/* ****************************************************************************** * * Copyright (C) 1999-2004, International Business Machines * Corporation and others. All Rights Reserved. * ****************************************************************************** * file name: ubidi.c * encoding: US-ASCII * tab size: 8 (not used) * indentation:4 * * created on: 1999jul27 * created by: Markus W. Scherer */ /* set import/export definitions */ #ifndef U_COMMON_IMPLEMENTATION # define U_COMMON_IMPLEMENTATION #endif #include "cmemory.h" #include "unicode/utypes.h" #include "unicode/ustring.h" #include "unicode/uchar.h" #include "unicode/ubidi.h" #include "ubidiimp.h" /* * General implementation notes: * * Throughout the implementation, there are comments like (W2) that refer to * rules of the BiDi algorithm in its version 5, in this example to the second * rule of the resolution of weak types. * * For handling surrogate pairs, where two UChar's form one "abstract" (or UTF-32) * character according to UTF-16, the second UChar gets the directional property of * the entire character assigned, while the first one gets a BN, a boundary * neutral, type, which is ignored by most of the algorithm according to * rule (X9) and the implementation suggestions of the BiDi algorithm. * * Later, adjustWSLevels() will set the level for each BN to that of the * following character (UChar), which results in surrogate pairs getting the * same level on each of their surrogates. * * In a UTF-8 implementation, the same thing could be done: the last byte of * a multi-byte sequence would get the "real" property, while all previous * bytes of that sequence would get BN. * * It is not possible to assign all those parts of a character the same real * property because this would fail in the resolution of weak types with rules * that look at immediately surrounding types. * * As a related topic, this implementation does not remove Boundary Neutral * types from the input, but ignores them whereever this is relevant. * For example, the loop for the resolution of the weak types reads * types until it finds a non-BN. * Also, explicit embedding codes are neither changed into BN nor removed. * They are only treated the same way real BNs are. * As stated before, adjustWSLevels() takes care of them at the end. * For the purpose of conformance, the levels of all these codes * do not matter. * * Note that this implementation never modifies the dirProps * after the initial setup. * * * In this implementation, the resolution of weak types (Wn), * neutrals (Nn), and the assignment of the resolved level (In) * are all done in one single loop, in resolveImplicitLevels(). * Changes of dirProp values are done on the fly, without writing * them back to the dirProps array. * * * This implementation contains code that allows to bypass steps of the * algorithm that are not needed on the specific paragraph * in order to speed up the most common cases considerably, * like text that is entirely LTR, or RTL text without numbers. * * Most of this is done by setting a bit for each directional property * in a flags variable and later checking for whether there are * any LTR characters or any RTL characters, or both, whether * there are any explicit embedding codes, etc. * * If the (Xn) steps are performed, then the flags are re-evaluated, * because they will then not contain the embedding codes any more * and will be adjusted for override codes, so that subsequently * more bypassing may be possible than what the initial flags suggested. * * If the text is not mixed-directional, then the * algorithm steps for the weak type resolution are not performed, * and all levels are set to the paragraph level. * * If there are no explicit embedding codes, then the (Xn) steps * are not performed. * * If embedding levels are supplied as a parameter, then all * explicit