summaryrefslogtreecommitdiff
path: root/xpcom/string/nsStringObsolete.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'xpcom/string/nsStringObsolete.cpp')
-rw-r--r--xpcom/string/nsStringObsolete.cpp1053
1 files changed, 1053 insertions, 0 deletions
diff --git a/xpcom/string/nsStringObsolete.cpp b/xpcom/string/nsStringObsolete.cpp
new file mode 100644
index 0000000000..bd6daacab3
--- /dev/null
+++ b/xpcom/string/nsStringObsolete.cpp
@@ -0,0 +1,1053 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim: set ts=8 sts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsString.h"
+
+
+/**
+ * nsTString obsolete API support
+ */
+
+#if MOZ_STRING_WITH_OBSOLETE_API
+
+#include "nsDependentString.h"
+#include "nsDependentSubstring.h"
+#include "nsReadableUtils.h"
+#include "nsCRT.h"
+#include "nsUTF8Utils.h"
+#include "prdtoa.h"
+
+/* ***** BEGIN RICKG BLOCK *****
+ *
+ * NOTE: This section of code was extracted from rickg's bufferRoutines.h file.
+ * For the most part it remains unmodified. We want to eliminate (or at
+ * least clean up) this code at some point. If you find the formatting
+ * in this section somewhat inconsistent, don't blame me! ;-)
+ */
+
+// avoid STDC's tolower since it may do weird things with non-ASCII bytes
+inline char
+ascii_tolower(char aChar)
+{
+ if (aChar >= 'A' && aChar <= 'Z')
+ return aChar + ('a' - 'A');
+ return aChar;
+}
+
+//-----------------------------------------------------------------------------
+//
+// This set of methods is used to search a buffer looking for a char.
+//
+
+
+/**
+ * This methods cans the given buffer for the given char
+ *
+ * @update gess 02/17/00
+ * @param aDest is the buffer to be searched
+ * @param aDestLength is the size (in char-units, not bytes) of the buffer
+ * @param anOffset is the start pos to begin searching
+ * @param aChar is the target character we're looking for
+ * @param aCount tells us how many characters to iterate through (which may be different than aLength); -1 means use full length.
+ * @return index of pos if found, else -1 (kNotFound)
+ */
+static int32_t
+FindChar1(const char* aDest,uint32_t aDestLength,int32_t anOffset,const char16_t aChar,int32_t aCount) {
+
+ if(anOffset < 0)
+ anOffset=0;
+
+ if(aCount < 0)
+ aCount = (int32_t)aDestLength;
+
+ if((aChar < 256) && (0 < aDestLength) && ((uint32_t)anOffset < aDestLength)) {
+
+ //We'll only search if the given aChar is within the normal ascii a range,
+ //(Since this string is definitely within the ascii range).
+
+ if(0<aCount) {
+
+ const char* left= aDest+anOffset;
+ const char* last= left+aCount;
+ const char* max = aDest+aDestLength;
+ const char* end = (last<max) ? last : max;
+
+ int32_t theMax = end-left;
+ if(0<theMax) {
+
+ unsigned char theChar = (unsigned char) aChar;
+ const char* result=(const char*)memchr(left, (int)theChar, theMax);
+
+ if(result)
+ return result-aDest;
+
+ }
+ }
+ }
+
+ return kNotFound;
+}
+
+
+/**
+ * This methods cans the given buffer for the given char
+ *
+ * @update gess 3/25/98
+ * @param aDest is the buffer to be searched
+ * @param aDestLength is the size (in char-units, not bytes) of the buffer
+ * @param anOffset is the start pos to begin searching
+ * @param aChar is the target character we're looking for
+ * @param aCount tells us how many characters to iterate through (which may be different than aLength); -1 means use full length.
