// © 2016 and later: Unicode, Inc. and others. // License & terms of use: http://www.unicode.org/copyright.html /* ****************************************************************************** * * Copyright (C) 1999-2015, International Business Machines * Corporation and others. All Rights Reserved. * ****************************************************************************** * file name: ubidi.c * encoding: UTF-8 * tab size: 8 (not used) * indentation:4 * * created on: 1999jul27 * created by: Markus W. Scherer, updated by Matitiahu Allouche * */ #include "cmemory.h" #include "unicode/utypes.h" #include "unicode/ustring.h" #include "unicode/uchar.h" #include "unicode/ubidi.h" #include "unicode/utf16.h" #include "ubidi_props.h" #include "ubidiimp.h" #include "uassert.h" /* * General implementation notes: * * Throughout the implementation, there are comments like (W2) that refer to * rules of the BiDi algorithm, in this example to the second rule of the * resolution of weak types. * * For handling surrogate pairs, where two char16_t's form one "abstract" (or UTF-32) * character according to UTF-16, the second char16_t 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 (char16_t), 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 wherever 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 modifies the dirProps * after the initial setup, when applying X5c (replace FSI by LRI or RLI), * X6, N0 (replace paired brackets by L or R). * * In this implementation, the resolution of weak types (W1 to W6), * neutrals (N1 and N2), 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]= …; static const Flags flagE[2]= …; static const Flags flagO[2]= …; #define DIRPROP_FLAG_LR(level) … #define DIRPROP_FLAG_E(level) … #define DIRPROP_FLAG_O(level) … #define DIR_FROM_STRONG(strong) … #define NO_OVERRIDE(level) … /* UBiDi object management -------------------------------------------------- */ U_CAPI UBiDi * U_EXPORT2 ubidi_open() { … } U_CAPI UBiDi * U_EXPORT2 ubidi_openSized(int32_t maxLength, int32_t maxRunCount, UErrorCode *pErrorCode) { … } /* * We are allowed to allocate memory if memory==nullptr or * mayAllocate==true for each array that we need. * We also try to grow memory as needed if we * allocate it. * * Assume sizeNeeded>0. * If *pMemory!=nullptr, 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(BidiMemoryForAllocation *bidiMem, int32_t *pSize, UBool mayAllocate, int32_t sizeNeeded) { … } U_CAPI void U_EXPORT2 ubidi_close(UBiDi *pBiDi) { … } /* set to approximate "inverse BiDi" ---------------------------------------- */ U_CAPI void U_EXPORT2 ubidi_setInverse(UBiDi *pBiDi, UBool isInverse) { … } U_CAPI UBool U_EXPORT2 ubidi_isInverse(UBiDi *pBiDi) { … } /* FOOD FOR THOUGHT: currently the reordering modes are a mixture of * algorithm for direct BiDi, algorithm for inverse BiDi and the bizarre * concept of RUNS_ONLY which is a double operation. * It could be advantageous to divide this into 3 concepts: * a) Operation: direct / inverse / RUNS_ONLY * b) Direct algorithm: default / NUMBERS_SPECIAL / GROUP_NUMBERS_WITH_R * c) Inverse algorithm: default / INVERSE_LIKE_DIRECT / NUMBERS_SPECIAL * This would allow combinations not possible today like RUNS_ONLY with * NUMBERS_SPECIAL. * Also allow to set INSERT_MARKS for the direct step of RUNS_ONLY and * REMOVE_CONTROLS for the inverse step. * Not all combinations would be supported, and probably not all do make sense. * This would need to document which ones are supported and what are the * fallbacks for unsupported combinations. */ U_CAPI void U_EXPORT2 ubidi_setReorderingMode(UBiDi *pBiDi, UBiDiReorderingMode reorderingMode) UPRV_NO_SANITIZE_UNDEFINED { … } U_CAPI UBiDiReorderingMode U_EXPORT2 ubidi_getReorderingMode(UBiDi *pBiDi) { … } U_CAPI void U_EXPORT2 ubidi_setReorderingOptions(UBiDi *pBiDi, uint32_t reorderingOptions) { … } U_CAPI uint32_t U_EXPORT2 ubidi_getReorderingOptions(UBiDi *pBiDi) { … } U_CAPI UBiDiDirection U_EXPORT2 ubidi_getBaseDirection(const char16_t *text, int32_t length){ … } /* perform (P2)..