3141 lines
127 KiB
C#
3141 lines
127 KiB
C#
// ==++==
|
||
//
|
||
// Copyright (c) Microsoft Corporation. All rights reserved.
|
||
//
|
||
// ==--==
|
||
/*============================================================
|
||
**
|
||
** Class: String
|
||
**
|
||
**
|
||
** Purpose: Contains headers for the String class. Actual implementations
|
||
** are in String.cpp
|
||
**
|
||
** Date: March 15, 1998
|
||
**
|
||
===========================================================*/
|
||
namespace System {
|
||
using System.Text;
|
||
using System;
|
||
using System.Diagnostics;
|
||
using System.Globalization;
|
||
using System.Threading;
|
||
using System.Collections;
|
||
using System.Runtime.CompilerServices;
|
||
using System.Runtime.InteropServices;
|
||
using Microsoft.Bartok.Runtime;
|
||
|
||
//
|
||
// For Information on these methods, please see COMString.cpp
|
||
//
|
||
// The String class represents a static string of characters. Many of
|
||
// the String methods perform some type of transformation on the current
|
||
// instance and return the result as a new String. All comparison methods are
|
||
// implemented as a part of String. As with arrays, character positions
|
||
// (indices) are zero-based.
|
||
//
|
||
//| <include path='docs/doc[@for="String"]/*' />
|
||
[CCtorIsRunDuringStartup]
|
||
public sealed class String : IComparable, ICloneable, IEnumerable {
|
||
|
||
//
|
||
//NOTE NOTE NOTE NOTE
|
||
//These fields map directly onto the fields in an EE StringObject. See object.h for the layout.
|
||
//Don't change these fields or add any new fields without first talking to JRoxe.
|
||
//
|
||
[RequiredByBartok]
|
||
private int m_arrayLength;
|
||
[AccessedByRuntime("referenced from c++")]
|
||
private int m_stringLength;
|
||
[AccessedByRuntime("referenced from c++")]
|
||
private char m_firstChar;
|
||
|
||
//private static readonly char FmtMsgMarkerChar='%';
|
||
//private static readonly char FmtMsgFmtCodeChar='!';
|
||
//These are defined in Com99/src/vm/COMStringCommon.h and must be kept in sync.
|
||
private const int TrimHead = 0;
|
||
private const int TrimTail = 1;
|
||
private const int TrimBoth = 2;
|
||
|
||
// The Empty constant holds the empty string value.
|
||
//We need to call the String constructor so that the compiler doesn't mark this as a literal.
|
||
//Marking this as a literal would mean that it doesn't show up as a field which we can access
|
||
//from native.
|
||
//| <include path='docs/doc[@for="String.Empty"]/*' />
|
||
public static readonly String Empty = "";
|
||
|
||
//
|
||
//Native Static Methods
|
||
//
|
||
|
||
// Joins an array of strings together as one string with a separator between each original string.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Join"]/*' />
|
||
public static String Join(String separator, String[] value) {
|
||
if (value==null) {
|
||
throw new ArgumentNullException("value");
|
||
}
|
||
return Join(separator, value, 0, value.Length);
|
||
}
|
||
|
||
/**
|
||
* Joins an array of strings together as one string with a separator
|
||
* between each original string.
|
||
*
|
||
* @param separator the string used as a separator.
|
||
* @param value the array of strings to be joined.
|
||
* @param startIndex the position within the array to start using
|
||
* strings.
|
||
* @param count the number of strings to take from the array.
|
||
*
|
||
* @return a string consisting of the strings contained in <i>value</i>
|
||
* from <EM>startIndex</EM> to <EM>startIndex</EM> + <EM>count</EM>
|
||
* joined together to form a single string, with each of the original
|
||
* strings separated by <i>separator</i>.
|
||
*/
|
||
//| <include path='docs/doc[@for="String.Join1"]/*' />
|
||
public static String Join(String separator, String[] value,
|
||
int startIndex, int count)
|
||
{
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::JoinArray
|
||
if (separator == null) {
|
||
separator = String.Empty;
|
||
}
|
||
if (value == null) {
|
||
throw new ArgumentNullException("value array is null");
|
||
}
|
||
if (startIndex < 0) {
|
||
throw new ArgumentOutOfRangeException("startIndex is negative");
|
||
}
|
||
if (count < 0) {
|
||
throw new ArgumentOutOfRangeException("count is negative");
|
||
}
|
||
if (startIndex + count > value.Length) {
|
||
throw new ArgumentOutOfRangeException("startIndex+count>value.Length");
|
||
}
|
||
// Special case the empty list of strings
|
||
if (count == 0) {
|
||
return String.Empty;
|
||
}
|
||
// Compute the length of the new string
|
||
int newLength = 0;
|
||
int limit = startIndex + count;
|
||
for (int i = startIndex; i < limit; i++) {
|
||
String s = value[i];
|
||
if (s != null) {
|
||
newLength += s.m_stringLength;
|
||
}
|
||
}
|
||
newLength += (count - 1) * separator.m_stringLength;
|
||
// Create a new String
|
||
String result = FastAllocateString(newLength);
|
||
if (newLength == 0) {
|
||
return result;
|
||
}
|
||
// Fill in the string
|
||
int dstIndex = 0;
|
||
String firstString = value[startIndex];
|
||
if (firstString != null) {
|
||
FillString(result, 0, firstString);
|
||
dstIndex = firstString.m_stringLength;
|
||
}
|
||
for (int i = startIndex+1; i < limit; i++) {
|
||
FillString(result, dstIndex, separator);
|
||
dstIndex += separator.m_stringLength;
|
||
String elementString = value[i];
|
||
if (elementString != null) {
|
||
FillString(result, dstIndex, elementString);
|
||
dstIndex += elementString.m_stringLength;
|
||
}
|
||
}
|
||
return result;
|
||
}
|
||
|
||
private unsafe static bool CaseInsensitiveCompHelper(char * strAChars,
|
||
char * strBChars,
|
||
int aLength,
|
||
int bLength,
|
||
out int result) {
|
||
char charA;
|
||
char charB;
|
||
char *strAStart;
|
||
|
||
strAStart = strAChars;
|
||
|
||
result = 0;
|
||
|
||
// setup the pointers so that we can always increment them.
|
||
// We never access these pointers at the negative offset.
|
||
strAChars--;
|
||
strBChars--;
|
||
|
||
do {
|
||
strAChars++; strBChars++;
|
||
|
||
charA = *strAChars;
|
||
charB = *strBChars;
|
||
|
||
//Case-insensitive comparison on chars greater than 0x80 requires
|
||
//a locale-aware casing operation and we're not going there.
|
||
if (charA>=0x80 || charB>=0x80) {
|
||
//[SOSP] we should be able to fix this.
|
||
return false;
|
||
}
|
||
|
||
// Do the right thing if they differ in case only.
|
||
// We depend on the fact that the uppercase and lowercase letters
|
||
// in the range which we care about (A-Z,a-z) differ only by the
|
||
// 0x20 bit.
|
||
// The check below takes the xor of the two characters and
|
||
// determines if this bit is only set on one of them.
|
||
// If they're different cases, we know that we need to execute
|
||
// only one of the conditions within block.
|
||
if (((charA^charB)&0x20) != 0) {
|
||
if (charA>='A' && charA<='Z') {
|
||
charA |= (char)0x20;
|
||
} else if (charB>='A' && charB<='Z') {
|
||
charB |= (char)0x20;
|
||
}
|
||
}
|
||
} while (charA==charB && charA!=0);
|
||
|
||
// Return the (case-insensitive) difference between them.
|
||
if (charA!=charB) {
|
||
result = (int)(charA-charB);
|
||
return true;
|
||
}
|
||
|
||
// The length of b was unknown because it was just a pointer to a
|
||
// null-terminated string.
|
||
// If we get here, we know that both A and B are pointing at a null.
|
||
// However, A can have an embedded null. Check the number of
|
||
// characters that we've walked in A against the expected length.
|
||
if (bLength==-1) {
|
||
if ((strAChars - strAStart)!=aLength) {
|
||
result = 1;
|
||
return true;
|
||
}
|
||
result=0;
|
||
return true;
|
||
}
|
||
|
||
result = (aLength - bLength);
|
||
return true;
|
||
}
|
||
|
||
internal unsafe static int nativeCompareOrdinal(String strA, String strB,
|
||
bool bIgnoreCase)
|
||
{
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::FCCompareOrdinal
|
||
int strALength = strA.Length;
|
||
int strBLength = strB.Length;
|
||
fixed(char * strAChars = &strA.m_firstChar) {
|
||
fixed(char * strBCharsStart = &strB.m_firstChar) {
|
||
|
||
// Need copy because fixed vars are readonly
|
||
char * strBChars = strBCharsStart;
|
||
|
||
// Handle the comparison where we wish to ignore case.
|
||
if (bIgnoreCase) {
|
||
int result;
|
||
if (CaseInsensitiveCompHelper(strAChars, strBChars,
|
||
strALength, strBLength,
|
||
out result)) {
|
||
return result;
|
||
} else {
|
||
// This will happen if we have characters greater
|
||
// than 0x7F.
|
||
throw new ArgumentException();
|
||
}
|
||
}
|
||
|
||
// If the strings are the same length, compare exactly the
|
||
// right # of chars. If they are different, compare the
|
||
// shortest # + 1 (the '\0').
|
||
int count = strALength;
|
||
if (count > strBLength)
|
||
count = strBLength;
|
||
|
||
long diff = (byte*)strAChars - (byte*)strBChars;
|
||
|
||
// Loop comparing a DWORD at a time.
|
||
while ((count -= 2) >= 0) {
|
||
if(((*(int*)((byte*)strBChars + diff))
|
||
- *(int*)strBChars) != 0) {
|
||
char * ptr1 = (char*)((byte*)strBChars + diff);
|
||
char * ptr2 = strBChars;
|
||
if (*ptr1 != *ptr2) {
|
||
return ((int)*ptr1 - (int)*ptr2);
|
||
}
|
||
return ((int)*(ptr1+1) - (int)*(ptr2+1));
|
||
}
|
||
|
||
// differs from COMString.cpp because they have DWORD*
|
||
// and we use char*
|
||
strBChars += 2;
|
||
}
|
||
|
||
// Handle an extra WORD.
|
||
int c;
|
||
if (count == -1) {
|
||
c = *((char*)((byte*)strBChars+diff)) - *strBChars;
|
||
if(c != 0) {
|
||
return c;
|
||
}
|
||
}
|
||
return strALength - strBLength;
|
||
}
|
||
}
|
||
}
|
||
|
||
internal static int nativeCompareOrdinalEx(String strA, int indexA,
|
||
String strB, int indexB,
|
||
int count)
|
||
{
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::CompareOrdinalEx
|
||
throw new Exception("System.String.nativeCompareOrdinalEx not implemented in Bartok!");
|
||
}
|
||
|
||
//
|
||
// This is a helper method for the security team. They need to uppercase some strings (guaranteed to be less
|
||
// than 0x80) before security is fully initialized. Without security initialized, we can't grab resources (the nlp's)
|
||
// from the assembly. This provides a workaround for that problem and should NOT be used anywhere else.
|
||
//
|
||
internal static String SmallCharToUpper(String strA)
|
||
{
|
||
String newString = FastAllocateString(strA.Length);
|
||
nativeSmallCharToUpper(strA, newString);
|
||
return newString;
|
||
}
|
||
|
||
private static void nativeSmallCharToUpper(String strIn, String strOut)
|
||
{
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::SmallCharToUpper
|
||
throw new Exception("System.String.nativeSmallCharToUpper not implemented in Bartok!");
|
||
}
|
||
|
||
// This is a helper method for the security team. They need to construct strings from a char[]
|
||
// within their homebrewed XML parser. They guarantee that the char[] they pass in isn't null and
|
||
// that the provided indices are valid so we just stuff real fast.
|
||
internal static String CreateFromCharArray(char[] array, int start, int count)
|
||
{
|
||
String newString = FastAllocateString(count);
|
||
FillStringArray(newString, 0, array, start, count);
|
||
return newString;
|
||
}
|
||
|
||
//
|
||
//
|
||
// NATIVE INSTANCE METHODS
|
||
//
|
||
//
|
||
|
||
//
|
||
// Search/Query methods
|
||
//
|
||
|
||
// Determines whether two strings match.
|
||
//| <include path='docs/doc[@for="String.Equals"]/*' />
|
||
public override bool Equals(Object obj) {
|
||
if (obj is String) {
|
||
return this.Equals((String) obj);
|
||
} else {
|
||
return false;
|
||
}
|
||
}
|
||
|
||
// Determines whether two strings match.
|
||
//| <include path='docs/doc[@for="String.Equals1"]/*' />
|
||
public unsafe bool Equals(String value) {
|
||
if (value == null) {
|
||
return false;
|
||
}
|
||
if (this.m_stringLength != value.m_stringLength) {
|
||
return false;
|
||
}
|
||
fixed (char *thisCharPtr = &this.m_firstChar) {
|
||
fixed (char *valueCharPtr = &value.m_firstChar) {
|
||
char *thisCursor = thisCharPtr;
|
||
char *valueCursor = valueCharPtr;
|
||
for (int i = this.m_stringLength; i > 0; i--) {
|
||
if (*thisCursor != *valueCursor) {
|
||
return false;
|
||
}
|
||
thisCursor++;
|
||
valueCursor++;
|
||
}
|
||
}
|
||
}
|
||
return true;
|
||
}
|
||
|
||
// Determines whether two Strings match.
|
||
//| <include path='docs/doc[@for="String.Equals2"]/*' />
|
||
public static bool Equals(String a, String b) {
|
||
if ((Object)a==(Object)b) {
|
||
return true;
|
||
}
|
||
|
||
if ((Object)a==null || (Object)b==null) {
|
||
return false;
|
||
}
|
||
|
||
return a.Equals(b);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.operatorEQ"]/*' />
|
||
public static bool operator == (String a, String b)
|
||
{
|
||
return String.Equals(a, b);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.operatorNE"]/*' />
|
||
public static bool operator != (String a, String b)
|
||
{
|
||
return !String.Equals(a, b);
|
||
}
|
||
|
||
internal unsafe char InternalGetChar(int index)
|
||
{
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::GetCharAt
|
||
if ((uint) index >= (uint) this.m_stringLength) {
|
||
throw new IndexOutOfRangeException();
|
||
}
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
return charPtr[index];
|
||
}
|
||
}
|
||
|
||
// Gets the character at a specified position.