+ * @return index of pos if found, else -1 (kNotFound)
+ */
+static int32_t
+FindChar2(const char16_t* aDest,uint32_t aDestLength,int32_t anOffset,const char16_t aChar,int32_t aCount) {
+
+ if(anOffset < 0)
+ anOffset=0;
+
+ if(aCount < 0)
+ aCount = (int32_t)aDestLength;
+
+ if((0<aDestLength) && ((uint32_t)anOffset < aDestLength)) {
+
+ if(0<aCount) {
+
+ const char16_t* root = aDest;
+ const char16_t* left = root+anOffset;
+ const char16_t* last = left+aCount;
+ const char16_t* max = root+aDestLength;
+ const char16_t* end = (last<max) ? last : max;
+
+ while(left<end){
+
+ if(*left==aChar)
+ return (left-root);
+
+ ++left;
+ }
+ }
+ }
+
+ return kNotFound;
+}
+
+
+/**
+ * This methods cans the given buffer (in reverse) for the given char
+ *
+ * @update gess 02/17/00
+ * @param aDest is the buffer to be searched
+ * @param aDestLength is the size (in char-units, not bytes) of the buffer
+ * @param anOffset is the start pos to begin searching
+ * @param aChar is the target character we're looking for
+ * @param aCount tells us how many characters to iterate through (which may be different than aLength); -1 means use full length.
+ * @return index of pos if found, else -1 (kNotFound)
+ */
+
+static int32_t
+RFindChar1(const char* aDest,uint32_t aDestLength,int32_t anOffset,const char16_t aChar,int32_t aCount) {
+
+ if(anOffset < 0)
+ anOffset=(int32_t)aDestLength-1;
+
+ if(aCount < 0)
+ aCount = int32_t(aDestLength);
+
+ if((aChar<256) && (0 < aDestLength) && ((uint32_t)anOffset < aDestLength)) {
+
+ //We'll only search if the given aChar is within the normal ascii a range,
+ //(Since this string is definitely within the ascii range).
+
+ if(0 < aCount) {
+
+ const char* rightmost = aDest + anOffset;
+ const char* min = rightmost - aCount + 1;
+ const char* leftmost = (min<aDest) ? aDest: min;
+
+ char theChar=(char)aChar;
+ while(leftmost <= rightmost){
+
+ if((*rightmost) == theChar)
+ return rightmost - aDest;
+
+ --rightmost;
+ }
+ }
+ }
+
+ return kNotFound;
+}
+
+
+/**
+ * This methods cans the given buffer for the given char
+ *
+ * @update gess 3/25/98
+ * @param aDest is the buffer to be searched
+ * @param aDestLength is the size (in char-units, not bytes) of the buffer
+ * @param anOffset is the start pos to begin searching
+ * @param aChar is the target character we're looking for
+ * @param aCount tells us how many characters to iterate through (which may be different than aLength); -1 means use full length.
+ * @return index of pos if found, else -1 (kNotFound)
+ */
+static int32_t
+RFindChar2(const char16_t* aDest,uint32_t aDestLength,int32_t anOffset,const char16_t aChar,int32_t aCount) {
+
+ if(anOffset < 0)
+ anOffset=(int32_t)aDestLength-1;
+
+ if(aCount < 0)
+ aCount = int32_t(aDestLength);
+
+ if((0 < aDestLength) && ((uint32_t)anOffset < aDestLength)) {
+
+ if(0 < aCount) {
+
+ const char16_t* root = aDest;
+ const char16_t* rightmost = root + anOffset;
+ const char16_t* min = rightmost - aCount + 1;
+ const char16_t* leftmost = (min<root) ? root: min;
+
+ while(leftmost <= rightmost){
+
+ if((*rightmost) == aChar)
+ return rightmost - root;
+
+ --rightmost;
+ }
+ }
+ }
+
+ return kNotFound;
+}
+
+//-----------------------------------------------------------------------------
+//
+// This set of methods is used to compare one buffer onto another. The
+// functions are differentiated by the size of source and dest character
+// sizes. WARNING: Your destination buffer MUST be big enough to hold all the
+// source bytes. We don't validate these ranges here (this should be done in
+// higher level routines).
