     {     L      }     ]
  65     0x5b, 0x5f, 0x5c, 0x70, 0x7b, 0x4c, 0x7d, 0x5d, 0 };
  66
  67 static const UChar kAny[] = {0x61, 0x6e, 0x79, 0x00};  // "any"
  68
  69
  70 U_CDECL_BEGIN
  71 static void U_CALLCONV RBBISetTable_deleter(void *p) {
  72     icu::RBBISetTableEl *px = (icu::RBBISetTableEl *)p;
  73     delete px->key;
  74     // Note:  px->val is owned by the linked list "fSetsListHead" in scanner.
  75     //        Don't delete the value nodes here.
  76     uprv_free(px);
  77 }
  78 U_CDECL_END
  79
  80 U_NAMESPACE_BEGIN
  81
  82 //------------------------------------------------------------------------------
  83 //
  84 //  Constructor.
  85 //
  86 //------------------------------------------------------------------------------
  87 RBBIRuleScanner57::RBBIRuleScanner57(RBBIRuleBuilder57 *rb)
  88 {
  89     fRB                 = rb;
  90     fScanIndex          = 0;
  91     fNextIndex          = 0;
  92     fQuoteMode          = FALSE;
  93     fLineNum            = 1;
  94     fCharNum            = 0;
  95     fLastChar           = 0;
  96
  97     fStateTable         = NULL;
  98     fStack[0]           = 0;
  99     fStackPtr           = 0;
 100     fNodeStack[0]       = NULL;
 101     fNodeStackPtr       = 0;
 102
 103     fReverseRule        = FALSE;
 104     fLookAheadRule      = FALSE;
 105     fNoChainInRule      = FALSE;
 106
 107     fSymbolTable        = NULL;
 108     fSetTable           = NULL;
 109     fRuleNum            = 0;
 110     fOptionStart        = 0;
 111
 112     // Do not check status until after all critical fields are sufficiently initialized
 113     //   that the destructor can run cleanly.
 114     if (U_FAILURE(*rb->fStatus)) {
 115         return;
 116     }
 117
 118     //
 119     //  Set up the constant Unicode Sets.
 120     //     Note:  These could be made static, lazily initialized, and shared among
 121     //            all instances of RBBIRuleScanners.  BUT this is quite a bit simpler,
 122     //            and the time to build these few sets should be small compared to a
 123     //            full break iterator build.
 124     fRuleSets[kRuleSet_rule_char-128]
 125         = UnicodeSet(UnicodeString(gRuleSet_rule_char_pattern),       *rb->fStatus);
 126     // fRuleSets[kRuleSet_white_space-128] = [:Pattern_White_Space:]
 127     fRuleSets[kRuleSet_white_space-128].
 128         add(9, 0xd).add(0x20).add(0x85).add(0x200e, 0x200f).add(0x2028, 0x2029);
 129     fRuleSets[kRuleSet_name_char-128]
 130         = UnicodeSet(UnicodeString(gRuleSet_name_char_pattern),       *rb->fStatus);
 131     fRuleSets[kRuleSet_name_start_char-128]
 132         = UnicodeSet(UnicodeString(gRuleSet_name_start_char_pattern), *rb->fStatus);
 133     fRuleSets[kRuleSet_digit_char-128]
 134         = UnicodeSet(UnicodeString(gRuleSet_digit_char_pattern),      *rb->fStatus);
 135     if (*rb->fStatus == U_ILLEGAL_ARGUMENT_ERROR) {
 136         // This case happens if ICU's data is missing.  UnicodeSet tries to look up property
 137         //   names from the init string, can't find them, and claims an illegal argument.
 138         //   Change the error so that the actual problem will be clearer to users.
 139         *rb->fStatus = U_BRK_INIT_ERROR;
 140     }
 141     if (U_FAILURE(*rb->fStatus)) {
 142         return;
 143     }
 144
 145     fSymbolTable = new RBBISymbolTable57(this, rb->fRules, *rb->fStatus);
 146     if (fSymbolTable == NULL) {
 147         *rb->fStatus = U_MEMORY_ALLOCATION_ERROR;
 148         return;
 149     }
 150     fSetTable    = uhash_open(uhash_hashUnicodeString, uhash_compareUnicodeString, NULL, rb->fStatus);
 151     if (U_FAILURE(*rb->fStatus)) {
 152         return;
 153     }
 154     uhash_setValueDeleter(fSetTable, RBBISetTable_deleter);
 155 }
 156
 157
 158
 159 //------------------------------------------------------------------------------
 160 //
 161 //  Destructor
 162 //
 163 //------------------------------------------------------------------------------
 164 RBBIRuleScanner57::~RBBIRuleScanner57() {
 165     delete fSymbolTable;
 166     if (fSetTable != NULL) {
 167          uhash_close(fSetTable);
 168          fSetTable = NULL;
 169
 170     }
 171
 172
 173     // Node Stack.
 174     //   Normally has one entry, which is the entire parse tree for the rules.
 175     //   If errors occured, there may be additional subtrees left on the stack.
 176     while (fNodeStackPtr > 0) {
 177         delete fNodeStack[fNodeStackPtr];
 178         fNodeStackPtr--;
 179     }
 180
 181 }
 182
 183 //------------------------------------------------------------------------------
 184 //
 185 //  doParseAction        Do some action during rule parsing.
 186 //                       Called by the parse state machine.
 187 //                       Actions build the parse tree and Unicode Sets,
 188 //                       and maintain the parse stack for nested expressions.
 189 //
 190 //                       TODO:  unify EParseAction and RBBI_RuleParseAction enum types.
 191 //                              They represent exactly the same thing.  They're separate
 192 //                              only to work around enum forward declaration restrictions
 193 //                              in some compilers, while at the same time avoiding multiple
 194 //                              definitions problems.  I'm sure that there's a better way.
