mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-11-01 06:35:42 +00:00
718 lines
23 KiB
C++
718 lines
23 KiB
C++
/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
|
|
/*
|
|
* The contents of this file are subject to the Netscape Public License
|
|
* Version 1.0 (the "NPL"); you may not use this file except in
|
|
* compliance with the NPL. You may obtain a copy of the NPL at
|
|
* http://www.mozilla.org/NPL/
|
|
*
|
|
* Software distributed under the NPL is distributed on an "AS IS" basis,
|
|
* WITHOUT WARRANTY OF ANY KIND, either express or implied. See the NPL
|
|
* for the specific language governing rights and limitations under the
|
|
* NPL.
|
|
*
|
|
* The Initial Developer of this code under the NPL is Netscape
|
|
* Communications Corporation. Portions created by Netscape are
|
|
* Copyright (C) 1998 Netscape Communications Corporation. All Rights
|
|
* Reserved.
|
|
*/
|
|
|
|
/******************************************************************************************
|
|
MODULE NOTES:
|
|
|
|
This file contains the nsStr data structure.
|
|
This general purpose buffer management class is used as the basis for our strings.
|
|
It's benefits include:
|
|
1. An efficient set of library style functions for manipulating nsStrs
|
|
2. Support for 1 and 2 byte character strings (which can easily be increased to n)
|
|
3. Unicode awareness and interoperability.
|
|
|
|
*******************************************************************************************/
|
|
|
|
#include "nsStr.h"
|
|
#include "bufferRoutines.h"
|
|
#include "stdio.h" //only used for printf
|
|
#include "nsCRT.h"
|
|
#include "nsDeque.h"
|
|
|
|
|
|
//static const char* kCallFindChar = "For better performance, call FindChar() for targets whose length==1.";
|
|
//static const char* kCallRFindChar = "For better performance, call RFindChar() for targets whose length==1.";
|
|
|
|
static const PRUnichar gCommonEmptyBuffer[1] = {0};
|
|
|
|
/**
|
|
* This method initializes all the members of the nsStr structure
|
|
*
|
|
* @update gess10/30/98
|
|
* @param
|
|
* @return
|
|
*/
|
|
void nsStr::Initialize(nsStr& aDest,eCharSize aCharSize) {
|
|
aDest.mStr=(char*)gCommonEmptyBuffer;
|
|
aDest.mLength=0;
|
|
aDest.mCapacity=0;
|
|
aDest.mCharSize=aCharSize;
|
|
aDest.mOwnsBuffer=0;
|
|
}
|
|
|
|
/**
|
|
* This method initializes all the members of the nsStr structure
|
|
* @update gess10/30/98
|
|
* @param
|
|
* @return
|
|
*/
|
|
void nsStr::Initialize(nsStr& aDest,char* aCString,PRUint32 aCapacity,PRUint32 aLength,eCharSize aCharSize,PRBool aOwnsBuffer){
|
|
aDest.mStr=(aCString) ? aCString : (char*)gCommonEmptyBuffer;
|
|
aDest.mLength=aLength;
|
|
aDest.mCapacity=aCapacity;
|
|
aDest.mCharSize=aCharSize;
|
|
aDest.mOwnsBuffer=aOwnsBuffer;
|
|
}
|
|
|
|
|
|
/**
|
|
* This member destroys the memory buffer owned by an nsStr object (if it actually owns it)
|
|
* @update gess10/30/98
|
|
* @param
|
|
* @return
|
|
*/
|
|
void nsStr::Destroy(nsStr& aDest) {
|
|
if((aDest.mStr) && (aDest.mStr!=(char*)gCommonEmptyBuffer)) {
|
|
Free(aDest);
|
|
}
|
|
}
|
|
|
|
|
|
/**
|
|
* This method gets called when the internal buffer needs
|
|
* to grow to a given size. The original contents are not preserved.