+//
+
+
+/**
+ * This method compares the data in one buffer with another
+ * @update gess 01/04/99
+ * @param aStr1 is the first buffer to be compared
+ * @param aStr2 is the 2nd buffer to be compared
+ * @param aCount is the number of chars to compare
+ * @param aIgnoreCase tells us whether to use a case-sensitive comparison
+ * @return -1,0,1 depending on <,==,>
+ */
+static
+#ifdef __SUNPRO_CC
+inline
+#endif /* __SUNPRO_CC */
+int32_t
+Compare1To1(const char* aStr1,const char* aStr2,uint32_t aCount,bool aIgnoreCase) {
+ int32_t result=0;
+ if(aIgnoreCase)
+ result=int32_t(PL_strncasecmp(aStr1, aStr2, aCount));
+ else
+ result=nsCharTraits<char>::compare(aStr1,aStr2,aCount);
+
+ // alien comparisons may return out-of-bound answers
+ // instead of the -1, 0, 1 expected by most clients
+ if ( result < -1 )
+ result = -1;
+ else if ( result > 1 )
+ result = 1;
+ return result;
+}
+
+/**
+ * This method compares the data in one buffer with another
+ * @update gess 01/04/99
+ * @param aStr1 is the first buffer to be compared
+ * @param aStr2 is the 2nd buffer to be compared
+ * @param aCount is the number of chars to compare
+ * @param aIgnoreCase tells us whether to use a case-sensitive comparison
+ * @return -1,0,1 depending on <,==,>
+ */
+static
+#ifdef __SUNPRO_CC
+inline
+#endif /* __SUNPRO_CC */
+int32_t
+Compare2To2(const char16_t* aStr1,const char16_t* aStr2,uint32_t aCount){
+ int32_t result;
+
+ if ( aStr1 && aStr2 )
+ result = nsCharTraits<char16_t>::compare(aStr1, aStr2, aCount);
+
+ // The following cases are rare and survivable caller errors.
+ // Two null pointers are equal, but any string, even 0 length
+ // is greater than a null pointer. It might not really matter,
+ // but we pick something reasonable anyway.
+ else if ( !aStr1 && !aStr2 )
+ result = 0;
+ else if ( aStr1 )
+ result = 1;
+ else
+ result = -1;
+
+ // alien comparisons may give answers outside the -1, 0, 1 expected by callers
+ if ( result < -1 )
+ result = -1;
+ else if ( result > 1 )
+ result = 1;
+ return result;
+}
+
+
+/**
+ * This method compares the data in one buffer with another
+ * @update gess 01/04/99
+ * @param aStr1 is the first buffer to be compared
+ * @param aStr2 is the 2nd buffer to be compared
+ * @param aCount is the number of chars to compare
+ * @param aIgnoreCase tells us whether to use a case-sensitive comparison
+ * @return -1,0,1 depending on <,==,>
+ */
+static
+#ifdef __SUNPRO_CC
+inline
+#endif /* __SUNPRO_CC */
+int32_t
+Compare2To1(const char16_t* aStr1,const char* aStr2,uint32_t aCount,bool aIgnoreCase){
+ const char16_t* s1 = aStr1;
+ const char *s2 = aStr2;
+
+ if (aStr1 && aStr2) {
+ if (aCount != 0) {
+ do {
+
+ char16_t c1 = *s1++;
+ char16_t c2 = char16_t((unsigned char)*s2++);
+
+ if (c1 != c2) {
+#ifdef DEBUG
+ // we won't warn on c1>=128 (the 2-byte value) because often
+ // it is just fine to compare an constant, ascii value (i.e. "body")
+ // against some non-ascii value (i.e. a unicode string that
+ // was downloaded from a web page)
+ if (aIgnoreCase && c2>=128)
+ NS_WARNING("got a non-ASCII string, but we can't do an accurate case conversion!");
+#endif
+
+ // can't do case conversion on characters out of our range
+ if (aIgnoreCase && c1<128 && c2<128) {
+
+ c1 = ascii_tolower(char(c1));
+ c2 = ascii_tolower(char(c2));
+
+ if (c1 == c2) continue;
+ }
+
+ if (c1 < c2) return -1;
+ return 1;
+ }
+ } while (--aCount);
+ }
+ }
+ return 0;
+}
+
+
+/**
+ * This method compares the data in one buffer with another
+ * @update gess 01/04/99
+ * @param aStr1 is the first buffer to be compared
+ * @param aStr2 is the 2nd buffer to be compared
+ * @param aCount is the number of chars to compare
+ * @param aIgnoreCase tells us whether to use a case-sensitive comparison
+ * @return -1,0,1 depending on <,==,>
+ */
+inline int32_t
+Compare1To2(const char* aStr1,const char16_t* aStr2,uint32_t aCount,bool aIgnoreCase){
+ return Compare2To1(aStr2, aStr1, aCount, aIgnoreCase) * -1;
+}
+
+
+//-----------------------------------------------------------------------------
+//
+// This set of methods is used compress char sequences in a buffer...