embedding codes are ignored, and the (Xn) steps * are not performed. * * White Space types could get the level of the run they belong to, * and are checked with a test of (flags&MASK_EMBEDDING) to * consider if the paragraph direction should be considered in * the flags variable. * * If there are no White Space types in the paragraph, then * (L1) is not necessary in adjustWSLevels(). */ /* to avoid some conditional statements, use tiny constant arrays */ static const Flags flagLR[2]={ DIRPROP_FLAG(L), DIRPROP_FLAG(R) }; static const Flags flagE[2]={ DIRPROP_FLAG(LRE), DIRPROP_FLAG(RLE) }; static const Flags flagO[2]={ DIRPROP_FLAG(LRO), DIRPROP_FLAG(RLO) }; #define DIRPROP_FLAG_LR(level) flagLR[(level)&1] #define DIRPROP_FLAG_E(level) flagE[(level)&1] #define DIRPROP_FLAG_O(level) flagO[(level)&1] /* UBiDi object management -------------------------------------------------- */ U_CAPI UBiDi * U_EXPORT2 ubidi_open(void) { UErrorCode errorCode=U_ZERO_ERROR; return ubidi_openSized(0, 0, &errorCode); } U_CAPI UBiDi * U_EXPORT2 ubidi_openSized(int32_t maxLength, int32_t maxRunCount, UErrorCode *pErrorCode) { UBiDi *pBiDi; /* check the argument values */ if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) { return NULL; } else if(maxLength<0 || maxRunCount<0) { *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR; return NULL; /* invalid arguments */ } /* allocate memory for the object */ pBiDi=(UBiDi *)uprv_malloc(sizeof(UBiDi)); if(pBiDi==NULL) { *pErrorCode=U_MEMORY_ALLOCATION_ERROR; return NULL; } /* reset the object, all pointers NULL, all flags FALSE, all sizes 0 */ uprv_memset(pBiDi, 0, sizeof(UBiDi)); /* allocate memory for arrays as requested */ if(maxLength>0) { if( !getInitialDirPropsMemory(pBiDi, maxLength) || !getInitialLevelsMemory(pBiDi, maxLength) ) { *pErrorCode=U_MEMORY_ALLOCATION_ERROR; } } else { pBiDi->mayAllocateText=TRUE; } if(maxRunCount>0) { if(maxRunCount==1) { /* use simpleRuns[] */ pBiDi->runsSize=sizeof(Run); } else if(!getInitialRunsMemory(pBiDi, maxRunCount)) { *pErrorCode=U_MEMORY_ALLOCATION_ERROR; } } else { pBiDi->mayAllocateRuns=TRUE; } if(U_SUCCESS(*pErrorCode)) { return pBiDi; } else { ubidi_close(pBiDi); return NULL; } } /* * We are allowed to allocate memory if memory==NULL or * mayAllocate==TRUE for each array that we need. * We also try to grow and shrink memory as needed if we * allocate it. * * Assume sizeNeeded>0. * If *pMemory!=NULL, then assume *pSize>0. * * ### this realloc() may unnecessarily copy the old data, * which we know we don't need any more; * is this the best way to do this?? */ U_CFUNC UBool ubidi_getMemory(void **pMemory, int32_t *pSize, UBool mayAllocate, int32_t sizeNeeded) { /* check for existing memory */ if(*pMemory==NULL) { /* we need to allocate memory */ if(mayAllocate && (*pMemory=uprv_malloc(sizeNeeded))!=NULL) { *pSize=sizeNeeded; return TRUE; } else { return FALSE; } } else { /* there is some memory, is it enough or too much? */ if(sizeNeeded>*pSize && !mayAllocate) { /* not enough memory, and we must not allocate */ return FALSE; } else if(sizeNeeded!=*pSize && mayAllocate) { /* we may try to grow or shrink */ void *memory; if((memory=uprv_realloc(*pMemory, sizeNeeded))!=NULL) { *pMemory=memory; *pSize=sizeNeeded; return TRUE; } else { /* we failed to grow */ return FALSE; } } else { /* we have at least enough memory and must not allocate */ return TRUE; } } } U_CAPI void U_EXPORT2 ubidi_close(UBiDi *pBiDi) { if(pBiDi!=NULL) { if(pBiDi->dirPropsMemory!=NULL) { uprv_free(pBiDi->dirPropsMemory); } if(pBiDi->levelsMemory!=NULL) { uprv_free(pBiDi->levelsMemory); } if(pBiDi->runsMemory!=NULL) { uprv_free(pBiDi->runsMemory); } uprv_free(pBiDi); } } /* set to approximate "inverse BiDi" ---------------------------------------- */ U_CAPI void U_EXPORT2 ubidi_setInverse(UBiDi *pBiDi, UBool isInverse) { if(pBiDi!