|
||
//
|
||
//| <include path='docs/doc[@for="String.this"]/*' />
|
||
public char this[int index] {
|
||
get { return InternalGetChar(index); }
|
||
}
|
||
|
||
internal unsafe int InternalGetChars(char *output, int maxput)
|
||
{
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
int i;
|
||
for (i = 0; i < m_stringLength && i < maxput; i++) {
|
||
output[i] = charPtr[i];
|
||
}
|
||
return i;
|
||
}
|
||
}
|
||
|
||
// Converts a substring of this string to an array of characters. Copies the
|
||
// characters of this string beginning at position startIndex and ending at
|
||
// startIndex + length - 1 to the character array buffer, beginning
|
||
// at bufferStartIndex.
|
||
//
|
||
//| <include path='docs/doc[@for="String.CopyTo"]/*' />
|
||
public void CopyTo(int sourceIndex, char[] destination, int destinationIndex, int count)
|
||
{
|
||
if (destination == null)
|
||
throw new ArgumentNullException("destination");
|
||
if (count < 0)
|
||
throw new ArgumentOutOfRangeException("count", "ArgumentOutOfRange_NegativeCount");
|
||
if (sourceIndex < 0)
|
||
throw new ArgumentOutOfRangeException("sourceIndex", "ArgumentOutOfRange_Index");
|
||
if (count > Length - sourceIndex)
|
||
throw new ArgumentOutOfRangeException("sourceIndex", "ArgumentOutOfRange_IndexCount");
|
||
if (destinationIndex > destination.Length-count || destinationIndex < 0)
|
||
throw new ArgumentOutOfRangeException("destinationIndex", "ArgumentOutOfRange_IndexCount");
|
||
InternalCopyTo(sourceIndex, destination, destinationIndex, count);
|
||
}
|
||
|
||
internal unsafe void InternalCopyTo(int sourceIndex, char[] destination,
|
||
int destinationIndex, int count) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::GetPreallocatedCharArray
|
||
if (sourceIndex + count > this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException();
|
||
}
|
||
if (count > 0) {
|
||
// Necessary test as &destination[0] is illegal for empty array
|
||
fixed (char *srcPtrFixed = &this.m_firstChar) {
|
||
fixed (char *dstPtrFixed = destination) {
|
||
char *srcPtr = srcPtrFixed + sourceIndex;
|
||
char *dstPtr = dstPtrFixed + destinationIndex;
|
||
Buffer.MoveMemory((byte *)dstPtr, (byte *)srcPtr,
|
||
count * sizeof(char));
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
internal unsafe void CopyToByteArray(int sourceIndex, byte[] destination,
|
||
int destinationIndex, int charCount)
|
||
{
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::InternalCopyToByteArray
|
||
// Rusa: apparently there are no calls to this method
|
||
// arlied: Actually, there is at least one call to this method, and I need to use it.
|
||
// arlied: It's called by System.Text.UnicodeEncoding.GetBytes() for little-endian Unicode.
|
||
|
||
if (destination == null)
|
||
throw new ArgumentNullException("destination");
|
||
if (charCount < 0)
|
||
throw new ArgumentOutOfRangeException("charCount");
|
||
if (destinationIndex < 0)
|
||
throw new ArgumentOutOfRangeException("destinationIndex");
|
||
if (sourceIndex < 0)
|
||
throw new ArgumentOutOfRangeException("sourceIndex");
|
||
if (sourceIndex + charCount > this.m_stringLength)
|
||
throw new ArgumentOutOfRangeException("charCount");
|
||
|
||
int byteCount = charCount * 2;
|
||
if (destinationIndex + byteCount > destination.Length)
|
||
throw new ArgumentOutOfRangeException("destinationLength");
|
||
|
||
if (charCount > 0) {
|
||
fixed (byte* dest = &destination[destinationIndex])
|
||
fixed (char* src_char0 = &this.m_firstChar) {
|
||
byte* src_bytes = (byte*)(&src_char0[sourceIndex]);
|
||
Buffer.MoveMemory(dest, src_bytes, byteCount);
|
||
}
|
||
}
|
||
}
|
||
|
||
// Returns the entire string as an array of characters.
|
||
//| <include path='docs/doc[@for="String.ToCharArray"]/*' />
|
||
public char[] ToCharArray() {
|
||
return ToCharArray(0,Length);
|
||
}
|
||
|
||
// Returns a substring of this string as an array of characters.
|
||
//
|
||
//| <include path='docs/doc[@for="String.ToCharArray1"]/*' />
|
||
public char[] ToCharArray(int startIndex, int length)
|
||
{
|
||
// Range check everything.
|
||
if (startIndex < 0 || startIndex > Length || startIndex > Length - length)
|
||
throw new ArgumentOutOfRangeException("startIndex", "ArgumentOutOfRange_Index");
|
||
if (length < 0)
|
||
throw new ArgumentOutOfRangeException("length", "ArgumentOutOfRange_Index");
|
||
|
||
char[] chars = new char[length];
|
||
InternalCopyTo(startIndex, chars, 0, length);
|
||
return chars;
|
||
}
|
||
|
||
// Gets a hash code for this string. If strings A and B are such that A.Equals(B), then
|
||
// they will return the same hash code.
|
||
//| <include path='docs/doc[@for="String.GetHashCode"]/*' />
|
||
public unsafe override int GetHashCode() {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::GetHashCode
|
||
fixed (char *charPtrFixed = &this.m_firstChar) {
|
||
char *charPtr = charPtrFixed;
|
||
if (charPtr[this.m_stringLength] != 0) {
|
||
throw new Exception("Bartok string is not null terminated");
|
||
}
|
||
// Rusa: see also Lightning\Src\UtilCode.h::HashString
|
||
uint hash = 5381;
|
||
char c = *charPtr;
|
||
while (c != 0) {
|
||
hash = ((hash << 5) + hash) ^ c;
|
||
charPtr++;
|
||
c = *charPtr;
|
||
}
|
||
return (int) hash;
|
||
}
|
||
}
|
||
|
||
// Gets the length of this string
|
||
//| <include path='docs/doc[@for="String.Length"]/*' />
|
||
public int Length {
|
||
get { return InternalLength(); }
|
||
}
|
||
|
||
/// This is a EE implemented function so that the JIT can recognise it and
|
||
/// treat it specially by eliminating checks on character fetches.
|
||
private int InternalLength() {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::Length
|
||
// Rusa: see also Lightning\Src\VM\object.h::GetStringLength
|
||
return this.m_stringLength;
|
||
}
|
||
|
||
///<internalonly/>
|
||
internal int ArrayLength {
|
||
get { return (m_arrayLength); }
|
||
}
|
||
|
||
// Used by StringBuilder
|
||
internal int Capacity {
|
||
get { return (m_arrayLength - 1); }
|
||
}
|
||
|
||
// Creates an array of strings by splitting this string at each
|
||
// occurrence of a separator. The separator is searched for, and if found,
|
||
// the substring preceding the occurrence is stored as the first element in
|
||
// the array of strings. We then continue in this manner by searching
|
||
// the substring that follows the occurrence. On the other hand, if the separator
|
||
// is not found, the array of strings will contain this instance as its only element.
|
||
// If the separator is null
|
||
// whitespace (i.e., Character.IsWhitespace) is used as the separator.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Split"]/*' />
|
||
public String [] Split(params char [] separator) {
|
||
return Split(separator, Int32.MaxValue);
|
||
}
|
||
|
||
/*==============================MakeSeparatorList========================
|
||
**Args: baseString -- the string to parse for the given list of
|
||
** separator chars.
|
||
** Separator -- A string containing all of the split characters.
|
||
** list -- a pointer to a caller-allocated array of ints for
|
||
** split char indices.
|
||
** listLength -- the number of slots allocated in list.
|
||
**Returns: A list of all of the places within baseString where instances
|
||
** of characters in Separator occur.
|
||
**Exceptions: None.
|
||
**N.B.: This just returns silently if the caller hasn't allocated
|
||
** enough space for the int list.
|
||
=======================================================================*/
|
||
|
||
// WCHAR * -> char *
|
||
// int * -> int[]
|
||
|
||
// CHARARRAYREF --> char[]
|
||
// c->GetNumComponents --> c.Length
|
||
|
||
// STRINGREF --> String
|
||
// s->GetStringLength --> s.Length
|
||
// s->GetBuffer --> fixed(&s.m_firstChar)
|
||
|
||
// COMNlsInfo::nativeIsWhiteSpace --> Char.IsWhiteSpace
|
||
// ArrayContains(char,char* start,char* end)
|
||
// --> ArrayContains(char, char[] buf)
|
||
private unsafe static int MakeSeparatorList(String baseString, char[] Separator, int[] list, int listLength) {
|
||
// markples: from Lightning\Src\VM\COMString.cpp::MakeSeparatorList
|
||
int i;
|
||
int foundCount=0;
|
||
fixed (char *thisChars = &baseString.m_firstChar) {
|
||
int thisLength = baseString.Length;
|
||
|
||
if (Separator==null || Separator.Length==0) {
|
||
//If they passed null or an empty string,
|
||
//look for whitespace.
|
||
for (i=0; i<thisLength && foundCount < listLength; i++) {
|
||
// was nativeIsWhiteSpace()
|
||
if (Char.IsWhiteSpace(thisChars[i])) {
|
||
list[foundCount++]=i;
|
||
}
|
||
}
|
||
} else {
|
||
//WCHAR *searchChars = (WCHAR *)Separator->GetDataPtr();
|
||
|
||
int searchLength = Separator.Length;
|
||
//If they passed in a string of chars,
|
||
//actually look for those chars.
|
||
for (i=0; i<thisLength && foundCount < listLength; i++) {
|
||
if (ArrayContains(thisChars[i],Separator) >= 0) {
|
||
list[foundCount++]=i;
|
||
}
|
||
}
|
||
}
|
||
return foundCount;
|
||
}
|
||
}
|
||
|
||
// Creates an array of strings by splitting this string at each
|
||
// occurrence of a separator. The separator is searched for, and if found,
|
||
// the substring preceding the occurrence is stored as the first element in
|
||
// the array of strings. We then continue in this manner by searching
|
||
// the substring that follows the occurrence. On the other hand, if the separator
|
||
// is not found, the array of strings will contain this instance as its only element.
|
||
// If the separator is the empty string (i.e., String.Empty), then
|
||
// whitespace (i.e., Character.IsWhitespace) is used as the separator.
|
||
// If there are more than count different strings, the last n-(count-1)
|
||
// elements are concatenated and added as the last String.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Split1"]/*' />
|
||
//| <include path='docs/doc[@for="String.Split1"]/*' />
|
||
|
||
// STRINGREF --> String
|
||
// PTRARRAYREF --> String[]
|
||
// p->SetAt(0, v) --> p[0] = v
|
||
//
|
||
// LPVOID --> gone
|
||
// CQuickBytes --> gone
|
||
|
||
// AllocateObjectArray(x,g_pStringClass) --> new String[x]
|
||
// NewString(&ref, index, size) --> ref.Substring(index, size)
|
||
|
||
public String[] Split(char[] separator, int count) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::Split
|
||
// markples: this implementation based on COMString
|
||
|
||
int numReplaces;
|
||
int numActualReplaces;
|
||
int[] sepList;
|
||
int currIndex=0;
|
||
int arrIndex=0;
|
||
//char *thisChars;
|
||
int thisLength;
|
||
int i;
|
||
String[] splitStrings;
|
||
String temp;
|
||
|
||
if (count<0) {
|
||
throw new ArgumentOutOfRangeException
|
||
("count", "ArgumentOutOfRange_Negative");
|
||
}
|
||
|
||
//Allocate space and fill an array of ints with a list of everyplace
|
||
//within our String that a separator character occurs.
|
||
sepList = new int[this.Length];
|
||
numReplaces = MakeSeparatorList(this, separator, sepList, this.Length);
|
||
//Handle the special case of no replaces.
|
||
if (0==numReplaces) {
|
||
splitStrings = new String[1];
|
||
splitStrings[0] = this;
|
||
return splitStrings;
|
||
}
|
||
thisLength = Length;
|
||
|
||
count--;
|
||
numActualReplaces = (numReplaces<count) ? numReplaces : count;
|
||
|
||
//Allocate space for the new array.
|
||
//+1 for the string from the end of the last replace to the end
|
||
//of the String.
|
||
splitStrings = new String[numActualReplaces+1];
|
||
|
||
for (i=0; i<numActualReplaces && currIndex<thisLength; i++) {
|
||
temp = this.Substring(currIndex, sepList[i]-currIndex);
|
||
|
||
splitStrings[arrIndex++] = temp;
|
||
currIndex=sepList[i]+1;
|
||
}
|
||
|
||
//Handle the last string at the end of the array if there is one.
|
||
|
||
if (currIndex<thisLength) {
|
||
temp = this.Substring(currIndex, thisLength-currIndex);
|
||
splitStrings[arrIndex] = temp;
|
||
} else if (arrIndex==numActualReplaces) {
|
||
//We had a separator character at the end of a string. Rather than just allowing
|
||
//a null character, we'll replace the last element in the array with an empty string.
|
||
temp = Empty;
|
||
splitStrings[arrIndex] = temp;
|
||
}
|
||
|
||
return splitStrings;
|
||
}
|
||
|
||
// Returns a substring of this string.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Substring"]/*' />
|
||
public String Substring(int startIndex) {
|
||
return this.Substring (startIndex, Length-startIndex);
|
||
}
|
||
|
||
// Returns a substring of this string.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Substring1"]/*' />
|
||
public String Substring(int startIndex, int length) {
|
||
|
||
int thisLength = Length;
|
||
|
||
//Bounds Checking.