|
|
* @update gess 3/30/98
|
|
* @param aNewLength -- new capacity of string in charSize units
|
|
* @return void
|
|
*/
|
|
PRBool nsStr::EnsureCapacity(nsStr& aString,PRUint32 aNewLength) {
|
|
PRBool result=PR_TRUE;
|
|
if(aNewLength>aString.mCapacity) {
|
|
result=Realloc(aString,aNewLength);
|
|
if(aString.mStr)
|
|
AddNullTerminator(aString);
|
|
}
|
|
return result;
|
|
}
|
|
|
|
/**
|
|
* This method gets called when the internal buffer needs
|
|
* to grow to a given size. The original contents ARE preserved.
|
|
* @update gess 3/30/98
|
|
* @param aNewLength -- new capacity of string in charSize units
|
|
* @return void
|
|
*/
|
|
PRBool nsStr::GrowCapacity(nsStr& aDest,PRUint32 aNewLength) {
|
|
PRBool result=PR_TRUE;
|
|
if(aNewLength>aDest.mCapacity) {
|
|
nsStr theTempStr;
|
|
nsStr::Initialize(theTempStr,aDest.mCharSize);
|
|
|
|
result=EnsureCapacity(theTempStr,aNewLength);
|
|
if(result) {
|
|
if(aDest.mLength) {
|
|
Append(theTempStr,aDest,0,aDest.mLength);
|
|
}
|
|
Free(aDest);
|
|
aDest.mStr = theTempStr.mStr;
|
|
theTempStr.mStr=0; //make sure to null this out so that you don't lose the buffer you just stole...
|
|
aDest.mLength=theTempStr.mLength;
|
|
aDest.mCapacity=theTempStr.mCapacity;
|
|
aDest.mOwnsBuffer=theTempStr.mOwnsBuffer;
|
|
}
|
|
}
|
|
return result;
|
|
}
|
|
|
|
/**
|
|
* Replaces the contents of aDest with aSource, up to aCount of chars.
|
|
* @update gess10/30/98
|
|
* @param aDest is the nsStr that gets changed.
|
|
* @param aSource is where chars are copied from
|
|
* @param aCount is the number of chars copied from aSource
|
|
*/
|
|
void nsStr::Assign(nsStr& aDest,const nsStr& aSource,PRUint32 anOffset,PRInt32 aCount){
|
|
if(&aDest!=&aSource){
|
|
Truncate(aDest,0);
|
|
Append(aDest,aSource,anOffset,aCount);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* This method appends the given nsStr to this one. Note that we have to
|
|
* pay attention to the underlying char-size of both structs.
|
|
* @update gess10/30/98
|
|
* @param aDest is the nsStr to be manipulated
|
|
* @param aSource is where char are copied from
|
|
* @aCount is the number of bytes to be copied
|
|
*/
|
|
void nsStr::Append(nsStr& aDest,const nsStr& aSource,PRUint32 anOffset,PRInt32 aCount){
|
|
if(anOffset<aSource.mLength){
|
|
PRUint32 theRealLen=(aCount<0) ? aSource.mLength : MinInt(aCount,aSource.mLength);
|
|
PRUint32 theLength=(anOffset+theRealLen<aSource.mLength) ? theRealLen : (aSource.mLength-anOffset);
|
|
if(0<theLength){
|
|
|
|
PRBool isBigEnough=PR_TRUE;
|
|
if(aDest.mLength+theLength > aDest.mCapacity) {
|
|
isBigEnough=GrowCapacity(aDest,aDest.mLength+theLength);
|
|
}
|
|
|
|
if(isBigEnough) {
|
|
//now append new chars, starting at offset
|
|
(*gCopyChars[aSource.mCharSize][aDest.mCharSize])(aDest.mStr,aDest.mLength,aSource.mStr,anOffset,theLength);
|
|
|
|
aDest.mLength+=theLength;
|
|
AddNullTerminator(aDest);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
/**
|
|
* This method inserts up to "aCount" chars from a source nsStr into a dest nsStr.
|
|
* @update gess10/30/98
|
|
* @param aDest is the nsStr that gets changed
|
|
* @param aDestOffset is where in aDest the insertion is to occur
|
|
* @param aSource is where chars are copied from
|
|
* @param aSrcOffset is where in aSource chars are copied from
|
|
* @param aCount is the number of chars from aSource to be inserted into aDest
|
|
*/
|
|
void nsStr::Insert( nsStr& aDest,PRUint32 aDestOffset,const nsStr& aSource,PRUint32 aSrcOffset,PRInt32 aCount){
|
|
//there are a few cases for insert:
|
|
// 1. You're inserting chars into an empty string (assign)
|
|
// 2. You're inserting onto the end of a string (append)
|
|
// 3. You're inserting onto the 1..n-1 pos of a string (the hard case).