=NULL) { pBiDi->isInverse=isInverse; } } U_CAPI UBool U_EXPORT2 ubidi_isInverse(UBiDi *pBiDi) { if(pBiDi!=NULL) { return pBiDi->isInverse; } else { return FALSE; } } /* perform (P2)..(P3) ------------------------------------------------------- */ /* * Get the directional properties for the text, * calculate the flags bit-set, and * determine the partagraph level if necessary. */ static void getDirProps(UBiDi *pBiDi, const UChar *text) { DirProp *dirProps=pBiDi->dirPropsMemory; /* pBiDi->dirProps is const */ int32_t i=0, i0, i1, length=pBiDi->length; Flags flags=0; /* collect all directionalities in the text */ UChar32 uchar; DirProp dirProp; if(IS_DEFAULT_LEVEL(pBiDi->paraLevel)) { /* determine the paragraph level (P2..P3) */ for(;;) { i0=i; /* index of first code unit */ UTF_NEXT_CHAR(text, i, length, uchar); i1=i-1; /* index of last code unit, gets the directional property */ flags|=DIRPROP_FLAG(dirProps[i1]=dirProp=u_charDirection(uchar)); if(i1>i0) { /* set previous code units' properties to BN */ flags|=DIRPROP_FLAG(BN); do { dirProps[--i1]=BN; } while(i1>i0); } if(dirProp==L) { pBiDi->paraLevel=0; break; } else if(dirProp==R || dirProp==AL) { pBiDi->paraLevel=1; break; } else if(i>=length) { /* * see comment in ubidi.h: * the DEFAULT_XXX values are designed so that * their bit 0 alone yields the intended default */ pBiDi->paraLevel&=1; break; } } } else { flags|=DIRPROP_FLAG_LR(pBiDi->paraLevel); } /* get the rest of the directional properties and the flags bits */ while(ii0) { /* set previous code units' properties to BN */ flags|=DIRPROP_FLAG(BN); do { dirProps[--i1]=BN; } while(i1>i0); } } if(flags&MASK_EMBEDDING) { flags|=DIRPROP_FLAG_LR(pBiDi->paraLevel); } pBiDi->flags=flags; } /* perform (X1)..(X9) ------------------------------------------------------- */ /* determine if the text is mixed-directional or single-directional */ static UBiDiDirection directionFromFlags(Flags flags) { /* if the text contains AN and neutrals, then some neutrals may become RTL */ if(!(flags&MASK_RTL || ((flags&DIRPROP_FLAG(AN)) && (flags&MASK_POSSIBLE_N)))) { return UBIDI_LTR; } else if(!(flags&MASK_LTR)) { return UBIDI_RTL; } else { return UBIDI_MIXED; } } /* * Resolve the explicit levels as specified by explicit embedding codes. * Recalculate the flags to have them reflect the real properties * after taking the explicit embeddings into account. * * The BiDi algorithm is designed to result in the same behavior whether embedding * levels are externally specified (from "styled text", supposedly the preferred * method) or set by explicit embedding codes (LRx, RLx, PDF) in the plain text. * That is why (X9) instructs to remove all explicit codes (and BN). * However, in a real implementation, this removal of these codes and their index * positions in the plain text is undesirable since it would result in * reallocated, reindexed text. * Instead, this implementation leaves the codes in there and just ignores them * in the subsequent processing. * In order to get the same reordering behavior, positions with a BN or an * explicit embedding code just get the same level assigned as the last "real" * character. * * Some implementations, not this one, then overwrite some of these * directionality properties at "real" same-level-run boundaries by * L or R codes so that the resolution of weak types can be performed on the * entire paragraph at once instead of having to parse it once more and * perform that resolution on same-level-runs. * This limits the scope of the implicit rules in effectively * the same way as the run limits. * * Instead, this implementation does not modify these codes. * On one hand, the paragraph has to be scanned for same-level-runs, but * on the other hand, this saves another loop to reset these codes, * or saves making and modifying a copy of dirProps[]. * * * Note that (Pn) and (Xn) changed significantly from version 