+//
+
+
+/**
+ * This method compresses duplicate runs of a given char from the given buffer
+ *
+ * @update rickg 03.23.2000
+ * @param aString is the buffer to be manipulated
+ * @param aLength is the length of the buffer
+ * @param aSet tells us which chars to compress from given buffer
+ * @param aEliminateLeading tells us whether to strip chars from the start of the buffer
+ * @param aEliminateTrailing tells us whether to strip chars from the start of the buffer
+ * @return the new length of the given buffer
+ */
+static int32_t
+CompressChars1(char* aString,uint32_t aLength,const char* aSet){
+
+ char* from = aString;
+ char* end = aString + aLength;
+ char* to = from;
+
+ //this code converts /n, /t, /r into normal space ' ';
+ //it also compresses runs of whitespace down to a single char...
+ if(aSet && aString && (0 < aLength)){
+ uint32_t aSetLen=strlen(aSet);
+
+ while (from < end) {
+ char theChar = *from++;
+
+ *to++=theChar; //always copy this char...
+
+ if((kNotFound!=FindChar1(aSet,aSetLen,0,theChar,aSetLen))){
+ while (from < end) {
+ theChar = *from++;
+ if(kNotFound==FindChar1(aSet,aSetLen,0,theChar,aSetLen)){
+ *to++ = theChar;
+ break;
+ }
+ } //while
+ } //if
+ } //if
+ *to = 0;
+ }
+ return to - aString;
+}
+
+
+
+/**
+ * This method compresses duplicate runs of a given char from the given buffer
+ *
+ * @update rickg 03.23.2000
+ * @param aString is the buffer to be manipulated
+ * @param aLength is the length of the buffer
+ * @param aSet tells us which chars to compress from given buffer
+ * @param aEliminateLeading tells us whether to strip chars from the start of the buffer
+ * @param aEliminateTrailing tells us whether to strip chars from the start of the buffer
+ * @return the new length of the given buffer
+ */
+static int32_t
+CompressChars2(char16_t* aString,uint32_t aLength,const char* aSet) {
+
+ char16_t* from = aString;
+ char16_t* end = from + aLength;
+ char16_t* to = from;
+
+ //this code converts /n, /t, /r into normal space ' ';
+ //it also compresses runs of whitespace down to a single char...
+ if(aSet && aString && (0 < aLength)){
+ uint32_t aSetLen=strlen(aSet);
+
+ while (from < end) {
+ char16_t theChar = *from++;
+
+ *to++=theChar; //always copy this char...
+
+ if((theChar<256) && (kNotFound!=FindChar1(aSet,aSetLen,0,theChar,aSetLen))){
+ while (from < end) {
+ theChar = *from++;
+ if(kNotFound==FindChar1(aSet,aSetLen,0,theChar,aSetLen)){
+ *to++ = theChar;
+ break;
+ }
+ } //while
+ } //if
+ } //if
+ *to = 0;
+ }
+ return to - (char16_t*)aString;
+}
+
+/**
+ * This method strips chars in a given set from the given buffer
+ *
+ * @update gess 01/04/99
+ * @param aString is the buffer to be manipulated
+ * @param aLength is the length of the buffer
+ * @param aSet tells us which chars to compress from given buffer
+ * @param aEliminateLeading tells us whether to strip chars from the start of the buffer
+ * @param aEliminateTrailing tells us whether to strip chars from the start of the buffer
+ * @return the new length of the given buffer
+ */
+static int32_t
+StripChars1(char* aString,uint32_t aLength,const char* aSet) {
+
+ // XXX(darin): this code should defer writing until necessary.