 195 //
 196 //------------------------------------------------------------------------------
 197 UBool RBBIRuleScanner57::doParseActions(int32_t action)
 198 {
 199     RBBINode *n       = NULL;
 200
 201     UBool   returnVal = TRUE;
 202
 203     switch (action) {
 204
 205     case doExprStart:
 206         pushNewNode(RBBINode::opStart);
 207         fRuleNum++;
 208         break;
 209
 210
 211     case doNoChain:
 212         // Scanned a '^' while on the rule start state.
 213         fNoChainInRule = TRUE;
 214         break;
 215
 216
 217     case doExprOrOperator:
 218         {
 219             fixOpStack(RBBINode::precOpCat);
 220             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 221             RBBINode  *orNode      = pushNewNode(RBBINode::opOr);
 222             if (U_FAILURE(*fRB->fStatus)) {
 223                 break;
 224             }
 225             orNode->fLeftChild     = operandNode;
 226             operandNode->fParent   = orNode;
 227         }
 228         break;
 229
 230     case doExprCatOperator:
 231         // concatenation operator.
 232         // For the implicit concatenation of adjacent terms in an expression that are
 233         //   not separated by any other operator.  Action is invoked between the
 234         //   actions for the two terms.
 235         {
 236             fixOpStack(RBBINode::precOpCat);
 237             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 238             RBBINode  *catNode     = pushNewNode(RBBINode::opCat);
 239             if (U_FAILURE(*fRB->fStatus)) {
 240                 break;
 241             }
 242             catNode->fLeftChild    = operandNode;
 243             operandNode->fParent   = catNode;
 244         }
 245         break;
 246
 247     case doLParen:
 248         // Open Paren.
 249         //   The openParen node is a dummy operation type with a low precedence,
 250         //     which has the affect of ensuring that any real binary op that
 251         //     follows within the parens binds more tightly to the operands than
 252         //     stuff outside of the parens.
 253         pushNewNode(RBBINode::opLParen);
 254         break;
 255
 256     case doExprRParen:
 257         fixOpStack(RBBINode::precLParen);
 258         break;
 259
 260     case doNOP:
 261         break;
 262
 263     case doStartAssign:
 264         // We've just scanned "$variable = "
 265         // The top of the node stack has the $variable ref node.
 266
 267         // Save the start position of the RHS text in the StartExpression node
 268         //   that precedes the $variableReference node on the stack.
 269         //   This will eventually be used when saving the full $variable replacement
 270         //   text as a string.
 271         n = fNodeStack[fNodeStackPtr-1];
 272         n->fFirstPos = fNextIndex;              // move past the '='
 273
 274         // Push a new start-of-expression node; needed to keep parse of the
 275         //   RHS expression happy.
 276         pushNewNode(RBBINode::opStart);
 277         break;
 278
 279
 280
 281
 282     case doEndAssign:
 283         {
 284             // We have reached the end of an assignement statement.
 285             //   Current scan char is the ';' that terminates the assignment.
 286
 287             // Terminate expression, leaves expression parse tree rooted in TOS node.
 288             fixOpStack(RBBINode::precStart);
 289
 290             RBBINode *startExprNode  = fNodeStack[fNodeStackPtr-2];
 291             RBBINode *varRefNode     = fNodeStack[fNodeStackPtr-1];
 292             RBBINode *RHSExprNode    = fNodeStack[fNodeStackPtr];
 293
 294             // Save original text of right side of assignment, excluding the terminating ';'
 295             //  in the root of the node for the right-hand-side expression.
 296             RHSExprNode->fFirstPos = startExprNode->fFirstPos;
 297             RHSExprNode->fLastPos  = fScanIndex;
 298             fRB->fRules.extractBetween(RHSExprNode->fFirstPos, RHSExprNode->fLastPos, RHSExprNode->fText);
 299
 300             // Expression parse tree becomes l. child of the $variable reference node.
 301             varRefNode->fLeftChild = RHSExprNode;
 302             RHSExprNode->fParent   = varRefNode;
 303
 304             // Make a symbol table entry for the $variableRef node.
 305             fSymbolTable->addEntry(varRefNode->fText, varRefNode, *fRB->fStatus);
 306             if (U_FAILURE(*fRB->fStatus)) {
 307                 // This is a round-about way to get the parse position set
 308                 //  so that duplicate symbols error messages include a line number.
 309                 UErrorCode t = *fRB->fStatus;
 310                 *fRB->fStatus = U_ZERO_ERROR;
 311                 error(t);
 312             }
 313
 314             // Clean up the stack.
 315             delete startExprNode;
 316             fNodeStackPtr-=3;
 317             break;
 318         }
 319
 320     case doEndOfRule:
 321         {
 322         fixOpStack(RBBINode::precStart);      // Terminate expression, leaves expression
 323         if (U_FAILURE(*fRB->fStatus)) {       //   parse tree rooted in TOS node.
 324             break;
 325         }
 326 #ifdef RBBI_DEBUG
 327         if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "rtree")) {printNodeStack("end of rule");}
 328 #endif
 329         U_ASSERT(fNodeStackPtr == 1);
 330         RBBINode *thisRule = fNodeStack[fNodeStackPtr];
 331
 332         // If this rule includes a look-ahead '/', add a endMark node to the
 333         //   expression tree.
 334         if (fLookAheadRule) {
 335             RBBINode  *endNode        = pushNewNode(RBBINode::endMark);
 336             RBBINode  *catNode        = pushNewNode(RBBINode::opCat);
 337             if (U_FAILURE(*fRB->fStatus)) {
 338                 break;
 339             }
 340             fNodeStackPtr -= 2;
 341             catNode->fLeftChild       = thisRule;
 342             catNode->fRightChild      = endNode;
 343             fNodeStack[fNodeStackPtr] = catNode;
 344             endNode->fVal             = fRuleNum;
 345             endNode->fLookAheadEnd    = TRUE;
 346             thisRule                  = catNode;
 347
 348             // TODO: Disable chaining out of look-ahead (hard break) rules.
 349             //   The break on rule match is forced, so there is no point in building up
 350             //   the state table to chain into another rule for a longer match.
 351         }
 352
 353         // Mark this node as being the root of a rule.
 354         thisRule->fRuleRoot = TRUE;
 355
 356         // Flag if chaining into this rule is wanted.