|
|
if(0<aSource.mLength){
|
|
if(aDest.mLength){
|
|
if(aDestOffset<aDest.mLength){
|
|
PRInt32 theRealLen=(aCount<0) ? aSource.mLength : MinInt(aCount,aSource.mLength);
|
|
PRInt32 theLength=(aSrcOffset+theRealLen<aSource.mLength) ? theRealLen : (aSource.mLength-aSrcOffset);
|
|
|
|
if(aSrcOffset<aSource.mLength) {
|
|
//here's the only new case we have to handle.
|
|
//chars are really being inserted into our buffer...
|
|
|
|
if(aDest.mLength+theLength > aDest.mCapacity) {
|
|
nsStr theTempStr;
|
|
nsStr::Initialize(theTempStr,aDest.mCharSize);
|
|
|
|
PRBool isBigEnough=EnsureCapacity(theTempStr,aDest.mLength+theLength); //grow the temp buffer to the right size
|
|
|
|
if(isBigEnough) {
|
|
if(aDestOffset) {
|
|
Append(theTempStr,aDest,0,aDestOffset); //first copy leftmost data...
|
|
}
|
|
|
|
Append(theTempStr,aSource,0,aSource.mLength); //next copy inserted (new) data
|
|
|
|
PRUint32 theRemains=aDest.mLength-aDestOffset;
|
|
if(theRemains) {
|
|
Append(theTempStr,aDest,aDestOffset,theRemains); //next copy rightmost data
|
|
}
|
|
|
|
Free(aDest);
|
|
aDest.mStr = theTempStr.mStr;
|
|
theTempStr.mStr=0; //make sure to null this out so that you don't lose the buffer you just stole...
|
|
aDest.mCapacity=theTempStr.mCapacity;
|
|
aDest.mOwnsBuffer=theTempStr.mOwnsBuffer;
|
|
}
|
|
|
|
}
|
|
|
|
else {
|
|
//shift the chars right by theDelta...
|
|
(*gShiftChars[aDest.mCharSize][KSHIFTRIGHT])(aDest.mStr,aDest.mLength,aDestOffset,theLength);
|
|
|
|
//now insert new chars, starting at offset
|
|
(*gCopyChars[aSource.mCharSize][aDest.mCharSize])(aDest.mStr,aDestOffset,aSource.mStr,aSrcOffset,theLength);
|
|
}
|
|
|
|
//finally, make sure to update the string length...
|
|
aDest.mLength+=theLength;
|
|
AddNullTerminator(aDest);
|
|
|
|
}//if
|
|
//else nothing to do!
|
|
}
|
|
else Append(aDest,aSource,0,aCount);
|
|
}
|
|
else Append(aDest,aSource,0,aCount);
|
|
}
|
|
}
|
|
|
|
|
|
/**
|
|
* This method deletes up to aCount chars from aDest
|
|
* @update gess10/30/98
|
|
* @param aDest is the nsStr to be manipulated
|
|
* @param aDestOffset is where in aDest deletion is to occur
|
|
* @param aCount is the number of chars to be deleted in aDest
|
|
*/
|
|
void nsStr::Delete(nsStr& aDest,PRUint32 aDestOffset,PRUint32 aCount){
|
|
if(aDestOffset<aDest.mLength){
|
|
|
|
PRUint32 theDelta=aDest.mLength-aDestOffset;
|
|
PRUint32 theLength=(theDelta<aCount) ? theDelta : aCount;
|
|
|
|
if(aDestOffset+theLength<aDest.mLength) {
|
|
|
|
//if you're here, it means we're cutting chars out of the middle of the string...
|
|
//so shift the chars left by theLength...