(P3) ------------------------------------------------------- */ /** * Returns the directionality of the first strong character * after the last B in prologue, if any. * Requires prologue!=null. */ static DirProp firstL_R_AL(UBiDi *pBiDi) { … } /* * Check that there are enough entries in the array pointed to by pBiDi->paras */ static UBool checkParaCount(UBiDi *pBiDi) { … } /* * Get the directional properties for the text, calculate the flags bit-set, and * determine the paragraph level if necessary (in pBiDi->paras[i].level). * FSI initiators are also resolved and their dirProp replaced with LRI or RLI. * When encountering an FSI, it is initially replaced with an LRI, which is the * default. Only if a strong R or AL is found within its scope will the LRI be * replaced by an RLI. */ static UBool getDirProps(UBiDi *pBiDi) { … } /* determine the paragraph level at position index */ U_CFUNC UBiDiLevel ubidi_getParaLevelAtIndex(const UBiDi *pBiDi, int32_t pindex) { … } /* Functions for handling paired brackets ----------------------------------- */ /* In the isoRuns array, the first entry is used for text outside of any isolate sequence. Higher entries are used for each more deeply nested isolate sequence. isoRunLast is the index of the last used entry. The openings array is used to note the data of opening brackets not yet matched by a closing bracket, or matched but still susceptible to change level. Each isoRun entry contains the index of the first and one-after-last openings entries for pending opening brackets it contains. The next openings entry to use is the one-after-last of the most deeply nested isoRun entry. isoRun entries also contain their current embedding level and the last encountered strong character, since these will be needed to resolve the level of paired brackets. */ static void bracketInit(UBiDi *pBiDi, BracketData *bd) { … } /* paragraph boundary */ static void bracketProcessB(BracketData *bd, UBiDiLevel level) { … } /* LRE, LRO, RLE, RLO, PDF */ static void bracketProcessBoundary(BracketData *bd, int32_t lastCcPos, UBiDiLevel contextLevel, UBiDiLevel embeddingLevel) { … } /* LRI or RLI */ static void bracketProcessLRI_RLI(BracketData *bd, UBiDiLevel level) { … } /* PDI */ static void bracketProcessPDI(BracketData *bd) { … } /* newly found opening bracket: create an openings entry */ static UBool /* return true if success */ bracketAddOpening(BracketData *bd, char16_t match, int32_t position) { … } /* change N0c1 to N0c2 when a preceding bracket is assigned the embedding level */ static void fixN0c(BracketData *bd, int32_t openingIndex, int32_t newPropPosition, DirProp newProp) { … } /* process closing bracket */ static DirProp /* return L or R if N0b or N0c, ON if N0d */ bracketProcessClosing(BracketData *bd, int32_t openIdx, int32_t position) { … } /* handle strong characters, digits and candidates for closing brackets */ static UBool /* return true if success */ bracketProcessChar(BracketData *bd, int32_t position) { … } /* perform (X1)..(X9) ------------------------------------------------------- */ /* determine if the text is mixed-directional or single-directional */ static UBiDiDirection directionFromFlags(UBiDi *pBiDi) { … } /* * 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, FSI, PDI) in the plain text. * That is why (X9) instructs to remove all not-isolate explicit codes (and BN). * However, in a real implementation, the 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 a not-isolate * 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, except for * paired brackets whose properties (ON) may be replaced by L or R. * 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, RLO, LRI, RLI and FSI and popped with each PDF and PDI, * the explicit level must