 357         //
 358         if (fRB->fChainRules &&         // If rule chaining is enabled globally via !!chain
 359                 !fNoChainInRule) {      //     and no '^' chain-in inhibit was on this rule
 360             thisRule->fChainIn = TRUE;
 361         }
 362
 363
 364         // All rule expressions are ORed together.
 365         // The ';' that terminates an expression really just functions as a '|' with
 366         //   a low operator prededence.
 367         //
 368         // Each of the four sets of rules are collected separately.
 369         //  (forward, reverse, safe_forward, safe_reverse)
 370         //  OR this rule into the appropriate group of them.
 371         //
 372         RBBINode **destRules = (fReverseRule? &fRB->fReverseTree : fRB->fDefaultTree);
 373
 374         if (*destRules != NULL) {
 375             // This is not the first rule encounted.
 376             // OR previous stuff  (from *destRules)
 377             // with the current rule expression (on the Node Stack)
 378             //  with the resulting OR expression going to *destRules
 379             //
 380             RBBINode  *thisRule    = fNodeStack[fNodeStackPtr];
 381             RBBINode  *prevRules   = *destRules;
 382             RBBINode  *orNode      = pushNewNode(RBBINode::opOr);
 383             if (U_FAILURE(*fRB->fStatus)) {
 384                 break;
 385             }
 386             orNode->fLeftChild     = prevRules;
 387             prevRules->fParent     = orNode;
 388             orNode->fRightChild    = thisRule;
 389             thisRule->fParent      = orNode;
 390             *destRules             = orNode;
 391         }
 392         else
 393         {
 394             // This is the first rule encountered (for this direction).
 395             // Just move its parse tree from the stack to *destRules.
 396             *destRules = fNodeStack[fNodeStackPtr];
 397         }
 398         fReverseRule   = FALSE;   // in preparation for the next rule.
 399         fLookAheadRule = FALSE;
 400         fNoChainInRule = FALSE;
 401         fNodeStackPtr  = 0;
 402         }
 403         break;
 404
 405
 406     case doRuleError:
 407         error(U_BRK_RULE_SYNTAX);
 408         returnVal = FALSE;
 409         break;
 410
 411
 412     case doVariableNameExpectedErr:
 413         error(U_BRK_RULE_SYNTAX);
 414         break;
 415
 416
 417     //
 418     //  Unary operands  + ? *
 419     //    These all appear after the operand to which they apply.
 420     //    When we hit one, the operand (may be a whole sub expression)
 421     //    will be on the top of the stack.
 422     //    Unary Operator becomes TOS, with the old TOS as its one child.
 423     case doUnaryOpPlus:
 424         {
 425             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 426             RBBINode  *plusNode    = pushNewNode(RBBINode::opPlus);
 427             if (U_FAILURE(*fRB->fStatus)) {
 428                 break;
 429             }
 430             plusNode->fLeftChild   = operandNode;
 431             operandNode->fParent   = plusNode;
 432         }
 433         break;
 434
 435     case doUnaryOpQuestion:
 436         {
 437             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 438             RBBINode  *qNode       = pushNewNode(RBBINode::opQuestion);
 439             if (U_FAILURE(*fRB->fStatus)) {
 440                 break;
 441             }
 442             qNode->fLeftChild      = operandNode;
 443             operandNode->fParent   = qNode;
 444         }
 445         break;
 446
 447     case doUnaryOpStar:
 448         {
 449             RBBINode  *operandNode = fNodeStack[fNodeStackPtr--];
 450             RBBINode  *starNode    = pushNewNode(RBBINode::opStar);
 451             if (U_FAILURE(*fRB->fStatus)) {
 452                 break;
 453             }
 454             starNode->fLeftChild   = operandNode;
 455             operandNode->fParent   = starNode;
 456         }
 457         break;
 458
 459     case doRuleChar:
 460         // A "Rule Character" is any single character that is a literal part
 461         // of the regular expression.  Like a, b and c in the expression "(abc*) | [:L:]"
 462         // These are pretty uncommon in break rules; the terms are more commonly
 463         //  sets.  To keep things uniform, treat these characters like as
 464         // sets that just happen to contain only one character.
 465         {
 466             n = pushNewNode(RBBINode::setRef);
 467             if (U_FAILURE(*fRB->fStatus)) {
 468                 break;
 469             }
 470             findSetFor(UnicodeString(fC.fChar), n);
 471             n->fFirstPos = fScanIndex;
 472             n->fLastPos  = fNextIndex;
 473             fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
 474             break;
 475         }
 476
 477     case doDotAny:
 478         // scanned a ".", meaning match any single character.
 479         {
 480             n = pushNewNode(RBBINode::setRef);
 481             if (U_FAILURE(*fRB->fStatus)) {
 482                 break;
 483             }
 484             findSetFor(UnicodeString(TRUE, kAny, 3), n);
 485             n->fFirstPos = fScanIndex;
 486             n->fLastPos  = fNextIndex;
 487             fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
 488             break;
 489         }
 490
 491     case doSlash:
 492         // Scanned a '/', which identifies a look-ahead break position in a rule.
 493         n = pushNewNode(RBBINode::lookAhead);
 494         if (U_FAILURE(*fRB->fStatus)) {
 495             break;
 496         }
 497         n->fVal      = fRuleNum;
 498         n->fFirstPos = fScanIndex;
 499         n->fLastPos  = fNextIndex;
 500         fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
 501         fLookAheadRule = TRUE;
 502         break;
 503
 504
 505     case doStartTagValue:
 506         // Scanned a '{', the opening delimiter for a tag value within a rule.