4 of the BiDi algorithm. * * * Handling the stack of explicit levels (Xn): * * With the BiDi stack of explicit levels, * as pushed with each LRE, RLE, LRO, and RLO and popped with each PDF, * the explicit level must never exceed UBIDI_MAX_EXPLICIT_LEVEL==61. * * In order to have a correct push-pop semantics even in the case of overflows, * there are two overflow counters: * - countOver60 is incremented with each LRx at level 60 * - from level 60, one RLx increases the level to 61 * - countOver61 is incremented with each LRx and RLx at level 61 * * Popping levels with PDF must work in the opposite order so that level 61 * is correct at the correct point. Underflows (too many PDFs) must be checked. * * This implementation assumes that UBIDI_MAX_EXPLICIT_LEVEL is odd. */ static UBiDiDirection resolveExplicitLevels(UBiDi *pBiDi) { const DirProp *dirProps=pBiDi->dirProps; UBiDiLevel *levels=pBiDi->levels; int32_t i=0, length=pBiDi->length; Flags flags=pBiDi->flags; /* collect all directionalities in the text */ DirProp dirProp; UBiDiLevel level=pBiDi->paraLevel; UBiDiDirection direction; /* determine if the text is mixed-directional or single-directional */ direction=directionFromFlags(flags); /* we may not need to resolve any explicit levels */ if(direction!=UBIDI_MIXED) { /* not mixed directionality: levels don't matter - trailingWSStart will be 0 */ } else if(!(flags&MASK_EXPLICIT) || pBiDi->isInverse) { /* mixed, but all characters are at the same embedding level */ /* or we are in "inverse BiDi" */ /* set all levels to the paragraph level */ for(i=0; i=UBIDI_MAX_EXPLICIT_LEVEL */ uint32_t countOver60=0, countOver61=0; /* count overflows of explicit levels */ /* recalculate the flags */ flags=0; /* since we assume that this is a single paragraph, we ignore (X8) */ for(i=0; i0) { --countOver61; } else if(countOver60>0 && (embeddingLevel&~UBIDI_LEVEL_OVERRIDE)!=UBIDI_MAX_EXPLICIT_LEVEL) { /* handle LRx overflows from level 60 */ --countOver60; } else if(stackTop>0) { /* this is the pop operation; it also pops level 61 while countOver60>0 */ --stackTop; embeddingLevel=stack[stackTop]; /* } else { (underflow) */ } flags|=DIRPROP_FLAG(BN); break; case B: /* * We do not really expect to see a paragraph separator (B), * but we should do something reasonable with it, * especially at the end of the text. */ stackTop=0; countOver60=countOver61=0; embeddingLevel=level=pBiDi->paraLevel; flags|=DIRPROP_FLAG(B); break; case BN: /* BN, LRE, RLE, and PDF are supposed to be removed (X9) */ /* they will get their levels set correctly in adjustWSLevels() */ flags|=DIRPROP_FLAG(BN); break; default: /* all other types get the "real" level */ if(level!=embeddingLevel) { level=embeddingLevel; if(level&UBIDI_LEVEL_OVERRIDE) { flags|=DIRPROP_FLAG_O(level)|DIRPROP_FLAG_MULTI_RUNS; } else { flags|=DIRPROP_FLAG_E(level)|DIRPROP_FLAG_MULTI_RUNS; } } if(!(level&UBIDI_LEVEL_OVERRIDE)) { flags|=DIRPROP_FLAG(dirProp); } break; } /* * We need to set reasonable levels even on BN codes and * explicit codes because we will later look at same-level runs (X10). */ levels[i]=level; } if(flags&MASK_EMBEDDING) { flags|=DIRPROP_FLAG_LR(pBiDi->paraLevel); } /* subsequently, ignore the explicit codes and BN (X9) */ /* again, determine if the text is mixed-directional or single-directional */ pBiDi->flags=flags; direction=directionFromFlags(flags); } return direction; } /* * Use a pre-specified embedding levels array: * * Adjust the directional properties for overrides (->LEVEL_OVERRIDE), * ignore all explicit codes (X9), * and check all the preset levels. * * Recalculate the flags to have them reflect the real properties * after taking the explicit embeddings into account. */ static UBiDiDirection checkExplicitLevels(UBiDi *pBiDi, UErrorCode *pErrorCode) { const DirProp *dirProps=pBiDi->dirProps; UBiDiLevel *levels=pBiDi->levels; int32_t