|
|
(*gShiftChars[aDest.mCharSize][KSHIFTLEFT])(aDest.mStr,aDest.mLength,aDestOffset,theLength);
|
|
aDest.mLength-=theLength;
|
|
AddNullTerminator(aDest);
|
|
}
|
|
else Truncate(aDest,aDestOffset);
|
|
}//if
|
|
}
|
|
|
|
/**
|
|
* This method truncates the given nsStr at given offset
|
|
* @update gess10/30/98
|
|
* @param aDest is the nsStr to be truncated
|
|
* @param aDestOffset is where in aDest truncation is to occur
|
|
*/
|
|
void nsStr::Truncate(nsStr& aDest,PRUint32 aDestOffset){
|
|
if(aDestOffset<aDest.mLength){
|
|
aDest.mLength=aDestOffset;
|
|
AddNullTerminator(aDest);
|
|
}
|
|
}
|
|
|
|
|
|
/**
|
|
* This method forces the given string to upper or lowercase
|
|
* @update gess1/7/99
|
|
* @param aDest is the string you're going to change
|
|
* @param aToUpper: if TRUE, then we go uppercase, otherwise we go lowercase
|
|
* @return
|
|
*/
|
|
void nsStr::ChangeCase(nsStr& aDest,PRBool aToUpper) {
|
|
// somehow UnicharUtil return failed, fallback to the old ascii only code
|
|
gCaseConverters[aDest.mCharSize](aDest.mStr,aDest.mLength,aToUpper);
|
|
}
|
|
|
|
/**
|
|
*
|
|
* @update gess1/7/99
|
|
* @param
|
|
* @return
|
|
*/
|
|
void nsStr::Trim(nsStr& aDest,const char* aSet,PRBool aEliminateLeading,PRBool aEliminateTrailing){
|
|
|
|
if((aDest.mLength>0) && aSet){
|
|
PRInt32 theIndex=-1;
|
|
PRInt32 theMax=aDest.mLength;
|
|
PRInt32 theSetLen=nsCRT::strlen(aSet);
|
|
|
|
if(aEliminateLeading) {
|
|
while(++theIndex<=theMax) {
|
|
PRUnichar theChar=GetCharAt(aDest,theIndex);
|
|
PRInt32 thePos=gFindChars[eOneByte](aSet,theSetLen,0,theChar,PR_FALSE);
|
|
if(kNotFound==thePos)
|
|
break;
|
|
}
|
|
if(0<theIndex) {
|
|
if(theIndex<theMax) {
|
|
Delete(aDest,0,theIndex);
|
|
}
|
|
else Truncate(aDest,0);
|
|
}
|
|
}
|
|
|
|
if(aEliminateTrailing) {
|
|
theIndex=aDest.mLength;
|
|
PRInt32 theNewLen=theIndex;
|
|
while(--theIndex>0) {
|
|
PRUnichar theChar=GetCharAt(aDest,theIndex); //read at end now...
|
|
PRInt32 thePos=gFindChars[eOneByte](aSet,theSetLen,0,theChar,PR_FALSE);
|
|
if(kNotFound<thePos)
|
|
theNewLen=theIndex;
|
|
else break;
|
|
}
|
|
if(theNewLen<theMax) {
|
|
Truncate(aDest,theNewLen);
|
|
}
|
|
}
|
|
|
|
}
|
|
}
|
|
|
|
/**
|
|
*
|
|
* @update gess1/7/99
|
|
* @param
|
|
* @return
|
|
*/
|
|
void nsStr::CompressSet(nsStr& aDest,const char* aSet,PRBool aEliminateLeading,PRBool aEliminateTrailing){
|
|
Trim(aDest,aSet,aEliminateLeading,aEliminateTrailing);
|
|
PRUint32 aNewLen=gCompressChars[aDest.mCharSize](aDest.mStr,aDest.mLength,aSet);
|
|
aDest.mLength=aNewLen;
|
|
}
|
|
|
|
|
|
/**
|
|
*
|
|
* @update gess1/7/99
|
|
* @param
|
|
* @return
|
|
*/
|
|
void nsStr::StripChars(nsStr& aDest,const char* aSet){
|
|
if((0<aDest.mLength) && (aSet)) {
|
|
PRUint32 aNewLen=gStripChars[aDest.mCharSize](aDest.mStr,aDest.mLength,aSet);
|
|
aDest.mLength=aNewLen;
|
|
}
|
|
}
|
|
|
|
|
|
/**************************************************************
|
|
Searching methods...