 507         n = pushNewNode(RBBINode::tag);
 508         if (U_FAILURE(*fRB->fStatus)) {
 509             break;
 510         }
 511         n->fVal      = 0;
 512         n->fFirstPos = fScanIndex;
 513         n->fLastPos  = fNextIndex;
 514         break;
 515
 516     case doTagDigit:
 517         // Just scanned a decimal digit that's part of a tag value
 518         {
 519             n = fNodeStack[fNodeStackPtr];
 520             uint32_t v = u_charDigitValue(fC.fChar);
 521             U_ASSERT(v < 10);
 522             n->fVal = n->fVal*10 + v;
 523             break;
 524         }
 525
 526     case doTagValue:
 527         n = fNodeStack[fNodeStackPtr];
 528         n->fLastPos = fNextIndex;
 529         fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
 530         break;
 531
 532     case doTagExpectedError:
 533         error(U_BRK_MALFORMED_RULE_TAG);
 534         returnVal = FALSE;
 535         break;
 536
 537     case doOptionStart:
 538         // Scanning a !!option.   At the start of string.
 539         fOptionStart = fScanIndex;
 540         break;
 541
 542     case doOptionEnd:
 543         {
 544             UnicodeString opt(fRB->fRules, fOptionStart, fScanIndex-fOptionStart);
 545             if (opt == UNICODE_STRING("chain", 5)) {
 546                 fRB->fChainRules = TRUE;
 547             } else if (opt == UNICODE_STRING("LBCMNoChain", 11)) {
 548                 fRB->fLBCMNoChain = TRUE;
 549             } else if (opt == UNICODE_STRING("RINoChain", 9)) {
 550                 fRB->fRINoChain = TRUE;
 551             } else if (opt == UNICODE_STRING("forward", 7)) {
 552                 fRB->fDefaultTree   = &fRB->fForwardTree;
 553             } else if (opt == UNICODE_STRING("reverse", 7)) {
 554                 fRB->fDefaultTree   = &fRB->fReverseTree;
 555             } else if (opt == UNICODE_STRING("safe_forward", 12)) {
 556                 fRB->fDefaultTree   = &fRB->fSafeFwdTree;
 557             } else if (opt == UNICODE_STRING("safe_reverse", 12)) {
 558                 fRB->fDefaultTree   = &fRB->fSafeRevTree;
 559             } else if (opt == UNICODE_STRING("lookAheadHardBreak", 18)) {
 560                 fRB->fLookAheadHardBreak = TRUE;
 561             } else {
 562                 error(U_BRK_UNRECOGNIZED_OPTION);
 563             }
 564         }
 565         break;
 566
 567     case doReverseDir:
 568         fReverseRule = TRUE;
 569         break;
 570
 571     case doStartVariableName:
 572         n = pushNewNode(RBBINode::varRef);
 573         if (U_FAILURE(*fRB->fStatus)) {
 574             break;
 575         }
 576         n->fFirstPos = fScanIndex;
 577         break;
 578
 579     case doEndVariableName:
 580         n = fNodeStack[fNodeStackPtr];
 581         if (n==NULL || n->fType != RBBINode::varRef) {
 582             error(U_BRK_INTERNAL_ERROR);
 583             break;
 584         }
 585         n->fLastPos = fScanIndex;
 586         fRB->fRules.extractBetween(n->fFirstPos+1, n->fLastPos, n->fText);
 587         // Look the newly scanned name up in the symbol table
 588         //   If there's an entry, set the l. child of the var ref to the replacement expression.
 589         //   (We also pass through here when scanning assignments, but no harm is done, other
 590         //    than a slight wasted effort that seems hard to avoid.  Lookup will be null)
 591         n->fLeftChild = fSymbolTable->lookupNode(n->fText);
 592         break;
 593
 594     case doCheckVarDef:
 595         n = fNodeStack[fNodeStackPtr];
 596         if (n->fLeftChild == NULL) {
 597             error(U_BRK_UNDEFINED_VARIABLE);
 598             returnVal = FALSE;
 599         }
 600         break;
 601
 602     case doExprFinished:
 603         break;
 604
 605     case doRuleErrorAssignExpr:
 606         error(U_BRK_ASSIGN_ERROR);
 607         returnVal = FALSE;
 608         break;
 609
 610     case doExit:
 611         returnVal = FALSE;
 612         break;
 613
 614     case doScanUnicodeSet:
 615         scanSet();
 616         break;
 617
 618     default:
 619         error(U_BRK_INTERNAL_ERROR);
 620         returnVal = FALSE;
 621         break;
 622     }
 623     return returnVal && U_SUCCESS(*fRB->fStatus);
 624 }
 625
 626
 627
 628
 629 //------------------------------------------------------------------------------
 630 //
 631 //  Error         Report a rule parse error.
 632 //                Only report it if no previous error has been recorded.
 633 //
 634 //------------------------------------------------------------------------------
 635 void RBBIRuleScanner57::error(UErrorCode e) {
 636     if (U_SUCCESS(*fRB->fStatus)) {
 637         *fRB->fStatus = e;
 638         if (fRB->fParseError) {
 639             fRB->fParseError->line  = fLineNum;
 640             fRB->fParseError->offset = fCharNum;
 641             fRB->fParseError->preContext[0] = 0;
 642             fRB->fParseError->postContext[0] = 0;
 643         }
 644     }
 645 }
 646
 647
 648
 649
 650 //------------------------------------------------------------------------------
 651 //
 652 //  fixOpStack   The parse stack holds partially assembled chunks of the parse tree.
 653 //               An entry on the stack may be as small as a single setRef node,
 654 //               or as large as the parse tree
 655 //               for an entire expression (this will be the one item left on the stack
 656 //               when the parsing of an RBBI rule completes.
 657 //
 658 //               This function is called when a binary operator is encountered.
 659 //               It looks back up the stack for operators that are not yet associated
 660 //               with a right operand, and if the precedence of the stacked operator >=
 661 //               the precedence of the current operator, binds the operand left,
 662 //               to the previously encountered operator.
 663 //
 664 //------------------------------------------------------------------------------
 665 void RBBIRuleScanner57::fixOpStack(RBBINode::OpPrecedence p) {
 666     RBBINode *n;
 667     // printNodeStack("entering fixOpStack()");
 668     for (;;) {
 669         n = fNodeStack[fNodeStackPtr-1];   // an operator node
 670         if (n->fPrecedence == 0) {
 671             RBBIDebugPuts("RBBIRuleScanner57::fixOpStack, bad operator node");
 672             error(U_BRK_INTERNAL_ERROR);
 673             return;
 674         }
 675
 676         if (n->fPrecedence < p || n->fPrecedence <= RBBINode::precLParen) {
 677             // The most recent operand goes with the current operator,
 678             //   not with the previously stacked one.