i, length=pBiDi->length; Flags flags=0; /* collect all directionalities in the text */ UBiDiLevel level, paraLevel=pBiDi->paraLevel; for(i=0; iparaLevel); } /* determine if the text is mixed-directional or single-directional */ pBiDi->flags=flags; return directionFromFlags(flags); } /* perform rules (Wn), (Nn), and (In) on a run of the text ------------------ */ /* * This implementation of the (Wn) rules applies all rules in one pass. * In order to do so, it needs a look-ahead of typically 1 character * (except for W5: sequences of ET) and keeps track of changes * in a rule Wp that affect a later Wq (pdirProps; UBiDiLevel *levels=pBiDi->levels; int32_t i, next, neutralStart=-1; DirProp prevDirProp, dirProp, nextDirProp, lastStrong, beforeNeutral=L; UBiDiLevel numberLevel; uint8_t historyOfEN; /* initialize: current at sor, next at start (it is startisInverse) { /* * For "inverse BiDi", we set the levels of numbers just like for * regular L characters, plus a flag that ubidi_getRuns() will use * to set a similar flag on the corresponding output run. */ numberLevel=levels[start]; if(numberLevel&1) { ++numberLevel; } } else { /* normal BiDi: least greater even level */ numberLevel=(UBiDiLevel)((levels[start]+2)&~1); } /* * In all steps of this implementation, BN and explicit embedding codes * must be treated as if they didn't exist (X9). * They will get levels set before a non-neutral character, and remain * undefined before a neutral one, but adjustWSLevels() will take care * of all of them. */ while(DIRPROP_FLAG(nextDirProp)&MASK_BN_EXPLICIT) { if(++next>=EN_SHIFT; /* * Technically, this should be done before the switch() in the form * if(nextDirProp==NSM) { * dirProps[next]=nextDirProp=dirProp; * } * * - effectively one iteration ahead. * However, whether the next dirProp is NSM or is equal to the current dirProp * does not change the outcome of any condition in (W2)..(W7). */ break; default: break; } /* here, it is always [prev,this,next]dirProp!=BN; it may be next>i+1 */ /* perform (Nn) - here, only L, R, EN, AN, and neutrals are left */ /* for "inverse BiDi", treat neutrals like L */ /* this is one iteration late for the neutrals */ if(DIRPROP_FLAG(dirProp)&MASK_N) { if(neutralStart<0) { /* start of a sequence of neutrals */ neutralStart=i; beforeNeutral=prevDirProp; } } else /* not a neutral, can be only one of { L, R, EN, AN } */ { /* * Note that all levels[] values are still the same at this * point because this function is called for an entire * same-level run. * Therefore, we need to read only one actual level. */ UBiDiLevel level=levels[i]; if(neutralStart>=0) { UBiDiLevel final; /* end of a sequence of neutrals (dirProp is "afterNeutral") */ if(!(pBiDi->isInverse)) { if(beforeNeutral==L) { if(dirProp==L) { final=0; /* make all neutrals L (N1) */ } else { final=level; /* make all neutrals "e" (N2) */ } } else /* beforeNeutral is one of { R, EN, AN } */ { if(dirProp==L) { final=level; /* make all neutrals "e" (N2) */ } else { final=1; /* make all neutrals R (N1) */ } } } else { /* "inverse BiDi": collapse [before]dirProps L, EN, AN into L */ if(beforeNeutral!=R) { if(dirProp!=R) { final=0; /* make all neutrals L (N1) */ } else { final=level; /* make all neutrals "e" (N2) */ } } else /* beforeNeutral is one of { R, EN, AN } */ { if(dirProp!=R) { final=level; /* make all neutrals "e" (N2) */ } else { final=1; /* make all neutrals R (N1) */ } } } /* perform (In) on the sequence of neutrals */ if((level^final)&1) { /* do something only if we need to _change_ the level */ do { ++levels[neutralStart]; } while(++neutralStart=0) { /* * Note that all levels[] values are still the same at this * point because this function is called for an entire * same-level run. * Therefore, we need to read only one actual level. */ UBiDiLevel level=levels[neutralStart], final; /* end of a sequence of neutrals (eor is "afterNeutral") */ if(!