|
||
if (startIndex<0) {
|
||
throw new ArgumentOutOfRangeException("startIndex", "ArgumentOutOfRange_StartIndex");
|
||
}
|
||
|
||
if (length<0) {
|
||
throw new ArgumentOutOfRangeException("length", "ArgumentOutOfRange_NegativeLength");
|
||
}
|
||
|
||
if (startIndex > thisLength-length) {
|
||
throw new ArgumentOutOfRangeException("length", "ArgumentOutOfRange_IndexLength");
|
||
}
|
||
|
||
String s = FastAllocateString(length);
|
||
FillSubstring(s, 0, this, startIndex, length);
|
||
|
||
return s;
|
||
}
|
||
|
||
internal String TrimHelper(char[] trimChars, int trimType)
|
||
{
|
||
// Rusa: see also Lighting\Src\VM\COMString.cpp::TrimHelper
|
||
int stringLength = this.m_stringLength;
|
||
int iLeft = 0;
|
||
int iRight = stringLength - 1;
|
||
this.TrimHelper(trimChars, trimType, ref iLeft, ref iRight);
|
||
int newLength = iRight - iLeft + 1;
|
||
if (newLength == stringLength) {
|
||
return this;
|
||
} else if (newLength == 0) {
|
||
return String.Empty;
|
||
} else {
|
||
String result = FastAllocateString(newLength);
|
||
FillSubstring(result, 0, this, iLeft, newLength);
|
||
return result;
|
||
}
|
||
}
|
||
|
||
private unsafe void TrimHelper(char[] trimChars, int trimType,
|
||
ref int iLeft, ref int iRight)
|
||
{
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
if (trimType == String.TrimHead ||
|
||
trimType == String.TrimBoth) {
|
||
while (iLeft <= iRight &&
|
||
ArrayContains(charPtr[iLeft], trimChars) >= 0) {
|
||
iLeft++;
|
||
}
|
||
}
|
||
if (trimType == String.TrimTail ||
|
||
trimType == String.TrimBoth) {
|
||
while (iRight >= iLeft &&
|
||
ArrayContains(charPtr[iRight], trimChars) >= 0) {
|
||
iRight--;
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
private static int ArrayContains(char c, char[] charArray) {
|
||
int limit = charArray.Length;
|
||
for (int i = 0; i < limit; i++) {
|
||
if (charArray[i] == c) {
|
||
return i;
|
||
}
|
||
}
|
||
return -1;
|
||
}
|
||
|
||
/**
|
||
* Creates a new string from the characters in a subarray. The new
|
||
* string will be created from the characters in <i>value</i> between
|
||
* <i>startIndex</i> and <i>startIndex</i> + <i>length</i> - 1.
|
||
*
|
||
* @param value an array of characters.
|
||
* @param startIndex the index at which the subarray begins.
|
||
* @param length the length of the subarray.
|
||
*
|
||
* @exception ArgumentException if value is <b>null</b>.
|
||
* @exception ArgumentException if <i>startIndex</i> or
|
||
* <i>startIndex</i>+<i>length</i>-1 are not valid indices of
|
||
* <i>value</i>.
|
||
*/
|
||
//| <include path='docs/doc[@for="String.String7"]/*' />
|
||
|
||
public static String StringCTOR(char[] value, int startIndex,
|
||
int length) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitCharArray
|
||
if (value == null) {
|
||
throw new ArgumentNullException("array value is null");
|
||
}
|
||
if (startIndex < 0) {
|
||
throw new ArgumentOutOfRangeException("startIndex is negative");
|
||
}
|
||
if (length < 0) {
|
||
throw new ArgumentOutOfRangeException("length is negative");
|
||
}
|
||
if (length == 0) {
|
||
return String.Empty;
|
||
} else {
|
||
String result = FastAllocateString(length);
|
||
FillStringArray(result, 0, value, startIndex, length);
|
||
return result;
|
||
}
|
||
}
|
||
|
||
/**
|
||
* Creates a new string from the characters in a subarray. The new
|
||
* string will be created from the characters in <i>value</i>.
|
||
*
|
||
* @param value an array of characters.
|
||
*
|
||
* @exception ArgumentException if value is <b>null</b>.
|
||
*/
|
||
//| <include path='docs/doc[@for="String.String5"]/*' />
|
||
public static String StringCTOR(char[] value) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitChars
|
||
if (value == null) {
|
||
return String.Empty;
|
||
}
|
||
int length = value.Length;
|
||
if (length == 0) {
|
||
return String.Empty;
|
||
}
|
||
String result = FastAllocateString(length);
|
||
FillStringArray(result, 0, value, 0, length);
|
||
return result;
|
||
}
|
||
|
||
internal static String NewString(String curr,int start,int copyLen,
|
||
int capacity) {
|
||
String result = FastAllocateString(capacity);
|
||
FillSubstring(result, 0, curr, start, copyLen);
|
||
return result;
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.String6"]/*' />
|
||
public static String StringCTOR(char c, int count) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitCharCount
|
||
if (count < 0) {
|
||
throw new ArgumentOutOfRangeException("count is negative");
|
||
}
|
||
if (count == 0) {
|
||
return String.Empty;
|
||
} else {
|
||
String result = FastAllocateString(count);
|
||
FillStringChar(result, 0, c, count);
|
||
return result;
|
||
}
|
||
}
|
||
|
||
// Removes a string of characters from the ends of this string.
|
||
//| <include path='docs/doc[@for="String.Trim"]/*' />
|
||
public String Trim(params char[] trimChars) {
|
||
if (null==trimChars || trimChars.Length == 0) {
|
||
trimChars=CharacterInfo.WhitespaceChars;
|
||
}
|
||
return TrimHelper(trimChars,TrimBoth);
|
||
}
|
||
|
||
// Removes a string of characters from the beginning of this string.
|
||
//| <include path='docs/doc[@for="String.TrimStart"]/*' />
|
||
public String TrimStart(params char[] trimChars) {
|
||
if (null==trimChars || trimChars.Length == 0) {
|
||
trimChars=CharacterInfo.WhitespaceChars;
|
||
}
|
||
return TrimHelper(trimChars,TrimHead);
|
||
}
|
||
|
||
|
||
// Removes a string of characters from the end of this string.
|
||
//| <include path='docs/doc[@for="String.TrimEnd"]/*' />
|
||
public String TrimEnd(params char[] trimChars) {
|
||
if (null==trimChars || trimChars.Length == 0) {
|
||
trimChars=CharacterInfo.WhitespaceChars;
|
||
}
|
||
return TrimHelper(trimChars,TrimTail);
|
||
}
|
||
|
||
|
||
// Creates a new string with the characters copied in from ptr. If
|
||
// ptr is null, a string initialized to ";<;No Object>;"; (i.e.,
|
||
// String.NullString) is created.
|
||
//
|
||
// Issue: This method is only accessible from VC.
|
||
//| <include path='docs/doc[@for="String.String"]/*' />
|
||
[CLSCompliant(false), MethodImpl(MethodImplOptions.InternalCall)]
|
||
unsafe public extern String(char *value);
|
||
//| <include path='docs/doc[@for="String.String1"]/*' />
|
||
[CLSCompliant(false), MethodImpl(MethodImplOptions.InternalCall)]
|
||
unsafe public extern String(char *value, int startIndex, int length);
|
||
|
||
//| <include path='docs/doc[@for="String.String2"]/*' />
|
||
[CLSCompliant(false), MethodImpl(MethodImplOptions.InternalCall)]
|
||
unsafe public extern String(sbyte *value);
|
||
//| <include path='docs/doc[@for="String.String3"]/*' />
|
||
[CLSCompliant(false), MethodImpl(MethodImplOptions.InternalCall)]
|
||
unsafe public extern String(sbyte *value, int startIndex, int length);
|
||
|
||
//| <include path='docs/doc[@for="String.String4"]/*' />
|
||
[CLSCompliant(false), MethodImpl(MethodImplOptions.InternalCall)]
|
||
unsafe public extern String(sbyte *value, int startIndex, int length, Encoding enc);
|
||
|
||
unsafe static private String CreateString(sbyte *value, int startIndex, int length, Encoding enc) {
|
||
if (enc == null)
|
||
return new String(value, startIndex, length); // default to ANSI
|
||
if (length < 0)
|
||
throw new ArgumentOutOfRangeException("length","ArgumentOutOfRange_NeedNonNegNum");
|
||
if (startIndex < 0) {
|
||
throw new ArgumentOutOfRangeException("startIndex","ArgumentOutOfRange_StartIndex");
|
||
}
|
||
if ((value + startIndex) < value) {
|
||
// overflow check
|
||
throw new ArgumentOutOfRangeException("startIndex","ArgumentOutOfRange_PartialWCHAR");
|
||
}
|
||
byte [] b = new byte[length];
|
||
if (length > 0) {
|
||
fixed(byte* pDest = b) {
|
||
Buffer.MoveMemory(pDest, ((byte*)value)+startIndex, length);
|
||
}
|
||
}
|
||
return enc.GetString(b);
|
||
}
|
||
|
||
// For ASCIIEncoding::GetString()
|
||
unsafe static internal String CreateStringFromASCII(byte[] bytes, int startIndex, int length) {
|
||
Debug.Assert(bytes != null, "need a byte[].");
|
||
Debug.Assert(startIndex >= 0 && (startIndex < bytes.Length || bytes.Length == 0), "startIndex >= 0 && startIndex < bytes.Length");
|
||
Debug.Assert(length >= 0 && length <= bytes.Length - startIndex, "length >= 0 && length <= bytes.Length - startIndex");
|
||
if (length == 0)
|
||
return String.Empty;
|
||
String s = FastAllocateString(length);
|
||
fixed(char* pChars = &s.m_firstChar) {
|
||
for(int i=0; i<length; i++)
|
||
pChars[i] = (char) (bytes[i+startIndex] & 0x7f);
|
||
}
|
||
return s;
|
||
}
|
||
|
||
// For Latin1Encoding::GetString()
|
||
unsafe static internal String CreateStringFromLatin1(byte[] bytes, int startIndex, int length) {
|
||
Debug.Assert(bytes != null, "need a byte[].");
|
||
Debug.Assert(startIndex >= 0 && (startIndex < bytes.Length || bytes.Length == 0), "startIndex >= 0 && startIndex < bytes.Length");
|
||
Debug.Assert(length >= 0 && length <= bytes.Length - startIndex, "length >= 0 && length <= bytes.Length - startIndex");
|
||
if (length == 0)
|
||
return String.Empty;
|
||
String s = FastAllocateString(length);
|
||
fixed(char* pChars = &s.m_firstChar) {
|
||
for(int i=0; i<length; i++)
|
||
pChars[i] = (char) (bytes[i+startIndex]);
|
||
}
|
||
return s;
|
||
}
|
||
|
||
/**
|
||
* Creates a new string with the characters copied in from <i>ptr</i>. If
|
||
* <i>ptr</i> is null, a string initialized to
|
||
* "<No Object>" (i.e., <code>String.NullString</code>)
|
||
* is created.
|
||
*
|
||
* Issue: This method is only accessible from VC.
|
||
* @issue The verifier needs to recognize that this is not necessarily
|
||
* secure.
|
||
*
|
||
* @param ptr this is a WCHAR *.
|
||
*
|
||
*/
|
||
//| <include path='docs/doc[@for="String.String"]/*' />
|
||
[CLSCompliant(false)]
|
||
public unsafe static String StringCTOR(char *valuePtr) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitWCHARPtr
|
||
if (valuePtr == null) {
|
||
return String.Empty;
|
||
}
|
||
// First figure out how many characters the string has
|
||
int count = 0;
|
||
char *cursor = valuePtr;
|
||
int c = *cursor;
|
||
while (c != 0) {
|
||
count++;
|
||
cursor++;
|
||
c = *cursor;
|
||
}
|
||
// Allocate and fill in the string object
|
||
if (count == 0) {
|
||
return String.Empty;
|
||
} else {
|
||
String result = FastAllocateString(count);
|
||
FillStringCharPtr(result, 0, valuePtr, count);
|
||
return result;
|
||
}
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.String1"]/*' />
|
||
[CLSCompliant(false)]
|
||
public unsafe static String StringCTOR(char *valuePtr, int startIndex,
|
||
int length) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitWCHARPtrPartial
|
||
if (valuePtr == null) {
|
||
return String.Empty;
|
||
}
|
||
if (length < 0) {
|
||
throw new ArgumentOutOfRangeException("length is negative");
|
||
}
|
||
if (startIndex < 0) {
|
||
throw new ArgumentOutOfRangeException("startIndex is negative");
|
||
}
|
||
if (length == 0) {
|
||
return String.Empty;
|
||
} else {
|
||
String result = FastAllocateString(length);
|
||
FillStringCharPtr(result, 0, valuePtr + startIndex, length);
|
||
return result;
|
||
}
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.String1"]/*' />
|
||
[CLSCompliant(false)]
|
||
public unsafe static String StringCTOR(byte *valuePtr, int startIndex,
|
||
int length) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitWCHARPtrPartial
|
||
if (valuePtr == null) {
|
||
return String.Empty;
|
||
}
|
||
if (length < 0) {
|
||
throw new ArgumentOutOfRangeException("length is negative");
|
||
}
|
||
if (startIndex < 0) {
|
||
throw new ArgumentOutOfRangeException("startIndex is negative");
|
||
}
|
||
if (length == 0) {
|
||
return String.Empty;
|
||
} else {
|
||
String result = FastAllocateString(length);
|
||
FillStringBytePtr(result, 0, valuePtr + startIndex, length);
|
||
return result;
|
||
}
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.String2"]/*' />
|
||
[CLSCompliant(false)]
|
||
public unsafe static String StringCTOR(sbyte *valuePtr) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitCharPtr
|
||
return StringInitCharHelper(valuePtr, -1);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.String3"]/*' />
|
||
[CLSCompliant(false)]
|
||
public unsafe static String StringCTOR(sbyte *valuePtr, int startIndex,
|
||
int length) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitCharPtrPartial
|
||
if (startIndex < 0) {
|
||
throw new ArgumentOutOfRangeException("startIndex is negative");
|
||
}
|
||
if (length < 0) {
|
||
throw new ArgumentOutOfRangeException("length is negative");
|
||
}
|
||
return StringInitCharHelper(valuePtr + startIndex, length);
|
||
}
|
||
|
||
private unsafe static String StringInitCharHelper(sbyte *valuePtr,
|
||
int length) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitCharHelper
|
||
throw new Exception("StringInitCharHelper is not implemented in Bartok!");
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.String4"]/*' />
|
||
[CLSCompliant(false)]
|
||
public unsafe static String StringCTOR(sbyte *value, int startIndex,
|
||
int length, Encoding enc) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::StringInitSBytPtrPartialEx
|
||
throw new Exception("String(sbyte*, int, int, Encoding) not implemented in Bartok!");
|
||
}
|
||
|
||
private static String FastAllocateString(int stringLength) {
|
||
return GC.AllocateString(stringLength);
|
||
}
|
||
|
||
[Inline]
|
||
internal void InitializeStringLength(int stringLength) {
|
||
this.m_arrayLength = stringLength+1;
|
||
this.m_stringLength = stringLength;
|
||
}
|
||
|
||
private unsafe static void FillString(String dest, int destPos,
|
||
String src) {
|
||
fixed (char *srcPtr = &src.m_firstChar) {
|
||
FillStringCharPtr(dest, destPos, srcPtr, src.m_stringLength);
|
||
}
|
||
}
|
||
|
||
private unsafe static void FillStringChecked(String dest, int destPos,
|
||
String src) {
|
||
if (! (src.Length <= dest.ArrayLength - destPos)) {
|
||
throw new IndexOutOfRangeException(); // REVIEW: param?