 679             break;
 680         }
 681             // Stack operator is a binary op  ( '|' or concatenation)
 682             //   TOS operand becomes right child of this operator.
 683             //   Resulting subexpression becomes the TOS operand.
 684             n->fRightChild = fNodeStack[fNodeStackPtr];
 685             fNodeStack[fNodeStackPtr]->fParent = n;
 686             fNodeStackPtr--;
 687         // printNodeStack("looping in fixOpStack()   ");
 688     }
 689
 690     if (p <= RBBINode::precLParen) {
 691         // Scan is at a right paren or end of expression.
 692         //  The scanned item must match the stack, or else there was an error.
 693         //  Discard the left paren (or start expr) node from the stack,
 694             //  leaving the completed (sub)expression as TOS.
 695             if (n->fPrecedence != p) {
 696                 // Right paren encountered matched start of expression node, or
 697                 // end of expression matched with a left paren node.
 698                 error(U_BRK_MISMATCHED_PAREN);
 699             }
 700             fNodeStack[fNodeStackPtr-1] = fNodeStack[fNodeStackPtr];
 701             fNodeStackPtr--;
 702             // Delete the now-discarded LParen or Start node.
 703             delete n;
 704     }
 705     // printNodeStack("leaving fixOpStack()");
 706 }
 707
 708
 709
 710
 711 //------------------------------------------------------------------------------
 712 //
 713 //   findSetFor    given a UnicodeString,
 714 //                  - find the corresponding Unicode Set  (uset node)
 715 //                         (create one if necessary)
 716 //                  - Set fLeftChild of the caller's node (should be a setRef node)
 717 //                         to the uset node
 718 //                 Maintain a hash table of uset nodes, so the same one is always used
 719 //                    for the same string.
 720 //                 If a "to adopt" set is provided and we haven't seen this key before,
 721 //                    add the provided set to the hash table.
 722 //                 If the string is one (32 bit) char in length, the set contains
 723 //                    just one element which is the char in question.
 724 //                 If the string is "any", return a set containing all chars.
 725 //
 726 //------------------------------------------------------------------------------
 727 void RBBIRuleScanner57::findSetFor(const UnicodeString &s, RBBINode *node, UnicodeSet *setToAdopt) {
 728
 729     RBBISetTableEl   *el;
 730
 731     // First check whether we've already cached a set for this string.
 732     // If so, just use the cached set in the new node.
 733     //   delete any set provided by the caller, since we own it.
 734     el = (RBBISetTableEl *)uhash_get(fSetTable, &s);
 735     if (el != NULL) {
 736         delete setToAdopt;
 737         node->fLeftChild = el->val;
 738         U_ASSERT(node->fLeftChild->fType == RBBINode::uset);
 739         return;
 740     }
 741
 742     // Haven't seen this set before.
 743     // If the caller didn't provide us with a prebuilt set,
 744     //   create a new UnicodeSet now.
 745     if (setToAdopt == NULL) {
 746         if (s.compare(kAny, -1) == 0) {
 747             setToAdopt = new UnicodeSet(0x000000, 0x10ffff);
 748         } else {
 749             UChar32 c;
 750             c = s.char32At(0);
 751             setToAdopt = new UnicodeSet(c, c);
 752         }
 753     }
 754
 755     //
 756     // Make a new uset node to refer to this UnicodeSet
 757     // This new uset node becomes the child of the caller's setReference node.
 758     //
 759     RBBINode *usetNode    = new RBBINode(RBBINode::uset);
 760     if (usetNode == NULL) {
 761         error(U_MEMORY_ALLOCATION_ERROR);
 762         return;
 763     }
 764     usetNode->fInputSet   = setToAdopt;
 765     usetNode->fParent     = node;
 766     node->fLeftChild      = usetNode;
 767     usetNode->fText = s;
 768
 769
 770     //
 771     // Add the new uset node to the list of all uset nodes.
 772     //
 773     fRB->fUSetNodes->addElement(usetNode, *fRB->fStatus);
 774
 775
 776     //
 777     // Add the new set to the set hash table.
 778     //
 779     el      = (RBBISetTableEl *)uprv_malloc(sizeof(RBBISetTableEl));
 780     UnicodeString *tkey = new UnicodeString(s);
 781     if (tkey == NULL || el == NULL || setToAdopt == NULL) {
 782         // Delete to avoid memory leak
 783         delete tkey;
 784         tkey = NULL;
 785         uprv_free(el);
 786         el = NULL;
 787         delete setToAdopt;
 788         setToAdopt = NULL;
 789
 790         error(U_MEMORY_ALLOCATION_ERROR);
 791         return;
 792     }
 793     el->key = tkey;
 794     el->val = usetNode;
 795     uhash_put(fSetTable, el->key, el, fRB->fStatus);
 796
 797     return;
 798 }
 799
 800
 801
 802 //
 803 //  Assorted Unicode character constants.
 804 //     Numeric because there is no portable way to enter them as literals.
 805 //     (Think EBCDIC).
 806 //
 807 static const UChar      chCR        = 0x0d;      // New lines, for terminating comments.
 808 static const UChar      chLF        = 0x0a;
 809 static const UChar      chNEL       = 0x85;      //    NEL newline variant
 810 static const UChar      chLS        = 0x2028;    //    Unicode Line Separator
 811 static const UChar      chApos      = 0x27;      //  single quote, for quoted chars.
 812 static const UChar      chPound     = 0x23;      // '#', introduces a comment.
 813 static const UChar      chBackSlash = 0x5c;      // '\'  introduces a char escape
 814 static const UChar      chLParen    = 0x28;
 815 static const UChar      chRParen    = 0x29;
 816
 817
 818 //------------------------------------------------------------------------------
 819 //
 820 //  stripRules    Return a rules string without unnecessary
 821 //                characters.