never exceed UBIDI_MAX_EXPLICIT_LEVEL. * * In order to have a correct push-pop semantics even in the case of overflows, * overflow counters and a valid isolate counter are used as described in UAX#9 * section 3.3.2 "Explicit Levels and Directions". * * This implementation assumes that UBIDI_MAX_EXPLICIT_LEVEL is odd. * * Returns normally the direction; -1 if there was a memory shortage * */ static UBiDiDirection resolveExplicitLevels(UBiDi *pBiDi, UErrorCode *pErrorCode) { … } /* * 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) { … } /****************************************************************** The Properties state machine table ******************************************************************* All table cells are 8 bits: bits 0..4: next state bits 5..7: action to perform (if > 0) Cells may be of format "n" where n represents the next state (except for the rightmost column). Cells may also be of format "s(x,y)" where x represents an action to perform and y represents the next state. ******************************************************************* Definitions and type for properties state table ******************************************************************* */ #define IMPTABPROPS_COLUMNS … #define IMPTABPROPS_RES … #define GET_STATEPROPS(cell) … #define GET_ACTIONPROPS(cell) … #define s … static const uint8_t groupProp[] = …/* dirProp regrouped */ { … }; enum { … }; /* reduced dirProp */ /****************************************************************** PROPERTIES STATE TABLE In table impTabProps, - the ON column regroups ON and WS, FSI, RLI, LRI and PDI - the BN column regroups BN, LRE, RLE, LRO, RLO, PDF - the Res column is the reduced property assigned to a run Action 1: process current run1, init new run1 2: init new run2 3: process run1, process run2, init new run1 4: process run1, set run1=run2, init new run2 Notes: 1) This table is used in resolveImplicitLevels(). 2) This table triggers actions when there is a change in the Bidi property of incoming characters (action 1). 3) Most such property sequences are processed immediately (in fact, passed to processPropertySeq(). 4) However, numbers are assembled as one sequence. This means that undefined situations (like CS following digits, until it is known if the next char will be a digit) are held until following chars define them. Example: digits followed by CS, then comes another CS or ON; the digits will be processed, then the CS assigned as the start of an ON sequence (action 3). 5) There are cases where more than one sequence must be processed, for instance digits followed by CS followed by L: the digits must be processed as one sequence, and the CS must be processed as an ON sequence, all this before starting assembling chars for the opening L sequence. */ static const uint8_t impTabProps[][IMPTABPROPS_COLUMNS] = …; /* we must undef macro s because the levels tables have a different * structure (4 bits for action and 4 bits for next state. */ #undef s /****************************************************************** The levels state machine tables ******************************************************************* All table cells are 8 bits: bits 0..3: next state bits 4..7: action to perform (if > 0) Cells may be of format "n" where n represents the next state (except for the rightmost column). Cells may also be of format "s(x,y)" where x represents an action to perform and y represents the next state. This format limits each table to 16 states each and to 15 actions. ******************************************************************* Definitions and type for levels state tables ******************************************************************* */ #define IMPTABLEVELS_COLUMNS … #define IMPTABLEVELS_RES … #define GET_STATE(cell) … #define GET_ACTION(cell) … #define s … ImpTab; ImpAct; /* FOOD FOR THOUGHT: each ImpTab should have its associated