+
+ char* to = aString;
+ char* from = aString-1;
+ char* end = aString + aLength;
+
+ if(aSet && aString && (0 < aLength)){
+ uint32_t aSetLen=strlen(aSet);
+ while (++from < end) {
+ char theChar = *from;
+ if(kNotFound==FindChar1(aSet,aSetLen,0,theChar,aSetLen)){
+ *to++ = theChar;
+ }
+ }
+ *to = 0;
+ }
+ return to - (char*)aString;
+}
+
+
+/**
+ * This method strips chars in a given set from the given buffer
+ *
+ * @update gess 01/04/99
+ * @param aString is the buffer to be manipulated
+ * @param aLength is the length of the buffer
+ * @param aSet tells us which chars to compress from given buffer
+ * @param aEliminateLeading tells us whether to strip chars from the start of the buffer
+ * @param aEliminateTrailing tells us whether to strip chars from the start of the buffer
+ * @return the new length of the given buffer
+ */
+static int32_t
+StripChars2(char16_t* aString,uint32_t aLength,const char* aSet) {
+
+ // XXX(darin): this code should defer writing until necessary.
+
+ char16_t* to = aString;
+ char16_t* from = aString-1;
+ char16_t* end = to + aLength;
+
+ if(aSet && aString && (0 < aLength)){
+ uint32_t aSetLen=strlen(aSet);
+ while (++from < end) {
+ char16_t theChar = *from;
+ //Note the test for ascii range below. If you have a real unicode char,
+ //and you're searching for chars in the (given) ascii string, there's no
+ //point in doing the real search since it's out of the ascii range.
+ if((255<theChar) || (kNotFound==FindChar1(aSet,aSetLen,0,theChar,aSetLen))){
+ *to++ = theChar;
+ }
+ }
+ *to = 0;
+ }
+ return to - (char16_t*)aString;
+}
+
+/* ***** END RICKG BLOCK ***** */
+
+static const char* kWhitespace="\f\t\r\n ";
+
+// This function is used to implement FindCharInSet and friends
+template <class CharT>
+#ifndef __SUNPRO_CC
+static
+#endif /* !__SUNPRO_CC */
+CharT
+GetFindInSetFilter( const CharT* set)
+{
+ CharT filter = ~CharT(0); // All bits set
+ while (*set) {
+ filter &= ~(*set);
+ ++set;
+ }
+ return filter;
+}
+
+// This template class is used by our code to access rickg's buffer routines.
+template <class CharT> struct nsBufferRoutines {};
+
+template <>
+struct nsBufferRoutines<char>
+{
+ static
+ int32_t compare( const char* a, const char* b, uint32_t max, bool ic )
+ {
+ return Compare1To1(a, b, max, ic);
+ }
+
+ static
+ int32_t compare( const char* a, const char16_t* b, uint32_t max, bool ic )
+ {
+ return Compare1To2(a, b, max, ic);
+ }
+
+ static
+ int32_t find_char( const char* s, uint32_t max, int32_t offset, const char16_t c, int32_t count )
+ {
+ return FindChar1(s, max, offset, c, count);
+ }
+
+ static
+ int32_t rfind_char( const char* s, uint32_t max, int32_t offset, const char16_t c, int32_t count )
+ {
+ return RFindChar1(s, max, offset, c, count);
+ }
+
+ static
+ char get_find_in_set_filter( const char* set )
+ {
+ return GetFindInSetFilter(set);
+ }
+
+ static
+ int32_t strip_chars( char* s, uint32_t len, const char* set )
+ {
+ return StripChars1(s, len, set);
+ }
+
+ static
+ int32_t compress_chars( char* s, uint32_t len, const char* set )
+ {
+ return CompressChars1(s, len, set);
+ }
+};
+
+template <>
+struct nsBufferRoutines<char16_t>
+{
+ static
+ int32_t compare( const char16_t* a, const char16_t* b, uint32_t max, bool ic )
+ {
+ NS_ASSERTION(!ic, "no case-insensitive compare here");
+ return Compare2To2(a, b, max);
+ }
+
+ static