|
||
}
|
||
fixed (char *srcPtr = &src.m_firstChar) {
|
||
FillStringCharPtr(dest, destPos, srcPtr, src.m_stringLength);
|
||
}
|
||
}
|
||
|
||
private unsafe static void FillStringEx(String dest, int destPos,
|
||
String src,int srcLength) {
|
||
// ASSERT(srcLength <= dest.ArrayLength - destPos);
|
||
fixed (char *srcPtr = &src.m_firstChar) {
|
||
FillStringCharPtr(dest, destPos, srcPtr, srcLength);
|
||
}
|
||
}
|
||
|
||
private unsafe static void FillStringChar(String dest, int destPos,
|
||
char c, int count) {
|
||
fixed (char *destPtr = &dest.m_firstChar) {
|
||
char *charPtr = destPtr + destPos;
|
||
// Set the odd leader char if necessary
|
||
if (destPos % 2 == 1) {
|
||
*charPtr = c;
|
||
charPtr++;
|
||
count--;
|
||
}
|
||
// Set the buffer 2 chars at a time
|
||
int c2 = (c << 16) | c;
|
||
int *intPtr = (int *) charPtr;
|
||
count--; // Prevent overruns from odd lengths
|
||
while (count > 0) {
|
||
*intPtr = c2;
|
||
intPtr++;
|
||
count -= 2;
|
||
}
|
||
// Set the odd trailer char if necessary
|
||
if (count == 0) {
|
||
*((char *) intPtr) = c;
|
||
}
|
||
}
|
||
}
|
||
|
||
private unsafe static void FillStringCharPtr(String dest, int destPos,
|
||
char *srcPtr, int count) {
|
||
fixed (char *charPtr = &dest.m_firstChar) {
|
||
char *destPtr = charPtr + destPos;
|
||
Buffer.MoveMemory((byte *)destPtr, (byte *)srcPtr, count * sizeof(char));
|
||
}
|
||
}
|
||
|
||
private unsafe static void FillStringBytePtr(String dest, int destPos,
|
||
byte *srcPtr, int count) {
|
||
fixed (char *charPtr = &dest.m_firstChar) {
|
||
char *destPtr = charPtr + destPos;
|
||
for (int i = 0; i < count; i++) {
|
||
*destPtr++ = (char)*srcPtr++;
|
||
}
|
||
}
|
||
}
|
||
|
||
private unsafe static void FillStringArray(String dest, int stringStart,
|
||
char[] array, int charStart,
|
||
int count) {
|
||
fixed (char *destPtr = &array[charStart]) {
|
||
FillStringCharPtr(dest, stringStart, destPtr, count);
|
||
}
|
||
}
|
||
|
||
private unsafe static void FillSubstring(String dest, int destPos,
|
||
String src, int startPos,
|
||
int count) {
|
||
fixed (char *srcPtr = &src.m_firstChar) {
|
||
FillStringCharPtr(dest, destPos, srcPtr + startPos, count);
|
||
}
|
||
}
|
||
|
||
// Creates a new string from the characters in a subarray. The new string will
|
||
// be created from the characters in value between startIndex and
|
||
// startIndex + length - 1.
|
||
//
|
||
//| <include path='docs/doc[@for="String.String7"]/*' />
|
||
[MethodImpl(MethodImplOptions.InternalCall)]
|
||
public extern String(char [] value, int startIndex, int length);
|
||
|
||
// Creates a new string from the characters in a subarray. The new string will be
|
||
// created from the characters in value.
|
||
//
|
||
//| <include path='docs/doc[@for="String.String5"]/*' />
|
||
[MethodImpl(MethodImplOptions.InternalCall)]
|
||
public extern String(char [] value);
|
||
|
||
//| <include path='docs/doc[@for="String.String6"]/*' />
|
||
[MethodImpl(MethodImplOptions.InternalCall)]
|
||
public extern String(char c, int count);
|
||
|
||
|
||
//
|
||
//
|
||
// INSTANCE METHODS
|
||
//
|
||
//
|
||
|
||
// Provides a culture-correct string comparison. StrA is compared to StrB
|
||
// to determine whether it is lexicographically less, equal, or greater, and then returns
|
||
// either a negative integer, 0, or a positive integer; respectively.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Compare"]/*' />
|
||
public static int Compare(String strA, String strB) {
|
||
return CompareInfo.Compare(strA, strB, CompareOptions.None);
|
||
}
|
||
|
||
// Provides a culture-correct string comparison. strA is compared to strB
|
||
// to determine whether it is lexicographically less, equal, or greater, and then a
|
||
// negative integer, 0, or a positive integer is returned; respectively.
|
||
// The case-sensitive option is set by ignoreCase
|
||
//
|
||
//| <include path='docs/doc[@for="String.Compare1"]/*' />
|
||
public static int Compare(String strA, String strB, bool ignoreCase) {
|
||
if (ignoreCase) {
|
||
return CompareInfo.Compare(strA, strB, CompareOptions.IgnoreCase);
|
||
}
|
||
return CompareInfo.Compare(strA, strB, CompareOptions.None);
|
||
}
|
||
|
||
//
|
||
public static int Compare(String strA, String strB, bool ignoreCase,
|
||
CultureInfo info)
|
||
{
|
||
return Compare(strA, strB, ignoreCase);
|
||
}
|
||
|
||
// Determines whether two string regions match. The substring of strA beginning
|
||
// at indexA of length count is compared with the substring of strB
|
||
// beginning at indexB of the same length.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Compare3"]/*' />
|
||
public static int Compare(String strA, int indexA, String strB, int indexB, int length) {
|
||
int lengthA = length;
|
||
int lengthB = length;
|
||
|
||
if (strA!=null) {
|
||
if (strA.Length - indexA < lengthA) {
|
||
lengthA = (strA.Length - indexA);
|
||
}
|
||
}
|
||
|
||
if (strB!=null) {
|
||
if (strB.Length - indexB < lengthB) {
|
||
lengthB = (strB.Length - indexB);
|
||
}
|
||
}
|
||
return CompareInfo.Compare(strA, indexA, lengthA, strB, indexB, lengthB, CompareOptions.None);
|
||
}
|
||
|
||
|
||
// Determines whether two string regions match. The substring of strA beginning
|
||
// at indexA of length count is compared with the substring of strB
|
||
// beginning at indexB of the same length. Case sensitivity is determined by the ignoreCase boolean.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Compare4"]/*' />
|
||
public static int Compare(String strA, int indexA, String strB, int indexB, int length, bool ignoreCase) {
|
||
int lengthA = length;
|
||
int lengthB = length;
|
||
|
||
if (strA!=null) {
|
||
if (strA.Length - indexA < lengthA) {
|
||
lengthA = (strA.Length - indexA);
|
||
}
|
||
}
|
||
|
||
if (strB!=null) {
|
||
if (strB.Length - indexB < lengthB) {
|
||
lengthB = (strB.Length - indexB);
|
||
}
|
||
}
|
||
|
||
if (ignoreCase) {
|
||
return CompareInfo.Compare(strA, indexA, lengthA, strB, indexB, lengthB, CompareOptions.IgnoreCase);
|
||
}
|
||
return CompareInfo.Compare(strA, indexA, lengthA, strB, indexB, lengthB, CompareOptions.None);
|
||
}
|
||
|
||
// Compares this object to another object, returning an integer that
|
||
// indicates the relationship. This method returns a value less than 0 if this is less than value, 0
|
||
// if this is equal to value, or a value greater than 0
|
||
// if this is greater than value. Strings are considered to be
|
||
// greater than all non-String objects. Note that this means sorted
|
||
// arrays would contain nulls, other objects, then Strings in that order.
|
||
//
|
||
//| <include path='docs/doc[@for="String.CompareTo"]/*' />
|
||
public int CompareTo(Object value) {
|
||
if (value == null) {
|
||
return 1;
|
||
}
|
||
|
||
if (!(value is String)) {
|
||
throw new ArgumentException("Arg_MustBeString");
|
||
}
|
||
|
||
return String.Compare(this,(String)value);
|
||
}
|
||
|
||
// Determines the sorting relation of StrB to the current instance.
|
||
//
|
||
//| <include path='docs/doc[@for="String.CompareTo1"]/*' />
|
||
public int CompareTo(String strB) {
|
||
if (strB==null) {
|
||
return 1;
|
||
}
|
||
return CompareInfo.Compare(this, strB, 0);
|
||
}
|
||
|
||
// Compares strA and strB using an ordinal (code-point) comparison.
|
||
//
|
||
//| <include path='docs/doc[@for="String.CompareOrdinal"]/*' />
|
||
public static int CompareOrdinal(String strA, String strB) {
|
||
if (strA == null || strB == null) {
|
||
if ((Object)strA==(Object)strB) { //they're both null;
|
||
return 0;
|
||
}
|
||
return (strA==null)? -1 : 1; //-1 if A is null, 1 if B is null.
|
||
}
|
||
|
||
return nativeCompareOrdinal(strA, strB, false);
|
||
}
|
||
|
||
// Compares strA and strB using an ordinal (code-point) comparison.
|
||
//
|
||
//| <include path='docs/doc[@for="String.CompareOrdinal"]/*' />
|
||
public static int CompareOrdinal(String strA, String strB, bool ignoreCase) {
|
||
if (strA == null || strB == null) {
|
||
if ((Object)strA==(Object)strB) { //they're both null;
|
||
return 0;
|
||
}
|
||
return (strA==null)? -1 : 1; //-1 if A is null, 1 if B is null.
|
||
}
|
||
|
||
return nativeCompareOrdinal(strA, strB, ignoreCase);
|
||
}
|
||
|
||
// Compares strA and strB using an ordinal (code-point) comparison.
|
||
//
|
||
//| <include path='docs/doc[@for="String.CompareOrdinal1"]/*' />
|
||
public static int CompareOrdinal(String strA, int indexA, String strB, int indexB, int length) {
|
||
if (strA == null || strB == null) {
|
||
if ((Object)strA==(Object)strB) { //they're both null;
|
||
return 0;
|
||
}
|
||
|
||
return (strA==null)? -1 : 1; //-1 if A is null, 1 if B is null.
|
||
}
|
||
|
||
return nativeCompareOrdinalEx(strA, indexA, strB, indexB, length);
|
||
}
|
||
|
||
|
||
// Determines whether a specified string is a suffix of the the current instance.
|
||
//
|
||
// The case-sensitive and culture-sensitive option is set by options,
|
||
// and the default culture is used.
|
||
//
|
||
//| <include path='docs/doc[@for="String.EndsWith"]/*' />
|
||
public bool EndsWith(String value) {
|
||
if (null==value) {
|
||
throw new ArgumentNullException("value");
|
||
}
|
||
int valueLen = value.Length;
|
||
int thisLen = this.Length;
|
||
if (valueLen>thisLen) {
|
||
return false;
|
||
}
|
||
return (0==Compare(this, thisLen-valueLen, value, 0, valueLen));
|
||
}
|
||
|
||
public bool EndsWith(char value) {
|
||
int thisLen = this.Length;
|
||
if (thisLen != 0) {
|
||
if (this[thisLen - 1] == value)
|
||
return true;
|
||
}
|
||
return false;
|
||
}
|
||
|
||
|
||
// Returns the index of the first occurrence of value in the current instance.
|
||
// The search starts at startIndex and runs thorough the next count characters.
|
||
//
|
||
//| <include path='docs/doc[@for="String.IndexOf"]/*' />
|
||
public int IndexOf(char value) {
|
||
return IndexOf(value, 0, this.Length);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.IndexOf1"]/*' />
|
||
public int IndexOf(char value, int startIndex) {
|
||
return IndexOf(value, startIndex, this.Length - startIndex);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.IndexOf2"]/*' />
|
||
public unsafe int IndexOf(char value, int startIndex, int count) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::IndexOfChar
|
||
if (startIndex < 0 || startIndex > this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException("startIndex out of range");
|
||
}
|
||
if (count < 0 || startIndex + count > this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException("count out of range");
|
||
}
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
char *cursor = charPtr + startIndex;
|
||
for (int i = count; i > 0; i--) {
|
||
if (*cursor == value) {
|
||
return (startIndex + (count - i));
|
||
}
|
||
cursor++;
|
||
}
|
||
}
|
||
return -1;
|
||
}
|
||
|
||
// Returns the index of the first occurrence of any character in value in the current instance.
|
||
// The search starts at startIndex and runs to endIndex-1. [startIndex,endIndex).
|
||
//
|
||
|
||
//| <include path='docs/doc[@for="String.IndexOfAny1"]/*' />
|
||
public int IndexOfAny(char [] anyOf) {
|
||
return IndexOfAny(anyOf,0, this.Length);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.IndexOfAny2"]/*' />
|
||
public int IndexOfAny(char [] anyOf, int startIndex) {
|
||
return IndexOfAny(anyOf, startIndex, this.Length - startIndex);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.IndexOfAny3"]/*' />
|
||
public unsafe int IndexOfAny(char [] anyOf, int startIndex, int count) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::IndexOfCharArray
|
||
if (anyOf == null) {
|
||
throw new ArgumentNullException("anyOf array is null");
|
||
}
|
||
if (startIndex < 0 || startIndex > this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException("startIndex out of range");
|
||
}
|
||
if (count < 0 || startIndex + count > this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException("count out of range");
|
||
}
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
char *cursor = charPtr + startIndex;
|
||
for (int i = count; i > 0; i--) {
|
||
if (ArrayContains(*cursor, anyOf) >= 0) {
|
||
return (startIndex + (count - i));
|
||
}
|
||
cursor++;
|
||
}
|
||
}
|
||
return -1;
|
||
}
|
||
|
||
// Determines the position within this string of the first occurrence of the specified
|
||
// string, according to the specified search criteria. The search begins at
|
||
// the first character of this string, it is case-sensitive and culture-sensitive,
|
||
// and the default culture is used.
|
||
//
|
||
//| <include path='docs/doc[@for="String.IndexOf6"]/*' />
|
||
public int IndexOf(String value) {
|
||
return CompareInfo.IndexOf(this,value);
|
||
}
|
||
|
||
// Determines the position within this string of the first occurrence of the specified
|
||
// string, according to the specified search criteria. The search begins at
|
||
// startIndex, it is case-sensitive and culture-sensitve, and the default culture is used.