|
|
**************************************************************/
|
|
|
|
|
|
/**
|
|
* This searches aDest for a given substring
|
|
*
|
|
* @update gess 3/25/98
|
|
* @param aDest string to search
|
|
* @param aTarget is the substring you're trying to find.
|
|
* @param aIgnorecase indicates case sensitivity of search
|
|
* @param anOffset tells us where to start the search
|
|
* @return index in aDest where member of aSet occurs, or -1 if not found
|
|
*/
|
|
PRInt32 nsStr::FindSubstr(const nsStr& aDest,const nsStr& aTarget, PRBool aIgnoreCase,PRInt32 anOffset) {
|
|
// NS_PRECONDITION(aTarget.mLength!=1,kCallFindChar);
|
|
|
|
PRInt32 result=kNotFound;
|
|
|
|
if((0<aDest.mLength) && (anOffset<(PRInt32)aDest.mLength)) {
|
|
PRInt32 theMax=aDest.mLength-aTarget.mLength;
|
|
PRInt32 index=(0<=anOffset) ? anOffset : 0;
|
|
|
|
if((aDest.mLength>=aTarget.mLength) && (aTarget.mLength>0) && (index>=0)){
|
|
PRInt32 theTargetMax=aTarget.mLength;
|
|
while(index<=theMax) {
|
|
PRInt32 theSubIndex=-1;
|
|
PRBool matches=PR_TRUE;
|
|
while((++theSubIndex<theTargetMax) && (matches)){
|
|
PRUnichar theChar=(aIgnoreCase) ? nsCRT::ToLower(GetCharAt(aDest,index+theSubIndex)) : GetCharAt(aDest,index+theSubIndex);
|
|
PRUnichar theTargetChar=(aIgnoreCase) ? nsCRT::ToLower(GetCharAt(aTarget,theSubIndex)) : GetCharAt(aTarget,theSubIndex);
|
|
matches=PRBool(theChar==theTargetChar);
|
|
}
|
|
if(matches) {
|
|
result=index;
|
|
break;
|
|
}
|
|
index++;
|
|
} //while
|
|
}//if
|
|
}//if
|
|
return result;
|
|
}
|
|
|
|
|
|
/**
|
|
* This searches aDest for a given character
|
|
*
|
|
* @update gess 3/25/98
|
|
* @param aDest string to search
|
|
* @param char is the character you're trying to find.
|
|
* @param aIgnorecase indicates case sensitivity of search
|
|
* @param anOffset tells us where to start the search
|
|
* @return index in aDest where member of aSet occurs, or -1 if not found
|
|
*/
|
|
PRInt32 nsStr::FindChar(const nsStr& aDest,PRUnichar aChar, PRBool aIgnoreCase,PRInt32 anOffset) {
|
|
PRInt32 result=kNotFound;
|
|
if((0<aDest.mLength) && (anOffset<(PRInt32)aDest.mLength)) {
|
|
PRUint32 index=(0<=anOffset) ? (PRUint32)anOffset : 0;
|
|
result=gFindChars[aDest.mCharSize](aDest.mStr,aDest.mLength,index,aChar,aIgnoreCase);
|
|
}
|
|
return result;
|
|
}
|
|
|
|
|
|
/**
|
|
* This searches aDest for a character found in aSet.
|
|
*
|
|
* @update gess 3/25/98
|
|
* @param aDest string to search
|
|
* @param aSet contains a list of chars to be searched for
|
|
* @param aIgnorecase indicates case sensitivity of search
|
|
* @param anOffset tells us where to start the search
|
|
* @return index in aDest where member of aSet occurs, or -1 if not found
|
|
*/
|
|
PRInt32 nsStr::FindCharInSet(const nsStr& aDest,const nsStr& aSet,PRBool aIgnoreCase,PRInt32 anOffset) {
|
|
//NS_PRECONDITION(aSet.mLength!=1,kCallFindChar);
|
|
|
|
PRInt32 index=(0<=anOffset) ? anOffset-1 : -1;
|
|
PRInt32 thePos;
|
|
|
|
//Note that the search is inverted here. We're scanning aDest, one char at a time
|
|
//but doing the search against the given set. That's why we use 0 as the offset below.