+ int32_t compare( const char16_t* a, const char* b, uint32_t max, bool ic )
+ {
+ return Compare2To1(a, b, max, ic);
+ }
+
+ static
+ int32_t find_char( const char16_t* s, uint32_t max, int32_t offset, const char16_t c, int32_t count )
+ {
+ return FindChar2(s, max, offset, c, count);
+ }
+
+ static
+ int32_t rfind_char( const char16_t* s, uint32_t max, int32_t offset, const char16_t c, int32_t count )
+ {
+ return RFindChar2(s, max, offset, c, count);
+ }
+
+ static
+ char16_t get_find_in_set_filter( const char16_t* set )
+ {
+ return GetFindInSetFilter(set);
+ }
+
+ static
+ char16_t get_find_in_set_filter( const char* set )
+ {
+ return (~char16_t(0)^~char(0)) | GetFindInSetFilter(set);
+ }
+
+ static
+ int32_t strip_chars( char16_t* s, uint32_t max, const char* set )
+ {
+ return StripChars2(s, max, set);
+ }
+
+ static
+ int32_t compress_chars( char16_t* s, uint32_t len, const char* set )
+ {
+ return CompressChars2(s, len, set);
+ }
+};
+
+//-----------------------------------------------------------------------------
+
+template <class L, class R>
+#ifndef __SUNPRO_CC
+static
+#endif /* !__SUNPRO_CC */
+int32_t
+FindSubstring( const L* big, uint32_t bigLen,
+ const R* little, uint32_t littleLen,
+ bool ignoreCase )
+{
+ if (littleLen > bigLen)
+ return kNotFound;
+
+ int32_t i, max = int32_t(bigLen - littleLen);
+ for (i=0; i<=max; ++i, ++big)
+ {
+ if (nsBufferRoutines<L>::compare(big, little, littleLen, ignoreCase) == 0)
+ return i;
+ }
+
+ return kNotFound;
+}
+
+template <class L, class R>
+#ifndef __SUNPRO_CC
+static
+#endif /* !__SUNPRO_CC */
+int32_t
+RFindSubstring( const L* big, uint32_t bigLen,
+ const R* little, uint32_t littleLen,
+ bool ignoreCase )
+{
+ if (littleLen > bigLen)
+ return kNotFound;
+
+ int32_t i, max = int32_t(bigLen - littleLen);
+
+ const L* iter = big + max;
+ for (i=max; iter >= big; --i, --iter)
+ {
+ if (nsBufferRoutines<L>::compare(iter, little, littleLen, ignoreCase) == 0)
+ return i;
+ }
+
+ return kNotFound;
+}
+
+template <class CharT, class SetCharT>
+#ifndef __SUNPRO_CC
+static
+#endif /* !__SUNPRO_CC */
+int32_t
+FindCharInSet( const CharT* data, uint32_t dataLen, const SetCharT* set )
+{
+ CharT filter = nsBufferRoutines<CharT>::get_find_in_set_filter(set);
+
+ const CharT* end = data + dataLen;
+ for (const CharT* iter = data; iter < end; ++iter)
+ {
+ CharT currentChar = *iter;
+ if (currentChar & filter)
+ continue; // char is not in filter set; go on with next char.
+
+ // test all chars
+ const SetCharT* charInSet = set;
+ CharT setChar = CharT(*charInSet);
+ while (setChar)
+ {
+ if (setChar == currentChar)
+ return iter - data; // found it! return index of the found char.
+
+ setChar = CharT(*(++charInSet));
+ }
+ }
+ return kNotFound;
+}
+
+template <class CharT, class SetCharT>
+#ifndef __SUNPRO_CC
+static
+#endif /* !__SUNPRO_CC */
+int32_t
+RFindCharInSet( const CharT* data, uint32_t dataLen, const SetCharT* set )
+{
+ CharT filter = nsBufferRoutines<CharT>::get_find_in_set_filter(set);
+
+ for (const CharT* iter = data + dataLen - 1; iter >= data; --iter)
+ {
+ CharT currentChar = *iter;
+ if (currentChar & filter)
+ continue; // char is not in filter set; go on with next char.
+
+ // test all chars
+ const CharT* charInSet = set;
+ CharT setChar = *charInSet;
+ while (setChar)
+ {
+ if (setChar == currentChar)
+ return iter - data; // found it! return index of the found char.