(pBiDi->isInverse)) { if(beforeNeutral==L) { if(eor==L) { final=0; /* make all neutrals L (N1) */ } else { final=level; /* make all neutrals "e" (N2) */ } } else /* beforeNeutral is one of { R, EN, AN } */ { if(eor==L) { final=level; /* make all neutrals "e" (N2) */ } else { final=1; /* make all neutrals R (N1) */ } } } else { /* "inverse BiDi": collapse [before]dirProps L, EN, AN into L */ if(beforeNeutral!=R) { if(eor!=R) { final=0; /* make all neutrals L (N1) */ } else { final=level; /* make all neutrals "e" (N2) */ } } else /* beforeNeutral is one of { R, EN, AN } */ { if(eor!=R) { final=level; /* make all neutrals "e" (N2) */ } else { final=1; /* make all neutrals R (N1) */ } } } /* perform (In) on the sequence of neutrals */ if((level^final)&1) { /* do something only if we need to _change_ the level */ do { ++levels[neutralStart]; } while(++neutralStartdirProps; UBiDiLevel *levels=pBiDi->levels; int32_t i; if(pBiDi->flags&MASK_WS) { UBiDiLevel paraLevel=pBiDi->paraLevel; Flags flag; i=pBiDi->trailingWSStart; while(i>0) { /* reset a sequence of WS/BN before eop and B/S to the paragraph paraLevel */ while(i>0 && DIRPROP_FLAG(dirProps[--i])&MASK_WS) { levels[i]=paraLevel; } /* reset BN to the next character's paraLevel until B/S, which restarts above loop */ /* here, i+1 is guaranteed to be 0) { flag=DIRPROP_FLAG(dirProps[--i]); if(flag&MASK_BN_EXPLICIT) { levels[i]=levels[i+1]; } else if(flag&MASK_B_S) { levels[i]=paraLevel; break; } } } } } /* ubidi_setPara ------------------------------------------------------------ */ U_CAPI void U_EXPORT2 ubidi_setPara(UBiDi *pBiDi, const UChar *text, int32_t length, UBiDiLevel paraLevel, UBiDiLevel *embeddingLevels, UErrorCode *pErrorCode) { UBiDiDirection direction; /* check the argument values */ if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) { return; } else if(pBiDi==NULL || text==NULL || ((UBIDI_MAX_EXPLICIT_LEVELtext=text; pBiDi->length=length; pBiDi->paraLevel=paraLevel; pBiDi->direction=UBIDI_LTR; pBiDi->trailingWSStart=length; /* the levels[] will reflect the WS run */ pBiDi->dirProps=NULL; pBiDi->levels=NULL; pBiDi->runs=NULL; if(length==0) { /* * For an empty paragraph, create a UBiDi object with the paraLevel and * the flags and the direction set but without allocating zero-length arrays. * There is nothing more to do. */ if(IS_DEFAULT_LEVEL(paraLevel)) { pBiDi->paraLevel&=1; } if(paraLevel&1) { pBiDi->flags=DIRPROP_FLAG(R); pBiDi->direction=UBIDI_RTL; } else { pBiDi->flags=DIRPROP_FLAG(L); pBiDi->direction=UBIDI_LTR; } pBiDi->runCount=0; return; } pBiDi->runCount=-1; /* * Get the directional properties, * the flags bit-set, and * determine the partagraph level if necessary. */ if(getDirPropsMemory(pBiDi, length)) { pBiDi->dirProps=pBiDi->dirPropsMemory; getDirProps(pBiDi, text); } else { *pErrorCode=U_MEMORY_ALLOCATION_ERROR; return; } /* are explicit levels specified? */ if(embeddingLevels==NULL) { /* no: determine explicit levels according to the (Xn) rules */\ if(getLevelsMemory(pBiDi, length)) { pBiDi->levels=pBiDi->levelsMemory; direction=resolveExplicitLevels(pBiDi); } else { *pErrorCode=U_MEMORY_ALLOCATION_ERROR; return; } } else { /* set BN for all explicit codes, check that all levels are paraLevel..UBIDI_MAX_EXPLICIT_LEVEL */ pBiDi->levels=embeddingLevels; direction=checkExplicitLevels(pBiDi, pErrorCode); if(U_FAILURE(*pErrorCode)) { return; } } /* * The steps after (X9) in the UBiDi algorithm are performed only if * the paragraph text has mixed directionality! */ pBiDi->direction=direction; switch(direction) { case UBIDI_LTR: /* make sure paraLevel is even */ pBiDi->paraLevel=(UBiDiLevel)((pBiDi->paraLevel+1)&~1); /* all levels are implicitly at paraLevel (important for ubidi_getLevels()) */ pBiDi->trailingWSStart=0; break; case UBIDI_RTL: /* make sure paraLevel is odd */ pBiDi->paraLevel|=1; /* all