|
||
//
|
||
//| <include path='docs/doc[@for="String.IndexOf7"]/*' />
|
||
public int IndexOf(String value, int startIndex){
|
||
return CompareInfo.IndexOf(this,value,startIndex);
|
||
}
|
||
|
||
// Determines the position within this string of the first occurrence of the specified
|
||
// string, according to the specified search criteria. The search begins at
|
||
// startIndex, ends at endIndex and the default culture is used.
|
||
//
|
||
//| <include path='docs/doc[@for="String.IndexOf8"]/*' />
|
||
public int IndexOf(String value, int startIndex, int count){
|
||
if (startIndex + count > this.Length) {
|
||
throw new ArgumentOutOfRangeException("count", "ArgumentOutOfRange_Index");
|
||
}
|
||
return CompareInfo.IndexOf(this, value, startIndex, count, CompareOptions.None);
|
||
}
|
||
|
||
|
||
// Returns the index of the last occurrence of value in the current instance.
|
||
// The search starts at startIndex and runs to endIndex. [startIndex,endIndex].
|
||
// The character at position startIndex is included in the search. startIndex is the larger
|
||
// index within the string.
|
||
//
|
||
//| <include path='docs/doc[@for="String.LastIndexOf"]/*' />
|
||
public int LastIndexOf(char value) {
|
||
return LastIndexOf(value, this.Length-1, this.Length);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.LastIndexOf1"]/*' />
|
||
public int LastIndexOf(char value, int startIndex){
|
||
return LastIndexOf(value,startIndex,startIndex + 1);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.LastIndexOf2"]/*' />
|
||
public unsafe int LastIndexOf(char value, int startIndex, int count) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::LastIndexOfChar
|
||
if (this.m_stringLength == 0) {
|
||
return -1;
|
||
}
|
||
if (startIndex < 0 || startIndex >= this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException("startIndex out of range");
|
||
}
|
||
if (count < 0 || count - 1 > startIndex) {
|
||
throw new ArgumentOutOfRangeException("count out of range");
|
||
}
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
char *cursor = charPtr + startIndex;
|
||
for (int i = count; i > 0; i--) {
|
||
if (*cursor == value) {
|
||
return (startIndex - (count - i));
|
||
}
|
||
cursor--;
|
||
}
|
||
}
|
||
return -1;
|
||
}
|
||
|
||
// Returns the index of the last occurrence of any character in value in the current instance.
|
||
// The search starts at startIndex and runs to endIndex. [startIndex,endIndex].
|
||
// The character at position startIndex is included in the search. startIndex is the larger
|
||
// index within the string.
|
||
//
|
||
|
||
//| <include path='docs/doc[@for="String.LastIndexOfAny1"]/*' />
|
||
public int LastIndexOfAny(char [] anyOf) {
|
||
return LastIndexOfAny(anyOf,this.Length-1,this.Length);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.LastIndexOfAny2"]/*' />
|
||
public int LastIndexOfAny(char [] anyOf, int startIndex) {
|
||
return LastIndexOfAny(anyOf,startIndex,startIndex + 1);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.LastIndexOfAny3"]/*' />
|
||
public unsafe int LastIndexOfAny(char [] anyOf, int startIndex,
|
||
int count) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::LastIndexOfCharArray
|
||
if (anyOf == null) {
|
||
throw new ArgumentNullException("anyOf array is null");
|
||
}
|
||
if (this.m_stringLength == 0) {
|
||
return -1;
|
||
}
|
||
if (startIndex < 0 || startIndex >= this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException("startIndex out of range");
|
||
}
|
||
if (count < 0 || count - 1 > startIndex) {
|
||
throw new ArgumentOutOfRangeException("count out of range");
|
||
}
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
char *cursor = charPtr + startIndex;
|
||
for (int i = count; i > 0; i--) {
|
||
if (ArrayContains(*cursor, anyOf) >= 0) {
|
||
return (startIndex - (count - 1));
|
||
}
|
||
cursor--;
|
||
}
|
||
}
|
||
return -1;
|
||
}
|
||
|
||
// Returns the index of the last occurrence of any character in value in the current instance.
|
||
// The search starts at startIndex and runs to endIndex. [startIndex,endIndex].
|
||
// The character at position startIndex is included in the search. startIndex is the larger
|
||
// index within the string.
|
||
//
|
||
//| <include path='docs/doc[@for="String.LastIndexOf6"]/*' />
|
||
public int LastIndexOf(String value) {
|
||
return LastIndexOf(value, this.Length-1,this.Length);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.LastIndexOf7"]/*' />
|
||
public int LastIndexOf(String value, int startIndex) {
|
||
return LastIndexOf(value, startIndex, startIndex + 1);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.LastIndexOf8"]/*' />
|
||
public int LastIndexOf(String value, int startIndex, int count) {
|
||
if (count<0) {
|
||
throw new ArgumentOutOfRangeException("count", "ArgumentOutOfRange_Count");
|
||
}
|
||
return CompareInfo.LastIndexOf(this, value, startIndex, count, CompareOptions.None);
|
||
}
|
||
|
||
|
||
//
|
||
//
|
||
//| <include path='docs/doc[@for="String.PadLeft"]/*' />
|
||
public String PadLeft(int totalWidth) {
|
||
return PadHelper(totalWidth, ' ', false);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.PadLeft1"]/*' />
|
||
public String PadLeft(int totalWidth, char paddingChar) {
|
||
return PadHelper(totalWidth, paddingChar, false);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.PadRight"]/*' />
|
||
public String PadRight(int totalWidth) {
|
||
return PadHelper(totalWidth, ' ', true);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.PadRight1"]/*' />
|
||
public String PadRight(int totalWidth, char paddingChar) {
|
||
return PadHelper(totalWidth, paddingChar, true);
|
||
}
|
||
|
||
private String PadHelper(int totalWidth, char paddingChar,
|
||
bool isRightPadded) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::PadHelper
|
||
if (totalWidth < 0) {
|
||
throw new ArgumentOutOfRangeException("totalWidth is negative");
|
||
}
|
||
if (totalWidth <= this.m_stringLength) {
|
||
return this;
|
||
}
|
||
int padCount = totalWidth - this.m_stringLength;
|
||
String result = FastAllocateString(totalWidth);
|
||
if (isRightPadded) {
|
||
FillString(result, 0, this);
|
||
FillStringChar(result, this.m_stringLength,
|
||
paddingChar, padCount);
|
||
} else {
|
||
|
||
FillStringChar(result, 0, paddingChar, padCount);
|
||
FillString(result, padCount, this);
|
||
}
|
||
return result;
|
||
}
|
||
|
||
// Determines whether a specified string is a prefix of the current instance
|
||
//
|
||
//| <include path='docs/doc[@for="String.StartsWith"]/*' />
|
||
public bool StartsWith(String value) {
|
||
if (null==value) {
|
||
throw new ArgumentNullException("value");
|
||
}
|
||
if (this.Length<value.Length) {
|
||
return false;
|
||
}
|
||
return (0==Compare(this,0, value,0, value.Length));
|
||
}
|
||
|
||
// Creates a copy of this string in lower case.
|
||
//| <include path='docs/doc[@for="String.ToLower"]/*' />
|
||
public String ToLower() {
|
||
return TextInfo.ToLower(this);
|
||
}
|
||
|
||
// Creates a copy of this string in upper case.
|
||
//| <include path='docs/doc[@for="String.ToUpper"]/*' />
|
||
public String ToUpper() {
|
||
return TextInfo.ToUpper(this);
|
||
}
|
||
|
||
// Returns this string.
|
||
//| <include path='docs/doc[@for="String.ToString"]/*' />
|
||
public override String ToString() {
|
||
return this;
|
||
}
|
||
|
||
// Method required for the ICloneable interface.
|
||
// There's no point in cloning a string since they're immutable, so we simply return this.
|
||
//| <include path='docs/doc[@for="String.Clone"]/*' />
|
||
public Object Clone() {
|
||
return this;
|
||
}
|
||
|
||
|
||
// Trims the whitespace from both ends of the string. Whitespace is defined by
|
||
// CharacterInfo.WhitespaceChars.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Trim1"]/*' />
|
||
public String Trim() {
|
||
return this.Trim(CharacterInfo.WhitespaceChars);
|
||
}
|
||
|
||
//
|
||
//
|
||
//| <include path='docs/doc[@for="String.Insert"]/*' />
|
||
public String Insert(int startIndex, String value) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::Insert
|
||
if (startIndex < 0 || startIndex > this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException("startIndex out of range");
|
||
}
|
||
if (value == null) {
|
||
throw new ArgumentNullException("value string is null");
|
||
}
|
||
int newLength = this.m_stringLength + value.m_stringLength;
|
||
String result = FastAllocateString(newLength);
|
||
FillSubstring(result, 0, this, 0, startIndex);
|
||
FillSubstring(result, startIndex, value, 0, value.m_stringLength);
|
||
FillSubstring(result, startIndex + value.m_stringLength,
|
||
this, startIndex, this.m_stringLength - startIndex);
|
||
return result;
|
||
}
|
||
|
||
internal unsafe void InsertHoleInString(int startIndex, int holeSize) {
|
||
int tail = this.m_stringLength - startIndex;
|
||
this.m_stringLength += holeSize;
|
||
if (tail > 0) {
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
char *srcPtr = charPtr + startIndex;
|
||
char *dstPtr = charPtr + holeSize;
|
||
Buffer.MoveMemory((byte *)dstPtr, (byte *)srcPtr,
|
||
tail * sizeof(char));
|
||
}
|
||
}
|
||
}
|
||
|
||
internal unsafe void InsertStringOverwrite(String value, int startIndex, int valueLength) {
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
char *dstPtr = charPtr + startIndex;
|
||
fixed (char *srcPtr = &value.m_firstChar) {
|
||
Buffer.MoveMemory((byte *)dstPtr, (byte *)srcPtr,
|
||
valueLength * sizeof(char));
|
||
}
|
||
}
|
||
}
|
||
|
||
// Replaces all instances of oldChar with newChar.
|
||
//
|
||
//| <include path='docs/doc[@for="String.Replace"]/*' />
|
||
public String Replace(char oldChar, char newChar) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::Replace
|
||
String result = FastAllocateString(this.m_stringLength);
|
||
this.Replace(oldChar, newChar, result);
|
||
return result;
|
||
}
|
||
|
||
private unsafe void Replace(char oldChar, char newChar, String newString) {
|
||
fixed (char *srcPtr = &this.m_firstChar) {
|
||
fixed (char *destPtr = &newString.m_firstChar) {
|
||
char *srcCursor = srcPtr;
|
||
char *destCursor = destPtr;
|
||
for (int i = this.m_stringLength; i > 0; i--) {
|
||
char c = *srcCursor;
|
||
*destCursor = (c == oldChar) ? newChar : c;
|
||
srcCursor++;
|
||
destCursor++;
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
// This method contains the same functionality as StringBuilder Replace. The only difference is that
|
||
// a new String has to be allocated since Strings are immutable
|
||
//| <include path='docs/doc[@for="String.Replace1"]/*' />
|
||
public String Replace(String oldValue, String newValue) {
|
||
// Rusa: see also Lightning\Src\VM\COMString.cpp::ReplaceString
|
||
if (oldValue == null) {
|
||
throw new ArgumentNullException("oldString is null");
|
||
}
|
||
if (newValue == null) {
|
||
newValue = String.Empty;
|
||
}
|
||
if (oldValue.m_stringLength == 0) {
|
||
throw new ArgumentException("oldString is empty string");
|
||
}
|
||
int matchIndex = LocalIndexOfString(oldValue, 0);
|
||
if (matchIndex < 0) {
|
||
return this;
|
||
}
|
||
// Compute a table of where to insert newValue
|
||
int replaceCount = 0;
|
||
int[] replacementTable = new int[(this.m_stringLength - matchIndex) / oldValue.m_stringLength + 1];
|
||
do {
|
||
replacementTable[replaceCount] = matchIndex;
|
||
replaceCount++;
|
||
matchIndex = LocalIndexOfString(oldValue, matchIndex+oldValue.m_stringLength);
|
||
} while (matchIndex >= 0);
|
||
int newLength = this.m_stringLength + replaceCount * (newValue.m_stringLength - oldValue.m_stringLength);
|
||
String result = FastAllocateString(newLength);
|
||
int srcIndex = 0;
|
||
int destIndex = 0;
|
||
for (int replIndex = 0; replIndex < replaceCount; replIndex++) {
|
||
int count = replacementTable[replIndex] - srcIndex;
|
||
FillSubstring(result, destIndex, this, srcIndex, count);
|
||
srcIndex += count + oldValue.m_stringLength;
|
||
destIndex += count;
|
||
FillString(result, destIndex, newValue);
|
||
destIndex += newValue.m_stringLength;
|
||
}
|
||
FillSubstring(result, destIndex, this, srcIndex,
|
||
this.m_stringLength - srcIndex);
|
||
return result;
|
||
}
|
||
|
||
private unsafe int LocalIndexOfString(String pattern, int startPos) {
|
||
fixed (char *stringPtr = &this.m_firstChar) {
|
||
fixed (char *patternPtr = &pattern.m_firstChar) {
|
||
char *stringCharPtr = stringPtr + startPos;
|
||
int count = this.m_stringLength - pattern.m_stringLength -
|
||
startPos + 1;
|
||
for (int index = 0; index < count; index++) {
|
||
char *stringCursor = stringCharPtr + index;
|
||
char *patternCursor = patternPtr;
|
||
int i = pattern.m_stringLength;
|
||
while (i > 0 && *stringCursor == *patternCursor) {
|
||
i--;
|
||
stringCursor++;
|
||
patternCursor++;
|
||
}
|
||
if (i == 0) {
|
||
return (startPos + index);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
return -1;
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Remove"]/*' />
|
||
public String Remove(int startIndex, int count) {
|
||
if (count < 0) {
|
||
throw new ArgumentOutOfRangeException("count is negative");
|
||
}
|
||
if (startIndex < 0) {
|
||