+
+ setChar = *(++charInSet);
+ }
+ }
+ return kNotFound;
+}
+
+/**
+ * this method changes the meaning of |offset| and |count|:
+ *
+ * upon return,
+ * |offset| specifies start of search range
+ * |count| specifies length of search range
+ */
+static void
+Find_ComputeSearchRange( uint32_t bigLen, uint32_t littleLen, int32_t& offset, int32_t& count )
+{
+ // |count| specifies how many iterations to make from |offset|
+
+ if (offset < 0)
+ {
+ offset = 0;
+ }
+ else if (uint32_t(offset) > bigLen)
+ {
+ count = 0;
+ return;
+ }
+
+ int32_t maxCount = bigLen - offset;
+ if (count < 0 || count > maxCount)
+ {
+ count = maxCount;
+ }
+ else
+ {
+ count += littleLen;
+ if (count > maxCount)
+ count = maxCount;
+ }
+}
+
+/**
+ * this method changes the meaning of |offset| and |count|:
+ *
+ * upon entry,
+ * |offset| specifies the end point from which to search backwards
+ * |count| specifies the number of iterations from |offset|
+ *
+ * upon return,
+ * |offset| specifies start of search range
+ * |count| specifies length of search range
+ *
+ *
+ * EXAMPLE
+ *
+ * + -- littleLen=4 -- +
+ * : :
+ * |____|____|____|____|____|____|____|____|____|____|____|____|
+ * : :
+ * offset=5 bigLen=12
+ *
+ * if count = 4, then we expect this function to return offset = 2 and
+ * count = 7.
+ *
+ */
+static void
+RFind_ComputeSearchRange( uint32_t bigLen, uint32_t littleLen, int32_t& offset, int32_t& count )
+{
+ if (littleLen > bigLen)
+ {
+ offset = 0;
+ count = 0;
+ return;
+ }
+
+ if (offset < 0)
+ offset = bigLen - littleLen;
+ if (count < 0)
+ count = offset + 1;
+
+ int32_t start = offset - count + 1;
+ if (start < 0)
+ start = 0;
+
+ count = offset + littleLen - start;
+ offset = start;
+}
+
+//-----------------------------------------------------------------------------
+
+// define nsString obsolete methods
+#include "string-template-def-unichar.h"
+#include "nsTStringObsolete.cpp"
+#include "string-template-undef.h"
+
+// define nsCString obsolete methods
+#include "string-template-def-char.h"
+#include "nsTStringObsolete.cpp"
+#include "string-template-undef.h"
+
+//-----------------------------------------------------------------------------
+
+// specialized methods:
+
+int32_t
+nsString::Find( const nsAFlatString& aString, int32_t aOffset, int32_t aCount ) const
+{
+ // this method changes the meaning of aOffset and aCount:
+ Find_ComputeSearchRange(mLength, aString.Length(), aOffset, aCount);
+
+ int32_t result = FindSubstring(mData + aOffset, aCount, static_cast<const char16_t*>(aString.get()), aString.Length(), false);
+ if (result != kNotFound)
+ result += aOffset;
+ return result;
+}
+
+int32_t
+nsString::Find( const char16_t* aString, int32_t aOffset, int32_t aCount ) const
+{
+ return Find(nsDependentString(aString), aOffset, aCount);
+}
+
+int32_t
+nsString::RFind( const nsAFlatString& aString, int32_t aOffset, int32_t aCount ) const
+{
+ // this method changes the meaning of aOffset and aCount:
+ RFind_ComputeSearchRange(mLength, aString.Length(), aOffset, aCount);
+
+ int32_t result = RFindSubstring(mData + aOffset, aCount, static_cast<const char16_t*>(aString.get()), aString.Length(), false);
+ if (result != kNotFound)
+ result += aOffset;
+ return result;
+}
+
+int32_t
+nsString::RFind( const char16_t* aString, int32_t aOffset, int32_t aCount ) const
+{
+ return RFind(nsDependentString(aString), aOffset, aCount);
+}
+
+int32_t
+nsString::FindCharInSet( const char16_t* aSet, int32_t aOffset ) const
+{
+ if (aOffset < 0)
+ aOffset = 0;
+ else if (aOffset >= int32_t(mLength))
+ return kNotFound;
+
+ int32_t result = ::FindCharInSet(mData + aOffset, mLength - aOffset, aSet);
+ if (result != kNotFound)
+ result += aOffset;
+ return result;
+}
+
+void
+nsString::ReplaceChar( const char16_t* aSet, char16_t aNewChar )
+{
+ if (!EnsureMutable()) // XXX do this lazily?
