// © 2016 and later: Unicode, Inc. and others. // License & terms of use: http://www.unicode.org/copyright.html /* ******************************************************************************* * Copyright (C) 1996-2014, International Business Machines Corporation and * others. All Rights Reserved. ******************************************************************************* */ #ifndef CANITER_H #define CANITER_H #include "unicode/utypes.h" #if U_SHOW_CPLUSPLUS_API #if !UCONFIG_NO_NORMALIZATION #include "unicode/uobject.h" #include "unicode/unistr.h" /** * \file * \brief C++ API: Canonical Iterator */ /** Should permutation skip characters with combining class zero * Should be either true or false. This is a compile time option * @stable ICU 2.4 */ #ifndef CANITER_SKIP_ZEROES #define CANITER_SKIP_ZEROES true #endif U_NAMESPACE_BEGIN class Hashtable; class Normalizer2; class Normalizer2Impl; /** * This class allows one to iterate through all the strings that are canonically equivalent to a given * string. For example, here are some sample results: Results for: {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA} 1: \\u0041\\u030A\\u0064\\u0307\\u0327 = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA} 2: \\u0041\\u030A\\u0064\\u0327\\u0307 = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE} 3: \\u0041\\u030A\\u1E0B\\u0327 = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA} 4: \\u0041\\u030A\\u1E11\\u0307 = {LATIN CAPITAL LETTER A}{COMBINING RING ABOVE}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE} 5: \\u00C5\\u0064\\u0307\\u0327 = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA} 6: \\u00C5\\u0064\\u0327\\u0307 = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE} 7: \\u00C5\\u1E0B\\u0327 = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA} 8: \\u00C5\\u1E11\\u0307 = {LATIN CAPITAL LETTER A WITH RING ABOVE}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE} 9: \\u212B\\u0064\\u0307\\u0327 = {ANGSTROM SIGN}{LATIN SMALL LETTER D}{COMBINING DOT ABOVE}{COMBINING CEDILLA} 10: \\u212B\\u0064\\u0327\\u0307 = {ANGSTROM SIGN}{LATIN SMALL LETTER D}{COMBINING CEDILLA}{COMBINING DOT ABOVE} 11: \\u212B\\u1E0B\\u0327 = {ANGSTROM SIGN}{LATIN SMALL LETTER D WITH DOT ABOVE}{COMBINING CEDILLA} 12: \\u212B\\u1E11\\u0307 = {ANGSTROM SIGN}{LATIN SMALL LETTER D WITH CEDILLA}{COMBINING DOT ABOVE} *
Note: the code is intended for use with small strings, and is not suitable for larger ones, * since it has not been optimized for that situation. * Note, CanonicalIterator is not intended to be subclassed. * @author M. Davis * @author C++ port by V. Weinstein * @stable ICU 2.4 */ class U_COMMON_API CanonicalIterator final : public UObject { public: /** * Construct a CanonicalIterator object * @param source string to get results for * @param status Fill-in parameter which receives the status of this operation. * @stable ICU 2.4 */ CanonicalIterator(const UnicodeString &source, UErrorCode &status); /** Destructor * Cleans pieces * @stable ICU 2.4 */ virtual ~CanonicalIterator(); /** * Gets the NFD form of the current source we are iterating over. * @return gets the source: NOTE: it is the NFD form of source * @stable ICU 2.4 */ UnicodeString getSource(); /** * Resets the iterator so that one can start again from the beginning. * @stable ICU 2.4 */ void reset(); /** * Get the next canonically equivalent string. *
Warning: The strings are not guaranteed to be in any particular order. * @return the next string that is canonically equivalent. A bogus string is returned when * the iteration is done. * @stable ICU 2.4 */ UnicodeString next(); /** * Set a new source for this iterator. Allows object reuse. * @param newSource the source string to iterate against. This allows the same iterator to be used * while changing the source string, saving object creation. * @param status Fill-in parameter which receives the status of this operation. * @stable ICU 2.4 */ void setSource(const UnicodeString &newSource, UErrorCode &status); #ifndef U_HIDE_INTERNAL_API /** * Dumb recursive implementation of permutation. * TODO: optimize * @param source the string to find permutations for * @param skipZeros determine if skip zeros * @param result the results in a set. * @param status Fill-in parameter which receives the status of this operation. * @param depth depth of the call. * @internal */ static void U_EXPORT2 permute(UnicodeString &source, UBool skipZeros, Hashtable *result, UErrorCode &status, int32_t depth=0); #endif /* U_HIDE_INTERNAL_API */ /** * ICU "poor man's RTTI", returns a UClassID for this class. * * @stable ICU 2.2 */ static UClassID U_EXPORT2 getStaticClassID(); /** * ICU "poor man's RTTI", returns a UClassID for the actual class. * * @stable ICU 2.2 */ virtual UClassID getDynamicClassID() const override; private: // ===================== PRIVATES ============================== // private default constructor CanonicalIterator() = delete; /** * Copy constructor. Private for now. * @internal (private) */ CanonicalIterator(const CanonicalIterator& other) = delete; /** * Assignment operator. Private for now. * @internal (private) */ CanonicalIterator& operator=(const CanonicalIterator& other) = delete; // fields UnicodeString source; UBool done; // 2 dimensional array holds the pieces of the string with // their different canonically equivalent representations UnicodeString **pieces; int32_t pieces_length; int32_t *pieces_lengths; // current is used in iterating to combine pieces int32_t *current; int32_t current_length; // transient fields UnicodeString buffer; const Normalizer2 *nfd; const Normalizer2Impl *nfcImpl; // we have a segment, in NFD. Find all the strings that are canonically equivalent to it. UnicodeString *getEquivalents(const UnicodeString &segment, int32_t &result_len, UErrorCode &status); //private String[] getEquivalents(String segment) //Set getEquivalents2(String segment); Hashtable *getEquivalents2(Hashtable *fillinResult, const char16_t *segment, int32_t segLen, UErrorCode &status); //Hashtable *getEquivalents2(const UnicodeString &segment, int32_t segLen, UErrorCode &status); /** * See if the decomposition of cp2 is at segment starting at segmentPos * (with canonical rearrangement!) * If so, take the remainder, and return the equivalents */ //Set extract(int comp, String segment, int segmentPos, StringBuffer buffer); Hashtable *extract(Hashtable *fillinResult, UChar32 comp, const char16_t *segment, int32_t segLen, int32_t segmentPos, UErrorCode &status); //Hashtable *extract(UChar32 comp, const UnicodeString &segment, int32_t segLen, int32_t segmentPos, UErrorCode &status); void cleanPieces(); }; U_NAMESPACE_END #endif /* #if !UCONFIG_NO_NORMALIZATION */ #endif /* U_SHOW_CPLUSPLUS_API */ #endif