throw new ArgumentOutOfRangeException("startIndex is negative");
|
||
}
|
||
if (startIndex + count > this.m_stringLength) {
|
||
throw new ArgumentOutOfRangeException("startIndex+count>Length");
|
||
}
|
||
int newLength = this.m_stringLength - count;
|
||
String result = FastAllocateString(newLength);
|
||
FillSubstring(result, 0, this, 0, startIndex);
|
||
FillSubstring(result, startIndex, this, startIndex + count,
|
||
newLength - startIndex);
|
||
return result;
|
||
}
|
||
|
||
internal unsafe void RemoveRange(int startIndex, int count) {
|
||
int tail = this.m_stringLength - startIndex;
|
||
this.m_stringLength -= count;
|
||
if (tail > 0) {
|
||
fixed (char *charPtr = &this.m_firstChar) {
|
||
char *dstPtr = charPtr + startIndex;
|
||
char *srcPtr = dstPtr + count;
|
||
Buffer.MoveMemory((byte *)dstPtr, (byte *)srcPtr,
|
||
tail * sizeof(char));
|
||
}
|
||
}
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Format"]/*' />
|
||
public static String Format(String format, Object arg0) {
|
||
return Format(format, new Object[] {arg0});
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Format1"]/*' />
|
||
public static String Format(String format, Object arg0, Object arg1) {
|
||
return Format(format, new Object[] {arg0, arg1});
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Format2"]/*' />
|
||
public static String Format(String format, Object arg0, Object arg1, Object arg2) {
|
||
return Format(format, new Object[] {arg0, arg1, arg2});
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Format3"]/*' />
|
||
public static String Format(String format, params Object[] args) {
|
||
if (format == null || args == null)
|
||
throw new ArgumentNullException((format==null)?"format":"args");
|
||
StringBuilder sb = new StringBuilder(format.Length + args.Length * 8);
|
||
sb.AppendFormat(format,args);
|
||
return sb.ToString();
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Copy"]/*' />
|
||
public static String Copy (String str) {
|
||
if (str==null) {
|
||
throw new ArgumentNullException("str");
|
||
}
|
||
|
||
int length = str.Length;
|
||
|
||
String result = FastAllocateString(length);
|
||
FillString(result, 0, str);
|
||
return result;
|
||
}
|
||
|
||
// Used by StringBuilder to avoid data corruption
|
||
internal static String InternalCopy (String str) {
|
||
int length = str.Length;
|
||
String result = FastAllocateString(length);
|
||
FillStringEx(result, 0, str, length); // The underlying's String can changed length is StringBuilder
|
||
return result;
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Concat"]/*' />
|
||
public static String Concat(Object arg0) {
|
||
if (arg0==null) {
|
||
return String.Empty;
|
||
}
|
||
return arg0.ToString();
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Concat1"]/*' />
|
||
public static String Concat(Object arg0, Object arg1) {
|
||
if (arg0==null) {
|
||
arg0 = String.Empty;
|
||
}
|
||
|
||
if (arg1==null) {
|
||
arg1 = String.Empty;
|
||
}
|
||
return Concat(arg0.ToString(), arg1.ToString());
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Concat2"]/*' />
|
||
public static String Concat(Object arg0, Object arg1, Object arg2) {
|
||
if (arg0==null) {
|
||
arg0 = String.Empty;
|
||
}
|
||
|
||
if (arg1==null) {
|
||
arg1 = String.Empty;
|
||
}
|
||
|
||
if (arg2==null) {
|
||
arg2 = String.Empty;
|
||
}
|
||
|
||
return Concat(arg0.ToString(), arg1.ToString(), arg2.ToString());
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Concat4"]/*' />
|
||
public static String Concat(params Object[] args) {
|
||
if (args==null) {
|
||
throw new ArgumentNullException("args");
|
||
}
|
||
|
||
String[] sArgs = new String[args.Length];
|
||
int totalLength=0;
|
||
|
||
for (int i=0; i<args.Length; i++) {
|
||
object value = args[i];
|
||
sArgs[i] = ((value==null)?(String.Empty):(value.ToString()));
|
||
totalLength += sArgs[i].Length;
|
||
// check for overflow
|
||
if (totalLength < 0) {
|
||
throw new OutOfMemoryException();
|
||
}
|
||
}
|
||
return ConcatArray(sArgs, totalLength);
|
||
}
|
||
|
||
|
||
//| <include path='docs/doc[@for="String.Concat5"]/*' />
|
||
public static String Concat(String str0, String str1) {
|
||
if (str0 == null) {
|
||
if (str1==null) {
|
||
return String.Empty;
|
||
}
|
||
return str1;
|
||
}
|
||
|
||
if (str1==null) {
|
||
return str0;
|
||
}
|
||
|
||
int str0Length = str0.Length;
|
||
|
||
String result = FastAllocateString(str0Length + str1.Length);
|
||
|
||
FillStringChecked(result, 0, str0);
|
||
FillStringChecked(result, str0Length, str1);
|
||
|
||
return result;
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Concat6"]/*' />
|
||
public static String Concat(String str0, String str1, String str2) {
|
||
if (str0==null && str1==null && str2==null) {
|
||
return String.Empty;
|
||
}
|
||
|
||
if (str0==null) {
|
||
str0 = String.Empty;
|
||
}
|
||
|
||
if (str1==null) {
|
||
str1 = String.Empty;
|
||
}
|
||
|
||
if (str2 == null) {
|
||
str2 = String.Empty;
|
||
}
|
||
|
||
int totalLength = str0.Length + str1.Length + str2.Length;
|
||
|
||
String result = FastAllocateString(totalLength);
|
||
FillStringChecked(result, 0, str0);
|
||
FillStringChecked(result, str0.Length, str1);
|
||
FillStringChecked(result, str0.Length + str1.Length, str2);
|
||
|
||
return result;
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Concat7"]/*' />
|
||
public static String Concat(String str0, String str1, String str2, String str3) {
|
||
if (str0==null && str1==null && str2==null && str3==null) {
|
||
return String.Empty;
|
||
}
|
||
|
||
if (str0==null) {
|
||
str0 = String.Empty;
|
||
}
|
||
|
||
if (str1==null) {
|
||
str1 = String.Empty;
|
||
}
|
||
|
||
if (str2 == null) {
|
||
str2 = String.Empty;
|
||
}
|
||
|
||
if (str3 == null) {
|
||
str3 = String.Empty;
|
||
}
|
||
|
||
int totalLength = str0.Length + str1.Length + str2.Length + str3.Length;
|
||
|
||
String result = FastAllocateString(totalLength);
|
||
FillStringChecked(result, 0, str0);
|
||
FillStringChecked(result, str0.Length, str1);
|
||
FillStringChecked(result, str0.Length + str1.Length, str2);
|
||
FillStringChecked(result, str0.Length + str1.Length + str2.Length, str3);
|
||
|
||
return result;
|
||
}
|
||
|
||
private static String ConcatArray(String[] values, int totalLength) {
|
||
String result = FastAllocateString(totalLength);
|
||
int currPos=0;
|
||
|
||
for (int i=0; i<values.Length; i++) {
|
||
Debug.Assert((currPos + values[i].Length <= totalLength),
|
||
"[String.ConcatArray](currPos + values[i].Length <= totalLength)");
|
||
|
||
FillStringChecked(result, currPos, values[i]);
|
||
currPos+=values[i].Length;
|
||
}
|
||
|
||
return result;
|
||
}
|
||
|
||
// poor man's varargs for String.Concat: 8 strings
|
||
public static String Concat(object obj0, object obj1, object obj2,
|
||
object obj3, string obj4, string obj5,
|
||
string obj6, string obj7) {
|
||
String[] strs = new String[8];
|
||
strs[0] = obj0.ToString();
|
||
strs[1] = obj1.ToString();
|
||
strs[2] = obj2.ToString();
|
||
strs[3] = obj3.ToString();
|
||
strs[4] = obj4.ToString();
|
||
strs[5] = obj5.ToString();
|
||
strs[6] = obj6.ToString();
|
||
strs[7] = obj7.ToString();
|
||
return Join(null,strs,0,8);
|
||
}
|
||
|
||
private static String[] CopyArrayOnNull(String[] values, out int totalLength) {
|
||
totalLength = 0;
|
||
|
||
String[] outValues = new String[values.Length];
|
||
|
||
for (int i=0; i<values.Length; i++) {
|
||
outValues[i] = ((values[i]==null)?String.Empty:values[i]);
|
||
totalLength += outValues[i].Length;
|
||
}
|
||
return outValues;
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Concat8"]/*' />
|
||
public static String Concat(params String[] values) {
|
||
int totalLength=0;
|
||
|
||
if (values==null) {
|
||
throw new ArgumentNullException("values");
|
||
}
|
||
|
||
// Always make a copy to prevent changing the array on another thread.
|
||
String[] internalValues = new String[values.Length];
|
||
|
||
for (int i = 0; i< values.Length; i++) {
|
||
string value = values[i];
|
||
internalValues[i] = ((value==null)?(String.Empty):(value));
|
||
totalLength += internalValues[i].Length;
|
||
// check for overflow
|
||
if (totalLength < 0) {
|
||
throw new OutOfMemoryException();
|
||
}
|
||
}
|
||
|
||
return ConcatArray(internalValues, totalLength);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.Intern"]/*' />
|
||
public static String Intern(String str) {
|
||
str.CheckHighChars();
|
||
if (str == null) {
|
||
throw new ArgumentNullException("str");
|
||
}
|
||
if (internedStringVector == null) {
|
||
InitializeInternedTable();
|
||
}
|
||
int code = str.GetHashCode();
|
||
lock (internedStringVector) {
|
||
int[] codeTable= internedStringHashCodes;
|
||
String[] stringTable = internedStringVector;
|
||
int tableSize = codeTable.Length;
|
||
int indexMask = tableSize - 1;
|
||
int hx = code & indexMask;
|
||
//Scan up from our initial hash index guess
|
||
while (true) { // It is guaranteed there are holes in table
|
||
int tableCode = codeTable[hx];
|
||
if (tableCode == code) {
|
||
String s = stringTable[hx];
|
||
if (str.Equals(s)) {
|
||
return s;
|
||
}
|
||
} else if (tableCode == 0 && stringTable[hx] == null) {
|
||
// We need to insert the string here!
|
||
int stringLength = str.Length;
|
||
if (str.m_arrayLength > stringLength+1) {
|
||
str = NewString(str, 0, stringLength, stringLength);
|
||
}
|
||
internedStringCount++;
|
||
stringTable[hx] = str; // Set string entry first
|
||
codeTable[hx] = code; // then set code!
|
||
if (internedStringCount >= internedStringBound) {
|
||
RehashInternedStrings();
|
||
}
|
||
return str;
|
||
}
|
||
hx++;
|
||
hx &= indexMask;
|
||
}
|
||
}
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.IsInterned"]/*' />
|
||
public static String IsInterned(String str) {
|
||
if (str == null) {
|
||
throw new ArgumentNullException("str");
|
||
}
|
||
if (internedStringVector == null) {
|
||
InitializeInternedTable();
|
||
}
|
||
int code = str.GetHashCode();
|
||
int[] codeTable;
|
||
String[] stringTable;
|
||
do {
|
||
codeTable = internedStringHashCodes;
|
||
stringTable = internedStringVector;
|
||
} while (codeTable.Length != stringTable.Length);
|
||
int tableSize = codeTable.Length;
|
||
int indexMask = tableSize - 1;
|
||
int hx = code & indexMask;
|
||
// Scan up from our initial hash index guess
|
||
while (true) { // It is guaranteed there are holes in table
|
||
int tableCode = codeTable[hx];
|
||
if (tableCode == code) {
|
||
String s = stringTable[hx];
|
||
if (str.Equals(s)) {
|
||
return s;
|
||
}
|
||
} else if (tableCode == 0 && stringTable[hx] == null) {
|
||
return null;
|
||
}
|
||
hx++;
|
||
hx &= indexMask;
|
||
}
|
||
}
|
||
|
||
private static void InitializeInternedTable() {
|
||
lock (typeof(String)) {
|
||
if (internedStringVector != null) {
|
||
return;
|
||
}
|
||
int tableSize = 128;
|
||
int constantCount = internedStringConstants.Length;
|
||
while (tableSize <= constantCount) {
|
||
tableSize <<= 1;
|
||
}
|
||
int indexMask = tableSize - 1;
|
||
String[] stringTable = new String[tableSize];
|
||
int[] codeTable = new int[tableSize];
|
||
for (int i = 0; i < constantCount; i++) {
|
||
String s = internedStringConstants[i];
|
||
int code = s.GetHashCode();
|
||
int hx = code & indexMask;
|
||
while (true) {
|
||
int tableCode = codeTable[hx];
|
||
VTable.Assert(tableCode != code ||
|
||
!s.Equals(stringTable[hx]),
|
||
"Duplicate interned strings!");
|
||
if (tableCode == 0 && stringTable[hx] == null) {
|
||
internedStringCount++;
|
||
stringTable[hx] = s;
|
||
codeTable[hx] = code;
|
||
break;
|
||
}
|
||
hx++;
|
||
hx &= indexMask;
|
||
}
|
||
}
|
||
internedStringBound = tableSize >> 1;
|
||
internedStringHashCodes = codeTable;
|
||
internedStringVector = stringTable;
|
||
}
|
||
}
|
||
|
||
private static void RehashInternedStrings() {
|
||
int[] oldCodeTable = internedStringHashCodes;
|
||
String[] oldStringTable = internedStringVector;
|
||
int oldSize = oldCodeTable.Length;
|
||
int oldIndexMask = oldSize - 1;
|
||
int newSize = oldSize << 1;
|
||
int newIndexMask = newSize -1;
|
||
int[] newCodeTable = new int[newSize];
|
||
String[] newStringTable = new String[newSize];
|
||
for (int i = 0; i < oldSize; i++) {
|
||
String s = oldStringTable[i];
|
||
if (s != null) {
|
||
int code = oldCodeTable[i];
|
||
int hx = code & newIndexMask;
|
||
while (true) {
|
||
if (newCodeTable[hx] == 0 &&
|
||
newStringTable[hx] == null) {
|
||
newStringTable[hx] = s;
|
||
newCodeTable[hx] = code;
|
||
break;
|
||
}
|
||
hx++;
|
||
hx &= newIndexMask;
|
||
}
|
||
}
|
||
}
|
||
internedStringBound = newSize >> 1;
|
||
internedStringHashCodes = newCodeTable;
|
||
internedStringVector = newStringTable;
|
||
}
|
||
|
||
private static int internedStringCount;
|
||
private static int internedStringBound;
|
||
private static int[] internedStringHashCodes;
|
||
private static String[] internedStringVector;
|
||
private static String[] internedStringConstants;
|
||
|
||
private static readonly bool [] HighCharTable = {
|
||
false, // 0x00, 0x0
|
||
true, // 0x01, .
|
||
true, // 0x02, .
|
||
true, // 0x03, .
|
||
true, // 0x04, .
|
||
true, // 0x05, .
|
||
true, // 0x06, .
|
||
true, // 0x07, .
|
||
true, // 0x08, .