ImpAct, * instead of having a pair of ImpTab and a pair of ImpAct. */ ImpTabPair; /****************************************************************** LEVELS STATE TABLES In all levels state tables, - state 0 is the initial state - the Res column is the increment to add to the text level for this property sequence. The impAct arrays for each table of a pair map the local action numbers of the table to the total list of actions. For instance, action 2 in a given table corresponds to the action number which appears in entry [2] of the impAct array for that table. The first entry of all impAct arrays must be 0. Action 1: init conditional sequence 2: prepend conditional sequence to current sequence 3: set ON sequence to new level - 1 4: init EN/AN/ON sequence 5: fix EN/AN/ON sequence followed by R 6: set previous level sequence to level 2 Notes: 1) These tables are used in processPropertySeq(). The input is property sequences as determined by resolveImplicitLevels. 2) Most such property sequences are processed immediately (levels are assigned). 3) However, some sequences cannot be assigned a final level till one or more following sequences are received. For instance, ON following an R sequence within an even-level paragraph. If the following sequence is R, the ON sequence will be assigned basic run level+1, and so will the R sequence. 4) S is generally handled like ON, since its level will be fixed to paragraph level in adjustWSLevels(). */ static const ImpTab impTabL_DEFAULT = …/* Even paragraph level */ /* In this table, conditional sequences receive the lower possible level until proven otherwise. */ { … }; static const ImpTab impTabR_DEFAULT = …/* Odd paragraph level */ /* In this table, conditional sequences receive the lower possible level until proven otherwise. */ { … }; static const ImpAct impAct0 = …; static const ImpTabPair impTab_DEFAULT = …; static const ImpTab impTabL_NUMBERS_SPECIAL = …/* Even paragraph level */ /* In this table, conditional sequences receive the lower possible level until proven otherwise. */ { … }; static const ImpTabPair impTab_NUMBERS_SPECIAL = …; static const ImpTab impTabL_GROUP_NUMBERS_WITH_R = …/* In this table, EN/AN+ON sequences receive levels as if associated with R until proven that there is L or sor/eor on both sides. AN is handled like EN. */ { … }; static const ImpTab impTabR_GROUP_NUMBERS_WITH_R = …/* In this table, EN/AN+ON sequences receive levels as if associated with R until proven that there is L on both sides. AN is handled like EN. */ { … }; static const ImpTabPair impTab_GROUP_NUMBERS_WITH_R = …; static const ImpTab impTabL_INVERSE_NUMBERS_AS_L = …/* This table is identical to the Default LTR table except that EN and AN are handled like L. */ { … }; static const ImpTab impTabR_INVERSE_NUMBERS_AS_L = …/* This table is identical to the Default RTL table except that EN and AN are handled like L. */ { … }; static const ImpTabPair impTab_INVERSE_NUMBERS_AS_L = …; static const ImpTab impTabR_INVERSE_LIKE_DIRECT = …/* Odd paragraph level */ /* In this table, conditional sequences receive the lower possible level until proven otherwise. */ { … }; static const ImpAct impAct1 = …; /* FOOD FOR THOUGHT: in LTR table below, check case "JKL 123abc" */ static const ImpTabPair impTab_INVERSE_LIKE_DIRECT = …; static const ImpTab impTabL_INVERSE_LIKE_DIRECT_WITH_MARKS = …/* The case handled in this table is (visually): R EN L */ { … }; static const ImpTab impTabR_INVERSE_LIKE_DIRECT_WITH_MARKS = …/* The cases handled in this table are (visually): R EN L R L AN L */ { … }; static const ImpAct impAct2 = …; static const ImpAct impAct3 = …; static const ImpTabPair impTab_INVERSE_LIKE_DIRECT_WITH_MARKS = …; static const ImpTabPair impTab_INVERSE_FOR_NUMBERS_SPECIAL = …; static const ImpTab impTabL_INVERSE_FOR_NUMBERS_SPECIAL_WITH_MARKS = …/* The case handled in this table is (visually): R EN L */ { … }; static