levels are implicitly at paraLevel (important for ubidi_getLevels()) */ pBiDi->trailingWSStart=0; break; default: /* * If there are no external levels specified and there * are no significant explicit level codes in the text, * then we can treat the entire paragraph as one run. * Otherwise, we need to perform the following rules on runs of * the text with the same embedding levels. (X10) * "Significant" explicit level codes are ones that actually * affect non-BN characters. * Examples for "insignificant" ones are empty embeddings * LRE-PDF, LRE-RLE-PDF-PDF, etc. */ if(embeddingLevels==NULL && !(pBiDi->flags&DIRPROP_FLAG_MULTI_RUNS)) { resolveImplicitLevels(pBiDi, 0, length, GET_LR_FROM_LEVEL(pBiDi->paraLevel), GET_LR_FROM_LEVEL(pBiDi->paraLevel)); } else { /* sor, eor: start and end types of same-level-run */ UBiDiLevel *levels=pBiDi->levels; int32_t start, limit=0; UBiDiLevel level, nextLevel; DirProp sor, eor; /* determine the first sor and set eor to it because of the loop body (sor=eor there) */ level=pBiDi->paraLevel; nextLevel=levels[0]; if(levelparaLevel; } /* determine eor from max(level, nextLevel); sor is last run's eor */ if((level&~UBIDI_LEVEL_OVERRIDE)<(nextLevel&~UBIDI_LEVEL_OVERRIDE)) { eor=GET_LR_FROM_LEVEL(nextLevel); } else { eor=GET_LR_FROM_LEVEL(level); } /* if the run consists of overridden directional types, then there are no implicit types to be resolved */ if(!(level&UBIDI_LEVEL_OVERRIDE)) { resolveImplicitLevels(pBiDi, start, limit, sor, eor); } else { /* remove the UBIDI_LEVEL_OVERRIDE flags */ do { levels[start++]&=~UBIDI_LEVEL_OVERRIDE; } while(startisInverse) { if(!ubidi_getRuns(pBiDi)) { *pErrorCode=U_MEMORY_ALLOCATION_ERROR; return; } } break; } } U_CAPI UBiDiDirection U_EXPORT2 ubidi_getDirection(const UBiDi *pBiDi) { if(pBiDi!=NULL) { return pBiDi->direction; } else { return UBIDI_LTR; } } U_CAPI const UChar * U_EXPORT2 ubidi_getText(const UBiDi *pBiDi) { if(pBiDi!=NULL) { return pBiDi->text; } else { return NULL; } } U_CAPI int32_t U_EXPORT2 ubidi_getLength(const UBiDi *pBiDi) { if(pBiDi!=NULL) { return pBiDi->length; } else { return 0; } } U_CAPI UBiDiLevel U_EXPORT2 ubidi_getParaLevel(const UBiDi *pBiDi) { if(pBiDi!=NULL) { return pBiDi->paraLevel; } else { return 0; } } /* statetable prototype ----------------------------------------------------- */ /* * This is here for possible future * performance work and is not compiled right now. */ #if 0 /* * This is a piece of code that could be part of ubidi.c/resolveImplicitLevels(). * It replaces in the (Wn) state machine the switch()-if()-cascade with * just a few if()s and a state table. */ /* use the state table only for the following dirProp's */ #define MASK_W_TABLE (FLAG(L)|FLAG(R)|FLAG(AL)|FLAG(EN)|FLAG(ES)|FLAG(CS)|FLAG(ET)|FLAG(AN)) /* * inputs: * * 0..1 historyOfEN - 2b * 2 prevDirProp==AN - 1b * 3..4 lastStrong, one of { L, R, AL, none } - 2b * 5..7 dirProp, one of { L, R, AL, EN, ES, CS, ET, AN } - 3b * 8..9 nextDirProp, one of { EN, AN, other } * * total: 10b=1024 states */ enum { _L, _R, _AL, _EN, _ES, _CS, _ET, _AN, _OTHER }; /* lastStrong, dirProp */ enum { __EN, __AN, __OTHER }; /* nextDirProp */ #define LAST_STRONG_SHIFT 3 #define DIR_PROP_SHIFT 5 #define NEXT_DIR_PROP_SHIFT 8 /* masks after shifting */ #define LAST_STRONG_MASK 3 #define DIR_PROP_MASK 7 #define STATE_MASK 0x1f /* convert dirProp into _dirProp (above enum) */ static DirProp inputDirProp[dirPropCount]={ _X<>DIR_PROP_SHIFT]; state&=STATE_MASK; } else if(dirProp==ET) { /* get sequence of ET; advance only next, not current, previous or historyOfEN */ while(next>DIR_PROP_SHIFT]; state&=STATE_MASK; /* apply the result of (W1), (W5)..(W7) to the entire sequence of ET */ } else if(dirProp==NSM) { /* (W1) */ dirProp=prevDirProp; /* keep prevDirProp's EN and AN states! */ } else /* other */ { /* set EN and AN states to 0 */ state&=LAST_STRONG_MASK<