|
||
false, // 0x09,
|
||
false, // 0x0A,
|
||
false, // 0x0B, .
|
||
false, // 0x0C, .
|
||
false, // 0x0D,
|
||
true, // 0x0E, .
|
||
true, // 0x0F, .
|
||
true, // 0x10, .
|
||
true, // 0x11, .
|
||
true, // 0x12, .
|
||
true, // 0x13, .
|
||
true, // 0x14, .
|
||
true, // 0x15, .
|
||
true, // 0x16, .
|
||
true, // 0x17, .
|
||
true, // 0x18, .
|
||
true, // 0x19, .
|
||
true, // 0x1A,
|
||
true, // 0x1B, .
|
||
true, // 0x1C, .
|
||
true, // 0x1D, .
|
||
true, // 0x1E, .
|
||
true, // 0x1F, .
|
||
false, // 0x20,
|
||
false, // 0x21, !
|
||
false, // 0x22, "
|
||
false, // 0x23, #
|
||
false, // 0x24, $
|
||
false, // 0x25, %
|
||
false, // 0x26, &
|
||
true, // 0x27, '
|
||
false, // 0x28, (
|
||
false, // 0x29, )
|
||
false, // 0x2A *
|
||
false, // 0x2B, +
|
||
false, // 0x2C, ,
|
||
true, // 0x2D, -
|
||
false, // 0x2E, .
|
||
false, // 0x2F, /
|
||
false, // 0x30, 0
|
||
false, // 0x31, 1
|
||
false, // 0x32, 2
|
||
false, // 0x33, 3
|
||
false, // 0x34, 4
|
||
false, // 0x35, 5
|
||
false, // 0x36, 6
|
||
false, // 0x37, 7
|
||
false, // 0x38, 8
|
||
false, // 0x39, 9
|
||
false, // 0x3A, :
|
||
false, // 0x3B, ;
|
||
false, // 0x3C, <
|
||
false, // 0x3D, =
|
||
false, // 0x3E, >
|
||
false, // 0x3F, ?
|
||
false, // 0x40, @
|
||
false, // 0x41, A
|
||
false, // 0x42, B
|
||
false, // 0x43, C
|
||
false, // 0x44, D
|
||
false, // 0x45, E
|
||
false, // 0x46, F
|
||
false, // 0x47, G
|
||
false, // 0x48, H
|
||
false, // 0x49, I
|
||
false, // 0x4A, J
|
||
false, // 0x4B, K
|
||
false, // 0x4C, L
|
||
false, // 0x4D, M
|
||
false, // 0x4E, N
|
||
false, // 0x4F, O
|
||
false, // 0x50, P
|
||
false, // 0x51, Q
|
||
false, // 0x52, R
|
||
false, // 0x53, S
|
||
false, // 0x54, T
|
||
false, // 0x55, U
|
||
false, // 0x56, V
|
||
false, // 0x57, W
|
||
false, // 0x58, X
|
||
false, // 0x59, Y
|
||
false, // 0x5A, Z
|
||
false, // 0x5B, [
|
||
false, // 0x5C, \
|
||
false, // 0x5D, ]
|
||
false, // 0x5E, ^
|
||
false, // 0x5F, _
|
||
false, // 0x60, `
|
||
false, // 0x61, a
|
||
false, // 0x62, b
|
||
false, // 0x63, c
|
||
false, // 0x64, d
|
||
false, // 0x65, e
|
||
false, // 0x66, f
|
||
false, // 0x67, g
|
||
false, // 0x68, h
|
||
false, // 0x69, i
|
||
false, // 0x6A, j
|
||
false, // 0x6B, k
|
||
false, // 0x6C, l
|
||
false, // 0x6D, m
|
||
false, // 0x6E, n
|
||
false, // 0x6F, o
|
||
false, // 0x70, p
|
||
false, // 0x71, q
|
||
false, // 0x72, r
|
||
false, // 0x73, s
|
||
false, // 0x74, t
|
||
false, // 0x75, u
|
||
false, // 0x76, v
|
||
false, // 0x77, w
|
||
false, // 0x78, x
|
||
false, // 0x79, y
|
||
false, // 0x7A, z
|
||
false, // 0x7B, {
|
||
false, // 0x7C, |
|
||
false, // 0x7D, }
|
||
false, // 0x7E, ~
|
||
true, // 0x7F,
|
||
};
|
||
|
||
//
|
||
// IValue implementation
|
||
//
|
||
|
||
//| <include path='docs/doc[@for="String.GetTypeCode"]/*' />
|
||
[NoHeapAllocation]
|
||
public override TypeCode GetTypeCode() {
|
||
return TypeCode.String;
|
||
}
|
||
|
||
// Is this a string that can be compared quickly (that is it has only characters > 0x80
|
||
// and not a - or '
|
||
internal bool IsFastSort() {
|
||
return ((StringState & StringState.SpecialSort) != 0);
|
||
}
|
||
|
||
internal bool IsSlowSort() {
|
||
return !IsFastSort();
|
||
}
|
||
|
||
internal bool IsFastOpsExceptSort() {
|
||
return ((StringState & StringState.SpecialSort) != 0 ||
|
||
(StringState & StringState.FastOps) != 0);
|
||
}
|
||
|
||
internal bool IsFastCasing() {
|
||
return ((StringState & StringState.SpecialSort) != 0 ||
|
||
(StringState & StringState.FastOps) != 0);
|
||
}
|
||
|
||
internal bool IsFastIndex() {
|
||
return ((StringState & StringState.SpecialSort) != 0 ||
|
||
(StringState & StringState.FastOps) != 0);
|
||
}
|
||
|
||
internal bool IsStringStateUndetermined() {
|
||
return (StringState == StringState.Undetermined);
|
||
}
|
||
|
||
internal bool HasHighChars() {
|
||
return (StringState == StringState.HighChars);
|
||
}
|
||
|
||
|
||
/// <summary>
|
||
/// Get or set the string state for this string.
|
||
/// </summary>
|
||
internal unsafe StringState StringState {
|
||
[NoLoggingForUndo]
|
||
get {
|
||
StringState result = MultiUseWord.GetStringState(this);
|
||
return (StringState)result;
|
||
}
|
||
[NoLoggingForUndo]
|
||
set {
|
||
MultiUseWord.SetStringState(this, value);
|
||
}
|
||
}
|
||
|
||
internal unsafe StringState CheckHighChars() {
|
||
char c;
|
||
StringState ss = StringState.FastOps;
|
||
int length = m_stringLength;
|
||
fixed(char *charPtr = &this.m_firstChar) {
|
||
for(int i = 0; i < length; i++) {
|
||
c = charPtr[i];
|
||
if(c>=0x80) {
|
||
StringState = StringState.HighChars;
|
||
return StringState;
|
||
}
|
||
else if (HighCharTable[(int)c]) {
|
||
ss = StringState.SpecialSort;
|
||
}
|
||
}
|
||
}
|
||
|
||
StringState = ss;
|
||
return StringState;
|
||
}
|
||
|
||
///<internalonly/>
|
||
unsafe internal void SetChar(int index, char value)
|
||
{
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
|
||
//Bounds check and then set the actual bit.
|
||
if ((UInt32)index >= (UInt32)Length)
|
||
throw new ArgumentOutOfRangeException("index", "ArgumentOutOfRange_Index");
|
||
|
||
fixed (char *p = &this.m_firstChar) {
|
||
// Set the character.
|
||
p[index] = value;
|
||
}
|
||
}
|
||
|
||
#if _DEBUG
|
||
// Only used in debug build. Insure that the HighChar state information for a string is not set as known
|
||
[MethodImpl(MethodImplOptions.InternalCall)]
|
||
private bool ValidModifiableString() {
|
||
throw new Exception("System.String.ValidModifiableString not implemented in Bartok!");
|
||
}
|
||
#endif
|
||
|
||
///<internalonly/>
|
||
unsafe internal void AppendInPlace(char value,int currentLength)
|
||
{
|
||
Debug.Assert(currentLength < m_arrayLength, "[String.AppendInPlace(char)currentLength < m_arrayLength");
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
|
||
fixed (char *p = &this.m_firstChar)
|
||
{
|
||
// Append the character.
|
||
p[currentLength] = value;
|
||
p[++currentLength] = '\0';
|
||
m_stringLength = currentLength;
|
||
}
|
||
}
|
||
|
||
|
||
///<internalonly/>
|
||
unsafe internal void AppendInPlace(char value, int repeatCount, int currentLength)
|
||
{
|
||
Debug.Assert(currentLength+repeatCount < m_arrayLength, "[String.AppendInPlace]currentLength+repeatCount < m_arrayLength");
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
|
||
int newLength = currentLength + repeatCount;
|
||
|
||
|
||
fixed (char *p = &this.m_firstChar)
|
||
{
|
||
int i;
|
||
for (i=currentLength; i<newLength; i++) {
|
||
p[i] = value;
|
||
}
|
||
p[i] = '\0';
|
||
}
|
||
this.m_stringLength = newLength;
|
||
}
|
||
|
||
///<internalonly/>
|
||
internal unsafe void AppendInPlace(String value, int currentLength) {
|
||
Debug.Assert(value!=null, "[String.AppendInPlace]value!=null");
|
||
Debug.Assert(value.Length + currentLength < this.m_arrayLength, "[String.AppendInPlace]Length is wrong.");
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
|
||
FillString(this, currentLength, value);
|
||
SetLength(currentLength + value.Length);
|
||
NullTerminate();
|
||
}
|
||
|
||
internal void AppendInPlace(String value, int startIndex, int count, int currentLength) {
|
||
Debug.Assert(value!=null, "[String.AppendInPlace]value!=null");
|
||
Debug.Assert(count + currentLength < this.m_arrayLength, "[String.AppendInPlace]count + currentLength < this.m_arrayLength");
|
||
Debug.Assert(count>=0, "[String.AppendInPlace]count>=0");
|
||
Debug.Assert(startIndex>=0, "[String.AppendInPlace]startIndex>=0");
|
||
Debug.Assert(startIndex <= (value.Length - count), "[String.AppendInPlace]startIndex <= (value.Length - count)");
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
|
||
FillSubstring(this, currentLength, value, startIndex, count);
|
||
SetLength(currentLength + count);
|
||
NullTerminate();
|
||
}
|
||
|
||
internal unsafe void AppendInPlace(char *value, int count,int currentLength) {
|
||
Debug.Assert(value!=null, "[String.AppendInPlace]value!=null");
|
||
Debug.Assert(count + currentLength < this.m_arrayLength, "[String.AppendInPlace]count + currentLength < this.m_arrayLength");
|
||
Debug.Assert(count>=0, "[String.AppendInPlace]count>=0");
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
fixed(char *p = &this.m_firstChar) {
|
||
int i;
|
||
for (i=0; i<count; i++) {
|
||
p[currentLength+i] = value[i];
|
||
}
|
||
}
|
||
SetLength(currentLength + count);
|
||
NullTerminate();
|
||
}
|
||
|
||
|
||
///<internalonly/>
|
||
internal unsafe void AppendInPlace(char[] value, int start, int count, int currentLength) {
|
||
Debug.Assert(value!=null, "[String.AppendInPlace]value!=null");
|
||
Debug.Assert(count + currentLength < this.m_arrayLength, "[String.AppendInPlace]Length is wrong.");
|
||
Debug.Assert(value.Length-count>=start, "[String.AppendInPlace]value.Length-count>=start");
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
|
||
FillStringArray(this, currentLength, value, start, count);
|
||
this.m_stringLength = (currentLength + count);
|
||
this.NullTerminate();
|
||
}
|
||
|
||
|
||
///<internalonly/>
|
||
unsafe internal void ReplaceCharInPlace(char oldChar, char newChar, int startIndex, int count,int currentLength) {
|
||
Debug.Assert(startIndex>=0, "[String.ReplaceCharInPlace]startIndex>0");
|
||
Debug.Assert(startIndex<=currentLength, "[String.ReplaceCharInPlace]startIndex>=Length");
|
||
Debug.Assert((startIndex<=currentLength-count), "[String.ReplaceCharInPlace]count>0 && startIndex<=currentLength-count");
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
|
||
int endIndex = startIndex+count;
|
||
|
||
fixed (char *p = &this.m_firstChar) {
|
||
for (int i=startIndex;i<endIndex; i++) {
|
||
if (p[i]==oldChar) {
|
||
p[i]=newChar;
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
|
||
///<internalonly/>
|
||
internal static String GetStringForStringBuilder(String value, int capacity) {
|
||
Debug.Assert(value!=null, "[String.GetStringForStringBuilder]value!=null");
|
||
Debug.Assert(capacity>=value.Length, "[String.GetStringForStringBuilder]capacity>=value.Length");
|
||
|
||
String newStr = FastAllocateString(capacity);
|
||
if (value.Length==0) {
|
||
newStr.m_stringLength=0;
|
||
newStr.m_firstChar='\0';
|
||
return newStr;
|
||
}
|
||
FillString(newStr, 0, value);
|
||
newStr.m_stringLength = value.m_stringLength;
|
||
return newStr;
|
||
}
|
||
|
||
///<internalonly/>
|
||
private unsafe void NullTerminate() {
|
||
fixed(char *p = &this.m_firstChar) {
|
||
p[m_stringLength] = '\0';
|
||
}
|
||
}
|
||
|
||
///<internalonly/>
|
||
unsafe internal void ClearPostNullChar() {
|
||
int newLength = Length+1;
|
||
if (newLength<m_arrayLength) {
|
||
fixed(char *p = &this.m_firstChar) {
|
||
p[newLength] = '\0';
|
||
}
|
||
}
|
||
}
|
||
|
||
///<internalonly/>
|
||
internal void SetLength(int newLength) {
|
||
Debug.Assert(newLength <= m_arrayLength, "newLength<=m_arrayLength");
|
||
m_stringLength = newLength;
|
||
}
|
||
|
||
|
||
|
||
//| <include path='docs/doc[@for="String.GetEnumerator"]/*' />
|
||
public CharEnumerator GetEnumerator() {
|
||
return new CharEnumerator(this);
|
||
}
|
||
|
||
//| <include path='docs/doc[@for="String.IEnumerable.GetEnumerator"]/*' />
|
||
/// <internalonly/>
|
||
IEnumerator IEnumerable.GetEnumerator() {
|
||
return new CharEnumerator(this);
|
||
}
|
||
|
||
//
|
||
// This is just designed to prevent compiler warnings.
|
||
// This field is used from native, but we need to prevent the compiler warnings.