|
|
if((0<aDest.mLength) && (0<aSet.mLength)){
|
|
while(++index<(PRInt32)aDest.mLength) {
|
|
PRUnichar theChar=GetCharAt(aDest,index);
|
|
thePos=gFindChars[aSet.mCharSize](aSet.mStr,aSet.mLength,0,theChar,aIgnoreCase);
|
|
if(kNotFound!=thePos)
|
|
return index;
|
|
} //while
|
|
}
|
|
return kNotFound;
|
|
}
|
|
|
|
/**************************************************************
|
|
Reverse Searching methods...
|
|
**************************************************************/
|
|
|
|
|
|
/**
|
|
* This searches aDest (in reverse) for a given substring
|
|
*
|
|
* @update gess 3/25/98
|
|
* @param aDest string to search
|
|
* @param aTarget is the substring you're trying to find.
|
|
* @param aIgnorecase indicates case sensitivity of search
|
|
* @param anOffset tells us where to start the search (counting from left)
|
|
* @return index in aDest where member of aSet occurs, or -1 if not found
|
|
*/
|
|
PRInt32 nsStr::RFindSubstr(const nsStr& aDest,const nsStr& aTarget, PRBool aIgnoreCase,PRInt32 anOffset) {
|
|
//NS_PRECONDITION(aTarget.mLength!=1,kCallRFindChar);
|
|
|
|
PRInt32 result=kNotFound;
|
|
|
|
if((0<aDest.mLength) && (anOffset<(PRInt32)aDest.mLength)) {
|
|
PRInt32 index=(0<=anOffset) ? anOffset : aDest.mLength-1;
|
|
|
|
if((aDest.mLength>=aTarget.mLength) && (aTarget.mLength>0) && (index>=0)){
|
|
|
|
nsStr theCopy;
|
|
nsStr::Initialize(theCopy,eOneByte);
|
|
nsStr::Assign(theCopy,aTarget,0,aTarget.mLength);
|
|
if(aIgnoreCase){
|
|
nsStr::ChangeCase(theCopy,PR_FALSE); //force to lowercase
|
|
}
|
|
|
|
PRInt32 theTargetMax=theCopy.mLength;
|
|
while(index>=0) {
|
|
PRInt32 theSubIndex=-1;
|
|
PRBool matches=PR_FALSE;
|
|
if(index+theCopy.mLength<=aDest.mLength) {
|
|
matches=PR_TRUE;
|
|
while((++theSubIndex<theTargetMax) && (matches)){
|
|
PRUnichar theDestChar=(aIgnoreCase) ? nsCRT::ToLower(GetCharAt(aDest,index+theSubIndex)) : GetCharAt(aDest,index+theSubIndex);
|
|
PRUnichar theTargetChar=GetCharAt(theCopy,theSubIndex);
|
|
matches=PRBool(theDestChar==theTargetChar);
|
|
} //while
|
|
} //if
|
|
if(matches) {
|
|
result=index;
|
|
break;
|
|
}
|
|
index--;
|
|
} //while
|
|
nsStr::Destroy(theCopy);
|
|
}//if
|
|
}//if
|
|
return result;
|
|
}
|
|
|
|
|
|
/**
|
|
* This searches aDest (in reverse) for a given character
|
|
*
|
|
* @update gess 3/25/98
|
|
* @param aDest string to search
|
|
* @param char is the character you're trying to find.
|
|
* @param aIgnorecase indicates case sensitivity of search
|
|
* @param anOffset tells us where to start the search
|
|
* @return index in aDest where member of aSet occurs, or -1 if not found
|
|
*/
|
|
PRInt32 nsStr::RFindChar(const nsStr& aDest,PRUnichar aChar, PRBool aIgnoreCase,PRInt32 anOffset) {
|
|
PRInt32 result=kNotFound;
|
|
if((0<aDest.mLength) && (anOffset<(PRInt32)aDest.mLength)) {
|
|
PRUint32 index=(0<=anOffset) ? anOffset : aDest.mLength-1;
|
|
result=gRFindChars[aDest.mCharSize](aDest.mStr,aDest.mLength,index,aChar,aIgnoreCase);
|
|
}
|
|
return result;
|
|
}
|
|
|
|
/**
|
|
* This searches aDest (in reverese) for a character found in aSet.