+ AllocFailed(mLength);
+
+ char16_t* data = mData;
+ uint32_t lenRemaining = mLength;
+
+ while (lenRemaining)
+ {
+ int32_t i = ::FindCharInSet(data, lenRemaining, aSet);
+ if (i == kNotFound)
+ break;
+
+ data[i++] = aNewChar;
+ data += i;
+ lenRemaining -= i;
+ }
+}
+
+
+/**
+ * nsTString::Compare,CompareWithConversion,etc.
+ */
+
+int32_t
+nsCString::Compare( const char* aString, bool aIgnoreCase, int32_t aCount ) const
+{
+ uint32_t strLen = char_traits::length(aString);
+
+ int32_t maxCount = int32_t(XPCOM_MIN(mLength, strLen));
+
+ int32_t compareCount;
+ if (aCount < 0 || aCount > maxCount)
+ compareCount = maxCount;
+ else
+ compareCount = aCount;
+
+ int32_t result =
+ nsBufferRoutines<char>::compare(mData, aString, compareCount, aIgnoreCase);
+
+ if (result == 0 &&
+ (aCount < 0 || strLen < uint32_t(aCount) || mLength < uint32_t(aCount)))
+ {
+ // Since the caller didn't give us a length to test, or strings shorter
+ // than aCount, and compareCount characters matched, we have to assume
+ // that the longer string is greater.
+
+ if (mLength != strLen)
+ result = (mLength < strLen) ? -1 : 1;
+ }
+ return result;
+}
+
+bool
+nsString::EqualsIgnoreCase( const char* aString, int32_t aCount ) const
+{
+ uint32_t strLen = nsCharTraits<char>::length(aString);
+
+ int32_t maxCount = int32_t(XPCOM_MIN(mLength, strLen));
+
+ int32_t compareCount;
+ if (aCount < 0 || aCount > maxCount)
+ compareCount = maxCount;
+ else
+ compareCount = aCount;
+
+ int32_t result =
+ nsBufferRoutines<char16_t>::compare(mData, aString, compareCount, true);
+
+ if (result == 0 &&
+ (aCount < 0 || strLen < uint32_t(aCount) || mLength < uint32_t(aCount)))
+ {
+ // Since the caller didn't give us a length to test, or strings shorter
+ // than aCount, and compareCount characters matched, we have to assume
+ // that the longer string is greater.
+
+ if (mLength != strLen)
+ result = 1; // Arbitrarily using any number != 0
+ }
+ return result == 0;
+}
+
+
+/**
+ * nsTString::ToDouble
+ */
+
+double
+nsCString::ToDouble(nsresult* aErrorCode) const
+{
+ double res = 0.0;
+ if (mLength > 0)
+ {
+ char *conv_stopped;
+ const char *str = mData;
+ // Use PR_strtod, not strtod, since we don't want locale involved.
+ res = PR_strtod(str, &conv_stopped);
+ if (conv_stopped == str+mLength)
+ *aErrorCode = NS_OK;
+ else // Not all the string was scanned
+ *aErrorCode = NS_ERROR_ILLEGAL_VALUE;
+ }
+ else
+ {
+ // The string was too short (0 characters)
+ *aErrorCode = NS_ERROR_ILLEGAL_VALUE;
+ }
+ return res;
+}
+
+double
+nsString::ToDouble(nsresult* aErrorCode) const
+{
+ return NS_LossyConvertUTF16toASCII(*this).ToDouble(aErrorCode);
+}
+
+
+/**
+ * nsTString::AssignWithConversion
+ */
+
+void
+nsCString::AssignWithConversion( const nsAString& aData )
+{
+ LossyCopyUTF16toASCII(aData, *this);
+}
+
+void
+nsString::AssignWithConversion( const nsACString& aData )
+{
+ CopyASCIItoUTF16(aData, *this);
+}
+
+#endif // !MOZ_STRING_WITH_OBSOLETE_API