 822 //
 823 //------------------------------------------------------------------------------
 824 UnicodeString RBBIRuleScanner57::stripRules(const UnicodeString &rules) {
 825     UnicodeString strippedRules;
 826     int rulesLength = rules.length();
 827     for (int idx = 0; idx < rulesLength; ) {
 828         UChar ch = rules[idx++];
 829         if (ch == chPound) {
 830             while (idx < rulesLength
 831                 && ch != chCR && ch != chLF && ch != chNEL)
 832             {
 833                 ch = rules[idx++];
 834             }
 835         }
 836         if (!u_isISOControl(ch)) {
 837             strippedRules.append(ch);
 838         }
 839     }
 840     // strippedRules = strippedRules.unescape();
 841     return strippedRules;
 842 }
 843
 844
 845 //------------------------------------------------------------------------------
 846 //
 847 //  nextCharLL    Low Level Next Char from rule input source.
 848 //                Get a char from the input character iterator,
 849 //                keep track of input position for error reporting.
 850 //
 851 //------------------------------------------------------------------------------
 852 UChar32  RBBIRuleScanner57::nextCharLL() {
 853     UChar32  ch;
 854
 855     if (fNextIndex >= fRB->fRules.length()) {
 856         return (UChar32)-1;
 857     }
 858     ch         = fRB->fRules.char32At(fNextIndex);
 859     fNextIndex = fRB->fRules.moveIndex32(fNextIndex, 1);
 860
 861     if (ch == chCR ||
 862         ch == chNEL ||
 863         ch == chLS   ||
 864         (ch == chLF && fLastChar != chCR)) {
 865         // Character is starting a new line.  Bump up the line number, and
 866         //  reset the column to 0.
 867         fLineNum++;
 868         fCharNum=0;
 869         if (fQuoteMode) {
 870             error(U_BRK_NEW_LINE_IN_QUOTED_STRING);
 871             fQuoteMode = FALSE;
 872         }
 873     }
 874     else {
 875         // Character is not starting a new line.  Except in the case of a
 876         //   LF following a CR, increment the column position.
 877         if (ch != chLF) {
 878             fCharNum++;
 879         }
 880     }
 881     fLastChar = ch;
 882     return ch;
 883 }
 884
 885
 886 //------------------------------------------------------------------------------
 887 //
 888 //   nextChar     for rules scanning.  At this level, we handle stripping
 889 //                out comments and processing backslash character escapes.
 890 //                The rest of the rules grammar is handled at the next level up.
 891 //
 892 //------------------------------------------------------------------------------
 893 void RBBIRuleScanner57::nextChar(RBBIRuleChar &c) {
 894
 895     // Unicode Character constants needed for the processing done by nextChar(),
 896     //   in hex because literals wont work on EBCDIC machines.
 897
 898     fScanIndex = fNextIndex;
 899     c.fChar    = nextCharLL();
 900     c.fEscaped = FALSE;
 901
 902     //
 903     //  check for '' sequence.
 904     //  These are recognized in all contexts, whether in quoted text or not.
 905     //
 906     if (c.fChar == chApos) {
 907         if (fRB->fRules.char32At(fNextIndex) == chApos) {
 908             c.fChar    = nextCharLL();        // get nextChar officially so character counts
 909             c.fEscaped = TRUE;                //   stay correct.
 910         }
 911         else
 912         {
 913             // Single quote, by itself.
 914             //   Toggle quoting mode.
 915             //   Return either '('  or ')', because quotes cause a grouping of the quoted text.
 916             fQuoteMode = !fQuoteMode;
 917             if (fQuoteMode == TRUE) {
 918                 c.fChar = chLParen;
 919             } else {
 920                 c.fChar = chRParen;
 921             }
 922             c.fEscaped = FALSE;      // The paren that we return is not escaped.
 923             return;
 924         }
 925     }
 926
 927     if (fQuoteMode) {
 928         c.fEscaped = TRUE;
 929     }
 930     else
 931     {
 932         // We are not in a 'quoted region' of the source.
 933         //
 934         if (c.fChar == chPound) {
 935             // Start of a comment.  Consume the rest of it.
 936             //  The new-line char that terminates the comment is always returned.
 937             //  It will be treated as white-space, and serves to break up anything
 938             //    that might otherwise incorrectly clump together with a comment in
 939             //    the middle (a variable name, for example.)
 940             for (;;) {
 941                 c.fChar = nextCharLL();
 942                 if (c.fChar == (UChar32)-1 ||  // EOF
 943                     c.fChar == chCR     ||
 944                     c.fChar == chLF     ||
 945                     c.fChar == chNEL    ||
 946                     c.fChar == chLS)       {break;}
 947             }
 948         }
 949         if (c.fChar == (UChar32)-1) {
 950             return;
 951         }
 952
 953         //
 954         //  check for backslash escaped characters.
 955         //  Use UnicodeString::unescapeAt() to handle them.
 956         //
 957         if (c.fChar == chBackSlash) {
 958             c.fEscaped = TRUE;
 959             int32_t startX = fNextIndex;
 960             c.fChar = fRB->fRules.unescapeAt(fNextIndex);
 961             if (fNextIndex == startX) {
 962                 error(U_BRK_HEX_DIGITS_EXPECTED);
 963             }
 964             fCharNum += fNextIndex-startX;
 965         }
 966     }
 967     // putc(c.fChar, stdout);
 968 }
 969
 970 //------------------------------------------------------------------------------
 971 //
 972 //  Parse RBBI rules.   The state machine for rules parsing is here.
 973 //                      The state tables are hand-written in the file rbbirpt.txt,
 974 //                      and converted to the form used here by a perl
 975 //                      script rbbicst.pl
 976 //
 977 //------------------------------------------------------------------------------
 978 void RBBIRuleScanner57::parse() {
 979     uint16_t                state;
 980     const RBBIRuleTableEl  *tableEl;
 981
 982     if (U_FAILURE(*fRB->fStatus)) {
 983         return;
 984     }
 985
 986     state = 1;
 987     nextChar(fC);
 988     //
 989     // Main loop for the rule parsing state machine.