|
|
*
|
|
* @update gess 3/25/98
|
|
* @param aDest string to search
|
|
* @param aSet contains a list of chars to be searched for
|
|
* @param aIgnorecase indicates case sensitivity of search
|
|
* @param anOffset tells us where to start the search
|
|
* @return index in aDest where member of aSet occurs, or -1 if not found
|
|
*/
|
|
PRInt32 nsStr::RFindCharInSet(const nsStr& aDest,const nsStr& aSet,PRBool aIgnoreCase,PRInt32 anOffset) {
|
|
//NS_PRECONDITION(aSet.mLength!=1,kCallRFindChar);
|
|
|
|
PRInt32 index=(0<=anOffset) ? anOffset : aDest.mLength;
|
|
PRInt32 thePos;
|
|
|
|
//note that the search is inverted here. We're scanning aDest, one char at a time
|
|
//but doing the search against the given set. That's why we use 0 as the offset below.
|
|
if(0<aDest.mLength) {
|
|
while(--index>=0) {
|
|
PRUnichar theChar=GetCharAt(aDest,index);
|
|
thePos=gFindChars[aSet.mCharSize](aSet.mStr,aSet.mLength,0,theChar,aIgnoreCase);
|
|
if(kNotFound!=thePos)
|
|
return index;
|
|
} //while
|
|
}
|
|
return kNotFound;
|
|
}
|
|
|
|
|
|
/**
|
|
* Compare source and dest strings, up to an (optional max) number of chars
|
|
* @param aDest is the first str to compare
|
|
* @param aSource is the second str to compare
|
|
* @param aCount -- if (-1), then we use length of longer string; if (0<aCount) then it gives the max # of chars to compare
|
|
* @param aIgnorecase tells us whether to search with case sensitivity
|
|
* @return aDest<aSource=-1;aDest==aSource==0;aDest>aSource=1
|
|
*/
|
|
PRInt32 nsStr::Compare(const nsStr& aDest,const nsStr& aSource,PRInt32 aCount,PRBool aIgnoreCase) {
|
|
PRInt32 result=0;
|
|
|
|
if(aCount) {
|
|
PRInt32 minlen=(aSource.mLength<aDest.mLength) ? aSource.mLength : aDest.mLength;
|
|
|
|
if(0==minlen) {
|
|
if ((aDest.mLength == 0) && (aSource.mLength == 0))
|
|
return 0;
|
|
if (aDest.mLength == 0)
|
|
return -1;
|
|
return 1;
|
|
}
|
|
|
|
PRInt32 maxlen=(aSource.mLength<aDest.mLength) ? aDest.mLength : aSource.mLength;
|
|
aCount = (aCount<0) ? maxlen : MinInt(aCount,maxlen);
|
|
result=(*gCompare[aDest.mCharSize][aSource.mCharSize])(aDest.mStr,aSource.mStr,aCount,aIgnoreCase);
|
|
}
|
|
return result;
|
|
}
|
|
|
|
//----------------------------------------------------------------------------------------
|
|
|
|
PRBool nsStr::Alloc(nsStr& aDest,PRUint32 aCount) {
|
|
|
|
static int mAllocCount=0;
|
|
mAllocCount++;
|
|
|
|
//we're given the acount value in charunits; now scale up to next multiple.