|
||
//
|
||
#if _DEBUG
|
||
private void DontTouchThis() {
|
||
m_arrayLength = 0;
|
||
m_stringLength = 0;
|
||
m_firstChar = m_firstChar;
|
||
}
|
||
#endif
|
||
|
||
internal unsafe void InternalSetCharNoBoundsCheck(int index, char value) {
|
||
fixed (char *p = &this.m_firstChar) {
|
||
p[index] = value;
|
||
}
|
||
}
|
||
|
||
|
||
#if false // Unused
|
||
|
||
// NOTE: len is not the length in characters, but the length in bytes
|
||
|
||
// Copies the source String (byte buffer) to the destination IntPtr memory allocated with len bytes.
|
||
internal unsafe static void InternalCopy(String src, IntPtr dest,int len)
|
||
{
|
||
if (len == 0)
|
||
return;
|
||
fixed(char* charPtr = &src.m_firstChar) {
|
||
byte* srcPtr = (byte*) charPtr;
|
||
byte* dstPtr = (byte*) dest.ToPointer();
|
||
Buffer.MoveMemory(dstPtr, srcPtr, len);
|
||
}
|
||
}
|
||
|
||
// memcopies characters inside a String.
|
||
internal unsafe static void InternalMemCpy(String src, int srcOffset, String dst, int destOffset, int len)
|
||
{
|
||
if (len == 0)
|
||
return;
|
||
fixed(char* srcPtr = &src.m_firstChar) {
|
||
fixed(char* dstPtr = &dst.m_firstChar) {
|
||
Buffer.MoveMemory((byte*)(dstPtr + destOffset),
|
||
(byte*)(srcPtr + srcOffset),
|
||
len);
|
||
}
|
||
}
|
||
}
|
||
#endif
|
||
|
||
// Copies the source String to the destination byte[]
|
||
internal unsafe static void InternalCopy(String src, byte[] dest, int stringLength)
|
||
{
|
||
if (stringLength == 0)
|
||
return;
|
||
int len = stringLength * sizeof(char);
|
||
|
||
Debug.Assert(dest.Length >= len);
|
||
|
||
fixed(char* charPtr = &src.m_firstChar) {
|
||
fixed(byte* destPtr = &dest[0]) {
|
||
byte* srcPtr = (byte*) charPtr;
|
||
Buffer.MoveMemory(destPtr, srcPtr, len);
|
||
}
|
||
}
|
||
}
|
||
|
||
internal unsafe void InsertInPlace(int index, String value, int repeatCount, int currentLength, int requiredLength) {
|
||
Debug.Assert(requiredLength < m_arrayLength, "[String.InsertString] requiredLength < m_arrayLength");
|
||
Debug.Assert(index + value.Length * repeatCount < m_arrayLength, "[String.InsertString] index + value.Length * repeatCount < m_arrayLength");
|
||
#if _DEBUG
|
||
Debug.Assert(ValidModifiableString(), "Modifiable string must not have highChars flags set");
|
||
#endif
|
||
//Copy the old characters over to make room and then insert the new characters.
|
||
fixed(char* srcPtr = &this.m_firstChar) {
|
||
fixed(char* valuePtr = &value.m_firstChar) {
|
||
Buffer.MoveMemory((byte*)(srcPtr + index + value.Length * repeatCount),
|
||
(byte*)(srcPtr + index),
|
||
(currentLength - index) * sizeof(char));
|
||
for (int i = 0; i < repeatCount; i++) {
|
||
Buffer.MoveMemory((byte*)(srcPtr + index + i * value.Length),
|
||
(byte*)valuePtr,
|
||
value.Length * sizeof(char));
|
||
}
|
||
}
|
||
srcPtr[requiredLength] = '\0';
|
||
}
|
||
this.m_stringLength = requiredLength;
|
||
}
|
||
|
||
[AccessedByRuntime("referenced from c++")]
|
||
[CLSCompliant(false)]
|
||
[NoHeapAllocation]
|
||
public static unsafe int LimitedFormatTo(String format,
|
||
ArgIterator args,
|
||
char *buffer,
|
||
int length)
|
||
{
|
||
unchecked {
|
||
|
||
fixed (char *fmtBeg = &format.m_firstChar) {
|
||
char *fmtPtr = fmtBeg;
|
||
char *fmtEnd = fmtBeg + format.m_stringLength;
|
||
|
||
char *outPtr = buffer;
|
||
char *outEnd = buffer + length;
|
||
|
||
while (fmtPtr < fmtEnd && outPtr < outEnd) {
|
||
if (*fmtPtr == '{') {
|
||
char * fmtPos = fmtPtr;
|
||
bool bad = false;
|
||
fmtPtr++;
|
||
|
||
int ndx = 0;
|
||
int aln = 0;
|
||
int wid = 0;
|
||
char fmt = 'd';
|
||
|
||
if (*fmtPtr == '{') {
|
||
*outPtr++ = *fmtPtr++;
|
||
}
|
||
else if (*fmtPtr >= '0' && *fmtPtr <= '9') {
|
||
|
||
// {index,alignment:type width}
|
||
// Get Index
|
||
while (*fmtPtr >= '0' && *fmtPtr <= '9') {
|
||
ndx = ndx * 10 + (*fmtPtr++ - '0');
|
||
}
|
||
|
||
// Get Alignment
|
||
if (*fmtPtr == ',') {
|
||
fmtPtr++;
|
||
if (*fmtPtr == '-') {
|
||
// We just ignore left alignment for now.
|
||
fmtPtr++;
|
||
}
|
||
while (*fmtPtr >= '0' && *fmtPtr <= '9') {
|
||
aln = aln * 10 + (*fmtPtr++ - '0');
|
||
}
|
||
}
|
||
|
||
// Get FormatString
|
||
if (*fmtPtr == ':') {
|
||
fmtPtr++;
|
||
if (*fmtPtr >= 'a' && *fmtPtr <= 'z') {
|
||
fmt = *fmtPtr++;
|
||
}
|
||
else if (*fmtPtr >= 'A' && *fmtPtr <= 'Z') {
|
||
fmt = (char)(*fmtPtr++ - ('A' + 'a'));
|
||
}
|
||
while (*fmtPtr >= '0' && *fmtPtr <= '9') {
|
||
wid = wid * 10 + (*fmtPtr++ - '0');
|
||
}
|
||
}
|
||
|
||
// Get closing brace.
|
||
if (*fmtPtr == '}') {
|
||
fmtPtr++;
|
||
}
|
||
else {
|
||
bad = true;
|
||
}
|
||
|
||
if (ndx >= args.Length) {
|
||
bad = true;
|
||
}
|
||
|
||
if (bad) {
|
||
for (; fmtPos < fmtPtr; fmtPos++) {
|
||
if (outPtr < outEnd) {
|
||
*outPtr++ = *fmtPos;
|
||
}
|
||
}
|
||
}
|
||
else {
|
||
// Get the value
|
||
char cvalue = '\0';
|
||
long ivalue = 0;
|
||
string svalue = null;
|
||
ulong value = 0;
|
||
IntPtr pvalue;
|
||
RuntimeType type;
|
||
|
||
type = args.GetArg(ndx, out pvalue);
|
||
|
||
switch (type.classVtable.structuralView) {
|
||
case StructuralType.Bool:
|
||
svalue =*(bool *)pvalue ? "true" : "false";
|
||
break;
|
||
case StructuralType.Char:
|
||
cvalue = *(char *)pvalue;
|
||
break;
|
||
case StructuralType.Int8:
|
||
ivalue = *(sbyte *)pvalue;
|
||
break;
|
||
case StructuralType.Int16:
|
||
ivalue = *(short *)pvalue;
|
||
break;
|
||
case StructuralType.Int32:
|
||
ivalue = *(int *)pvalue;
|
||
break;
|
||
case StructuralType.Int64:
|
||
ivalue = *(long *)pvalue;
|
||
break;
|
||
case StructuralType.UInt8:
|
||
value = *(byte *)pvalue;
|
||
break;
|
||
case StructuralType.UInt16:
|
||
value = *(ushort *)pvalue;
|
||
break;
|
||
case StructuralType.UInt32:
|
||
value = *(uint *)pvalue;
|
||
break;
|
||
case StructuralType.UInt64:
|
||
value = *(ulong *)pvalue;
|
||
break;
|
||
case StructuralType.IntPtr:
|
||
value = (ulong)*(IntPtr *)pvalue;
|
||
break;
|
||
case StructuralType.UIntPtr:
|
||
value = (ulong)*(UIntPtr *)pvalue;
|
||
break;
|
||
case StructuralType.Reference:
|
||
if (type == typeof(String)) {
|
||
svalue = Magic.toString(
|
||
Magic.fromAddress(*(UIntPtr *)pvalue));
|
||
}
|
||
else {
|
||
svalue = type.Name;
|
||
}
|
||
break;
|
||
default:
|
||
svalue = "???";
|
||
break;
|
||
}
|
||
|
||
if (cvalue != '\0') {
|
||
outPtr = AddChar(outPtr, outEnd, *(char *)pvalue, aln);
|
||
}
|
||
else if (svalue != null) {
|
||
outPtr = AddString(outPtr, outEnd, svalue, aln);
|
||
}
|
||
else {
|
||
if (aln < wid) {
|
||
aln = wid;
|
||
}
|
||
|
||
if (fmt == 'x') {
|
||
if (ivalue != 0) {
|
||
value = (ulong)ivalue;
|
||
}
|
||
outPtr = AddNumber(outPtr, outEnd, value, aln, 16, '0', '\0');
|
||
}
|
||
else {
|
||
char sign = '\0';
|
||
if (ivalue < 0) {
|
||
sign = '-';
|
||
value = unchecked((ulong)-ivalue);
|
||
}
|
||
else if (ivalue > 0) {
|
||
value = unchecked((ulong)ivalue);
|
||
}
|
||
outPtr = AddNumber(outPtr, outEnd, value, aln, 10, ' ', sign);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
}
|
||
else if (*fmtPtr == '}') {
|
||
if (outPtr < outEnd) {
|
||
*outPtr++ = *fmtPtr;
|
||
}
|
||
fmtPtr++;
|
||
}
|
||
else {
|
||
if (outPtr < outEnd) {
|
||
*outPtr++ = *fmtPtr;
|
||
}
|
||
fmtPtr++;
|
||
}
|
||
}
|
||
return (int)(outPtr - buffer);
|
||
}
|
||
}
|
||
}
|
||
|
||
[NoHeapAllocation]
|
||
private static int CountDigits(ulong value, uint valueBase)
|
||
{
|
||
int used = 0;
|
||
do {
|
||
value /= valueBase;
|
||
used++;
|
||
} while (value != 0);
|
||
|
||
return used;
|
||
}
|
||
|
||
[NoHeapAllocation]
|
||
private static unsafe char * AddNumber(char *output, char *limit,
|
||
ulong value, int width, uint valueBase,
|
||
char fill, char sign)
|
||
{
|
||
// Make sure we won't overfill the buffer.
|
||
if (output >= limit) {
|
||
return output;
|
||
}
|
||
|
||
// Figure out how wide the string will be.
|
||
int used = CountDigits(value, valueBase);
|
||
|
||
// Check for overflow.
|
||
if (output + used + ((sign != '\0') ? 1 : 0) > limit) {
|
||
while (width > 0) {
|
||
*output++ = '#';
|
||
width--;
|
||
}
|
||
return output;
|
||
}
|
||
|
||
// Handle sign and space fill.
|
||
if (sign != '\0' && used < width) {
|
||
if (fill == ' ') {
|
||
while (width > used + 1) {
|
||
*output++ = fill;
|
||
width--;
|
||
}
|
||
fill = '\0';
|
||
}
|
||
*output++ = sign;
|
||
}
|
||
|
||
// Handle other non-zero fills.
|
||
if (fill != '\0') {
|
||
while (width > used) {
|
||
*output++ = fill;
|
||
width--;
|
||
}
|
||
}
|
||
|
||
// Write the characters into the buffer.
|
||
char *outp = output + used;
|
||
do {
|
||
uint digit = (uint)(value % valueBase);
|
||
value /= valueBase;
|
||
|
||
if (digit >= 0 && digit <= 9) {
|
||
*--outp = (char)('0' + digit);
|
||
}
|
||
else {
|
||
*--outp = (char)('a' + (digit - 10));
|
||
}
|
||
} while (value != 0);
|
||
|
||
return output + used;
|
||
}
|
||
|
||
[NoHeapAllocation]
|
||
private static unsafe char * AddString(char *output, char *limit,
|
||
string value, int width)
|
||
{
|
||
fixed (char *pwString = &value.m_firstChar) {
|
||
return AddString(output, limit, pwString, value.m_stringLength, width);
|
||
}
|
||
}
|
||
|
||
[NoHeapAllocation]
|
||
private static unsafe char * AddString(char *output, char *limit,
|
||
char *pwString, int cwString,
|
||
int width)
|
||
{
|
||
// width < 0: left justified at -width.
|
||
// width = 0: no specified width.
|
||
// width > 0: right justified at width.
|
||
if (width == 0) {
|
||
width = cwString;
|
||
}
|
||
else if (width < 0) {
|
||
width = -width;
|
||
for (; width > cwString; width--) {
|
||
if (output < limit) {
|
||
*output++ = ' ';
|
||
}
|
||
}
|
||
}
|
||
|
||
while (cwString > 0 && width > 0) {
|
||
if (output < limit) {
|
||
*output++ = (char)*pwString++;
|
||
}
|
||
width--;
|
||
cwString--;
|
||
}
|
||
|
||
for (; width > 0; width--) {
|
||
if (output < limit) {
|
||
*output++ = ' ';
|
||
}
|
||
}
|
||
return output;
|
||
}
|
||
|
||
[NoHeapAllocation]
|
||
private static unsafe char * AddChar(char *output, char *limit, char value, int width)
|
||
{
|
||
// width < 0: left justified at -width.
|
||
// width = 0: no specified width.
|
||
// width > 0: right justified at width.
|
||
if (width == 0) {
|
||
width = 1;
|
||
}
|
||
else if (width < 0) {
|
||
width = -width;
|
||
for (; width > 1; width--) {
|
||
if (output < limit) {
|
||
*output++ = ' ';
|
||
}
|
||
}
|
||
}
|
||
|
||
if (width > 0) {
|
||
if (output < limit) {
|
||
*output++ = (char)value;
|
||
}
|
||
width--;
|
||
}
|
||
|
||
for (; width > 0; width--) {
|
||
if (output < limit) {
|
||
*output++ = ' ';
|
||
}
|
||
}
|
||
return output;
|
||
}
|
||
}
|
||
}
|