 990     //   Runs once per state transition.
 991     //   Each time through optionally performs, depending on the state table,
 992     //      - an advance to the the next input char
 993     //      - an action to be performed.
 994     //      - pushing or popping a state to/from the local state return stack.
 995     //
 996     for (;;) {
 997         //  Bail out if anything has gone wrong.
 998         //  RBBI rule file parsing stops on the first error encountered.
 999         if (U_FAILURE(*fRB->fStatus)) {
1000             break;
1001         }
1002
1003         // Quit if state == 0.  This is the normal way to exit the state machine.
1004         //
1005         if (state == 0) {
1006             break;
1007         }
1008
1009         // Find the state table element that matches the input char from the rule, or the
1010         //    class of the input character.  Start with the first table row for this
1011         //    state, then linearly scan forward until we find a row that matches the
1012         //    character.  The last row for each state always matches all characters, so
1013         //    the search will stop there, if not before.
1014         //
1015         tableEl = &gRuleParseStateTable[state];
1016         #ifdef RBBI_DEBUG
1017             if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "scan")) {
1018                 RBBIDebugPrintf("char, line, col = (\'%c\', %d, %d)    state=%s ",
1019                     fC.fChar, fLineNum, fCharNum, RBBIRuleStateNames[state]);
1020             }
1021         #endif
1022
1023         for (;;) {
1024             #ifdef RBBI_DEBUG
1025                 if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "scan")) { RBBIDebugPrintf("."); fflush(stdout);}
1026             #endif
1027             if (tableEl->fCharClass < 127 && fC.fEscaped == FALSE &&   tableEl->fCharClass == fC.fChar) {
1028                 // Table row specified an individual character, not a set, and
1029                 //   the input character is not escaped, and
1030                 //   the input character matched it.
1031                 break;
1032             }
1033             if (tableEl->fCharClass == 255) {
1034                 // Table row specified default, match anything character class.
1035                 break;
1036             }
1037             if (tableEl->fCharClass == 254 && fC.fEscaped)  {
1038                 // Table row specified "escaped" and the char was escaped.
1039                 break;
1040             }
1041             if (tableEl->fCharClass == 253 && fC.fEscaped &&
1042                 (fC.fChar == 0x50 || fC.fChar == 0x70 ))  {
1043                 // Table row specified "escaped P" and the char is either 'p' or 'P'.
1044                 break;
1045             }
1046             if (tableEl->fCharClass == 252 && fC.fChar == (UChar32)-1)  {
1047                 // Table row specified eof and we hit eof on the input.
1048                 break;
1049             }
1050
1051             if (tableEl->fCharClass >= 128 && tableEl->fCharClass < 240 &&   // Table specs a char class &&
1052                 fC.fEscaped == FALSE &&                                      //   char is not escaped &&
1053                 fC.fChar != (UChar32)-1) {                                   //   char is not EOF
1054                 U_ASSERT((tableEl->fCharClass-128) < UPRV_LENGTHOF(fRuleSets));
1055                 if (fRuleSets[tableEl->fCharClass-128].contains(fC.fChar)) {
1056                     // Table row specified a character class, or set of characters,
1057                     //   and the current char matches it.
1058                     break;
1059                 }
1060             }
1061
1062             // No match on this row, advance to the next  row for this state,
1063             tableEl++;
1064         }
1065         if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "scan")) { RBBIDebugPuts("");}
1066
1067         //
1068         // We've found the row of the state table that matches the current input
1069         //   character from the rules string.
1070         // Perform any action specified  by this row in the state table.
1071         if (doParseActions((int32_t)tableEl->fAction) == FALSE) {
1072             // Break out of the state machine loop if the
1073             //   the action signalled some kind of error, or
1074             //   the action was to exit, occurs on normal end-of-rules-input.
1075             break;
1076         }
1077
1078         if (tableEl->fPushState != 0) {
1079             fStackPtr++;
1080             if (fStackPtr >= kStackSize) {
1081                 error(U_BRK_INTERNAL_ERROR);
1082                 RBBIDebugPuts("RBBIRuleScanner57::parse() - state stack overflow.");
1083                 fStackPtr--;
1084             }
1085             fStack[fStackPtr] = tableEl->fPushState;
1086         }
1087
1088         if (tableEl->fNextChar) {
1089             nextChar(fC);
1090         }
1091
1092         // Get the next state from the table entry, or from the
1093         //   state stack if the next state was specified as "pop".
1094         if (tableEl->fNextState != 255) {
1095             state = tableEl->fNextState;
1096         } else {
1097             state = fStack[fStackPtr];
1098             fStackPtr--;
1099             if (fStackPtr < 0) {
1100                 error(U_BRK_INTERNAL_ERROR);
1101                 RBBIDebugPuts("RBBIRuleScanner57::parse() - state stack underflow.");
1102                 fStackPtr++;
1103             }
1104         }
1105
1106     }
1107
1108     //
1109     // If there were NO user specified reverse rules, set up the equivalent of ".*;"
1110     //
1111     if (fRB->fReverseTree == NULL) {
1112         fRB->fReverseTree  = pushNewNode(RBBINode::opStar);
1113         RBBINode  *operand = pushNewNode(RBBINode::setRef);
1114         if (U_FAILURE(*fRB->fStatus)) {
1115             return;
1116         }
1117         findSetFor(UnicodeString(TRUE, kAny, 3), operand);
1118         fRB->fReverseTree->fLeftChild = operand;
1119         operand->fParent              = fRB->fReverseTree;
1120         fNodeStackPtr -= 2;
1121     }
1122
1123
1124     //
1125     // Parsing of the input RBBI rules is complete.
1126     // We now have a parse tree for the rule expressions
1127     // and a list of all UnicodeSets that are referenced.