const ImpTabPair impTab_INVERSE_FOR_NUMBERS_SPECIAL_WITH_MARKS = …; #undef s LevState; /*------------------------------------------------------------------------*/ static void addPoint(UBiDi *pBiDi, int32_t pos, int32_t flag) /* param pos: position where to insert param flag: one of LRM_BEFORE, LRM_AFTER, RLM_BEFORE, RLM_AFTER */ { … } static void setLevelsOutsideIsolates(UBiDi *pBiDi, int32_t start, int32_t limit, UBiDiLevel level) { … } /* 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 (p<q). * * The (Nn) and (In) rules are also performed in that same single loop, * but effectively one iteration behind for white space. * * Since all implicit rules are performed in one step, it is not necessary * to actually store the intermediate directional properties in dirProps[]. */ static void processPropertySeq(UBiDi *pBiDi, LevState *pLevState, uint8_t _prop, int32_t start, int32_t limit) { … } /** * Returns the directionality of the last strong character at the end of the prologue, if any. * Requires prologue!=null. */ static DirProp lastL_R_AL(UBiDi *pBiDi) { … } /** * Returns the directionality of the first strong character, or digit, in the epilogue, if any. * Requires epilogue!=null. */ static DirProp firstL_R_AL_EN_AN(UBiDi *pBiDi) { … } static void resolveImplicitLevels(UBiDi *pBiDi, int32_t start, int32_t limit, DirProp sor, DirProp eor) { … } /* perform (L1) and (X9) ---------------------------------------------------- */ /* * Reset the embedding levels for some non-graphic characters (L1). * This function also sets appropriate levels for BN, and * explicit embedding types that are supposed to have been removed * from the paragraph in (X9). */ static void adjustWSLevels(UBiDi *pBiDi) { … } U_CAPI void U_EXPORT2 ubidi_setContext(UBiDi *pBiDi, const char16_t *prologue, int32_t proLength, const char16_t *epilogue, int32_t epiLength, UErrorCode *pErrorCode) { … } static void setParaSuccess(UBiDi *pBiDi) { … } #define BIDI_MIN(x, y) … #define BIDI_ABS(x) … static void setParaRunsOnly(UBiDi *pBiDi, const char16_t *text, int32_t length, UBiDiLevel paraLevel, UErrorCode *pErrorCode) { … } /* ubidi_setPara ------------------------------------------------------------ */ U_CAPI void U_EXPORT2 ubidi_setPara(UBiDi *pBiDi, const char16_t *text, int32_t length, UBiDiLevel paraLevel, UBiDiLevel *embeddingLevels, UErrorCode *pErrorCode) { … } U_CAPI void U_EXPORT2 ubidi_orderParagraphsLTR(UBiDi *pBiDi, UBool orderParagraphsLTR) { … } U_CAPI UBool U_EXPORT2 ubidi_isOrderParagraphsLTR(UBiDi *pBiDi) { … } U_CAPI UBiDiDirection U_EXPORT2 ubidi_getDirection(const UBiDi *pBiDi) { … } U_CAPI const char16_t * U_EXPORT2 ubidi_getText(const UBiDi *pBiDi) { … } U_CAPI int32_t U_EXPORT2 ubidi_getLength(const UBiDi *pBiDi) { … } U_CAPI int32_t U_EXPORT2 ubidi_getProcessedLength(const UBiDi *pBiDi) { … } U_CAPI int32_t U_EXPORT2 ubidi_getResultLength(const UBiDi *pBiDi) { … } /* paragraphs API functions ------------------------------------------------- */ U_CAPI UBiDiLevel U_EXPORT2 ubidi_getParaLevel(const UBiDi *pBiDi) { … } U_CAPI int32_t U_EXPORT2 ubidi_countParagraphs(UBiDi *pBiDi) { … } U_CAPI void U_EXPORT2 ubidi_getParagraphByIndex(const UBiDi *pBiDi, int32_t paraIndex, int32_t *pParaStart, int32_t *pParaLimit, UBiDiLevel *pParaLevel, UErrorCode *pErrorCode) { … } U_CAPI int32_t U_EXPORT2 ubidi_getParagraph(const UBiDi *pBiDi, int32_t charIndex, int32_t *pParaStart, int32_t *pParaLimit, UBiDiLevel *pParaLevel, UErrorCode *pErrorCode) { … } U_CAPI void U_EXPORT2 ubidi_setClassCallback(UBiDi *pBiDi, UBiDiClassCallback *newFn, const void *newContext, UBiDiClassCallback **oldFn, const void **oldContext, UErrorCode *pErrorCode) { … } U_CAPI void U_EXPORT2 ubidi_getClassCallback(UBiDi *pBiDi, UBiDiClassCallback **fn, const void **context) { … } U_CAPI UCharDirection U_EXPORT2 ubidi_getCustomizedClass(UBiDi *pBiDi, UChar32 c) { … }
Codebase Browser
chromium