|
|
PRUint32 theNewCapacity=kDefaultStringSize;
|
|
while(theNewCapacity<aCount){
|
|
theNewCapacity<<=1;
|
|
}
|
|
|
|
aDest.mCapacity=theNewCapacity++;
|
|
PRUint32 theSize=(theNewCapacity<<aDest.mCharSize);
|
|
aDest.mStr = (char*)nsAllocator::Alloc(theSize);
|
|
|
|
PRBool result=PR_FALSE;
|
|
if(aDest.mStr) {
|
|
aDest.mOwnsBuffer=1;
|
|
result=PR_TRUE;
|
|
}
|
|
return result;
|
|
}
|
|
|
|
PRBool nsStr::Free(nsStr& aDest){
|
|
if(aDest.mStr){
|
|
if(aDest.mOwnsBuffer){
|
|
nsAllocator::Free(aDest.mStr);
|
|
}
|
|
aDest.mStr=0;
|
|
aDest.mOwnsBuffer=0;
|
|
return PR_TRUE;
|
|
}
|
|
return PR_FALSE;
|
|
}
|
|
|
|
PRBool nsStr::Realloc(nsStr& aDest,PRUint32 aCount){
|
|
|
|
nsStr temp;
|
|
memcpy(&temp,&aDest,sizeof(aDest));
|
|
|
|
PRBool result=Alloc(temp,aCount);
|
|
if(result) {
|
|
Free(aDest);
|
|
aDest.mStr=temp.mStr;
|
|
aDest.mCapacity=temp.mCapacity;
|
|
aDest.mOwnsBuffer=temp.mOwnsBuffer;
|
|
}
|
|
return result;
|
|
}
|
|
|
|
//----------------------------------------------------------------------------------------
|
|
|
|
CBufDescriptor::CBufDescriptor(char* aString,PRBool aStackBased,PRUint32 aCapacity,PRInt32 aLength) {
|
|
mBuffer=aString;
|
|
mCharSize=eOneByte;
|
|
mStackBased=aStackBased;
|
|
mIsConst=PR_FALSE;
|
|
mLength=mCapacity=0;
|
|
if(aString && aCapacity>1) {
|
|
mCapacity=aCapacity-1;
|
|
mLength=(-1==aLength) ? strlen(aString) : aLength;
|
|
if(mLength>PRInt32(mCapacity))
|
|
mLength=mCapacity;
|
|
}
|
|
}
|
|
|
|
CBufDescriptor::CBufDescriptor(const char* aString,PRBool aStackBased,PRUint32 aCapacity,PRInt32 aLength) {
|
|
mBuffer=(char*)aString;
|
|
mCharSize=eOneByte;
|
|
mStackBased=aStackBased;
|
|
mIsConst=PR_TRUE;
|
|
mLength=mCapacity=0;
|
|
if(aString && aCapacity>1) {
|
|
mCapacity=aCapacity-1;
|
|
mLength=(-1==aLength) ? strlen(aString) : aLength;
|
|
if(mLength>PRInt32(mCapacity))
|
|
mLength=mCapacity;
|
|
}
|
|
}
|
|
|
|
|
|
CBufDescriptor::CBufDescriptor(PRUnichar* aString,PRBool aStackBased,PRUint32 aCapacity,PRInt32 aLength) {
|
|
mBuffer=(char*)aString;
|
|
mCharSize=eTwoByte;
|
|
mStackBased=aStackBased;
|
|
mLength=mCapacity=0;
|
|
mIsConst=PR_FALSE;
|
|
if(aString && aCapacity>1) {
|
|
mCapacity=aCapacity-1;
|
|
mLength=(-1==aLength) ? nsCRT::strlen(aString) : aLength;
|
|
if(mLength>PRInt32(mCapacity))
|
|
mLength=mCapacity;
|
|
}
|
|
}
|
|
|
|
CBufDescriptor::CBufDescriptor(const PRUnichar* aString,PRBool aStackBased,PRUint32 aCapacity,PRInt32 aLength) {
|
|
mBuffer=(char*)aString;
|
|
mCharSize=eTwoByte;
|
|
mStackBased=aStackBased;
|
|
mLength=mCapacity=0;
|
|
mIsConst=PR_TRUE;
|
|
if(aString && aCapacity>1) {
|
|
mCapacity=aCapacity-1;
|
|
mLength=(-1==aLength) ? nsCRT::strlen(aString) : aLength;
|
|
if(mLength>PRInt32(mCapacity))
|
|
mLength=mCapacity;
|
|
}
|
|
}
|
|
|
|
//----------------------------------------------------------------------------------------
|
|
|
|
|