1128     //
1129 #ifdef RBBI_DEBUG
1130     if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "symbols")) {fSymbolTable->rbbiSymtablePrint();}
1131     if (fRB->fDebugEnv && uprv_strstr(fRB->fDebugEnv, "ptree"))
1132     {
1133         RBBIDebugPrintf("Completed Forward Rules Parse Tree...\n");
1134         fRB->fForwardTree->printTree(TRUE);
1135         RBBIDebugPrintf("\nCompleted Reverse Rules Parse Tree...\n");
1136         fRB->fReverseTree->printTree(TRUE);
1137         RBBIDebugPrintf("\nCompleted Safe Point Forward Rules Parse Tree...\n");
1138         fRB->fSafeFwdTree->printTree(TRUE);
1139         RBBIDebugPrintf("\nCompleted Safe Point Reverse Rules Parse Tree...\n");
1140         fRB->fSafeRevTree->printTree(TRUE);
1141     }
1142 #endif
1143 }
1144
1145
1146 //------------------------------------------------------------------------------
1147 //
1148 //  printNodeStack     for debugging...
1149 //
1150 //------------------------------------------------------------------------------
1151 #ifdef RBBI_DEBUG
1152 void RBBIRuleScanner57::printNodeStack(const char *title) {
1153     int i;
1154     RBBIDebugPrintf("%s.  Dumping node stack...\n", title);
1155     for (i=fNodeStackPtr; i>0; i--) {fNodeStack[i]->printTree(TRUE);}
1156 }
1157 #endif
1158
1159
1160
1161
1162 //------------------------------------------------------------------------------
1163 //
1164 //  pushNewNode   create a new RBBINode of the specified type and push it
1165 //                onto the stack of nodes.
1166 //
1167 //------------------------------------------------------------------------------
1168 RBBINode  *RBBIRuleScanner57::pushNewNode(RBBINode::NodeType  t) {
1169     if (U_FAILURE(*fRB->fStatus)) {
1170         return NULL;
1171     }
1172     fNodeStackPtr++;
1173     if (fNodeStackPtr >= kStackSize) {
1174         error(U_BRK_INTERNAL_ERROR);
1175         RBBIDebugPuts("RBBIRuleScanner57::pushNewNode - stack overflow.");
1176         *fRB->fStatus = U_BRK_INTERNAL_ERROR;
1177         return NULL;
1178     }
1179     fNodeStack[fNodeStackPtr] = new RBBINode(t);
1180     if (fNodeStack[fNodeStackPtr] == NULL) {
1181         *fRB->fStatus = U_MEMORY_ALLOCATION_ERROR;
1182     }
1183     return fNodeStack[fNodeStackPtr];
1184 }
1185
1186
1187
1188 //------------------------------------------------------------------------------
1189 //
1190 //  scanSet    Construct a UnicodeSet from the text at the current scan
1191 //             position.  Advance the scan position to the first character
1192 //             after the set.
1193 //
1194 //             A new RBBI setref node referring to the set is pushed onto the node
1195 //             stack.
1196 //
1197 //             The scan position is normally under the control of the state machine
1198 //             that controls rule parsing.  UnicodeSets, however, are parsed by
1199 //             the UnicodeSet constructor, not by the RBBI rule parser.
1200 //
1201 //------------------------------------------------------------------------------
1202 void RBBIRuleScanner57::scanSet() {
1203     UnicodeSet    *uset;
1204     ParsePosition  pos;
1205     int            startPos;
1206     int            i;
1207
1208     if (U_FAILURE(*fRB->fStatus)) {
1209         return;
1210     }
1211
1212     pos.setIndex(fScanIndex);
1213     startPos = fScanIndex;
1214     UErrorCode localStatus = U_ZERO_ERROR;
1215     uset = new UnicodeSet();
1216     if (uset == NULL) {
1217         localStatus = U_MEMORY_ALLOCATION_ERROR;
1218     } else {
1219         uset->applyPatternIgnoreSpace(fRB->fRules, pos, fSymbolTable, localStatus);
1220     }
1221     if (U_FAILURE(localStatus)) {
1222         //  TODO:  Get more accurate position of the error from UnicodeSet's return info.
1223         //         UnicodeSet appears to not be reporting correctly at this time.
1224         #ifdef RBBI_DEBUG
1225             RBBIDebugPrintf("UnicodeSet parse postion.ErrorIndex = %d\n", pos.getIndex());
1226         #endif
1227         error(localStatus);
1228         delete uset;
1229         return;
1230     }
1231
1232     // Verify that the set contains at least one code point.
1233     //
1234     U_ASSERT(uset!=NULL);
1235     if (uset->isEmpty()) {
1236         // This set is empty.
1237         //  Make it an error, because it almost certainly is not what the user wanted.
1238         //  Also, avoids having to think about corner cases in the tree manipulation code
1239         //   that occurs later on.
1240         error(U_BRK_RULE_EMPTY_SET);
1241         delete uset;
1242         return;
1243     }
1244
1245
1246     // Advance the RBBI parse postion over the UnicodeSet pattern.
1247     //   Don't just set fScanIndex because the line/char positions maintained
1248     //   for error reporting would be thrown off.
1249     i = pos.getIndex();
1250     for (;;) {
1251         if (fNextIndex >= i) {
1252             break;
1253         }
1254         nextCharLL();
1255     }
1256
1257     if (U_SUCCESS(*fRB->fStatus)) {
1258         RBBINode         *n;
1259
1260         n = pushNewNode(RBBINode::setRef);
1261         if (U_FAILURE(*fRB->fStatus)) {
1262             return;
1263         }
1264         n->fFirstPos = startPos;
1265         n->fLastPos  = fNextIndex;
1266         fRB->fRules.extractBetween(n->fFirstPos, n->fLastPos, n->fText);
1267         //  findSetFor() serves several purposes here:
1268         //     - Adopts storage for the UnicodeSet, will be responsible for deleting.
1269         //     - Mantains collection of all sets in use, needed later for establishing
1270         //          character categories for run time engine.
1271         //     - Eliminates mulitiple instances of the same set.
1272         //     - Creates a new uset node if necessary (if this isn't a duplicate.)
1273         findSetFor(n->fText, n, uset);
1274     }
1275
1276 }
1277
1278 U_NAMESPACE_END
1279
1280 #endif /* #if !UCONFIG_NO_BREAK_ITERATION */