Return to String.cpp CVS log

Up to [Pegasus] / pegasus / src / Pegasus / Common

BUG#: 4200
TITLE: Pegasus String functions perform poorly

DESCRIPTION: Updates to fix bugs when ICU is used.  Also fixed an OS/400 problem
with operator<<

//%2005////////////////////////////////////////////////////////////////////////
//
// Copyright (c) 2000, 2001, 2002 BMC Software; Hewlett-Packard Development
// Company, L.P.; IBM Corp.; The Open Group; Tivoli Systems.
// Copyright (c) 2003 BMC Software; Hewlett-Packard Development Company, L.P.;
// IBM Corp.; EMC Corporation, The Open Group.
// Copyright (c) 2004 BMC Software; Hewlett-Packard Development Company, L.P.;
// IBM Corp.; EMC Corporation; VERITAS Software Corporation; The Open Group.
// Copyright (c) 2005 Hewlett-Packard Development Company, L.P.; IBM Corp.;
// EMC Corporation; VERITAS Software Corporation; The Open Group.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to
// deal in the Software without restriction, including without limitation the
// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
// sell copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// THE ABOVE COPYRIGHT NOTICE AND THIS PERMISSION NOTICE SHALL BE INCLUDED IN
// ALL COPIES OR SUBSTANTIAL PORTIONS OF THE SOFTWARE. THE SOFTWARE IS PROVIDED
// "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT
// LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR
// PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
// ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
//
//==============================================================================
//
// Author: Mike Brasher (mbrasher@bmc.com)
//
// Modified By: 
//     Roger Kumpf, Hewlett-Packard Company (roger_kumpf@hp.com)
//     Josephine Eskaline Joyce, IBM (jojustin@in.ibm.com) for Bug#3297
//     David Dillard, VERITAS Software Corp. (david.dillard@veritas.com)
//     Mike Brasher (mike-brasher@austin.rr.com)
//
//%/////////////////////////////////////////////////////////////////////////////

#ifndef PEGASUS_USE_INTERNAL_INLINES
# define PEGASUS_USE_INTERNAL_INLINES
#endif

#include "String.h"
#include <cassert>
#include "InternalException.h"
#include "CommonUTF.h"
#include "MessageLoader.h"
#include "CharSet.h"

#ifdef PEGASUS_STRING_ENABLE_ICU
#include <unicode/ustring.h>
#include <unicode/uchar.h>
#endif

PEGASUS_NAMESPACE_BEGIN

//==============================================================================
//
// Compile-time macros (undefined by default).
//
//     PEGASUS_STRING_ENABLE_ICU -- enables use of ICU package
//
//     PEGASUS_STRING_NO_THROW -- suppresses throwing of exceptions
//      
//     PEGASUS_STRING_NO_UTF8 -- don't generate slower UTF8 code.
//
//     PEGASUS_USE_INTERNAL_INLINES -- enables internal inlining feature.
//
//==============================================================================

//==============================================================================
//
// File-scope definitions:
//
//==============================================================================

// Converts 16-bit characters to upper case.
inline Uint16 _to_upper(Uint16 x)
{
    return (x & 0xFF00) ? x : CharSet::to_upper(x);
}

// Converts 16-bit characters to lower case.
inline Uint16 _to_lower(Uint16 x)
{
    return (x & 0xFF00) ? x : CharSet::to_lower(x);
}

// Rounds x to the next power of two (or just returns 8 if x < 8).
static Uint32 _next_pow_2(Uint32 x)
{
    if (x < 8)
	return 8;

    x--;
    x |= (x >> 1);
    x |= (x >> 2);
    x |= (x >> 4);
    x |= (x >> 8);
    x |= (x >> 16);
    x++;

    return x;
}

template<class P, class Q>
static void _copy(P* p, const Q* q, size_t n)
{
    // Use loop unrolling.

    while (n >= 8)
    {
	p[0] = q[0];
	p[1] = q[1];
	p[2] = q[2];
	p[3] = q[3];
	p[4] = q[4];
	p[5] = q[5];
	p[6] = q[6];
	p[7] = q[7];
	p += 8;
	q += 8;
	n -= 8;
    }

    while (n >= 4)
    {
	p[0] = q[0];
	p[1] = q[1];
	p[2] = q[2];
	p[3] = q[3];
	p += 4;
	q += 4;
	n -= 4;
    }

    while (n--)
	*p++ = *q++;
}

static Uint16* _find(const Uint16* s, size_t n, Uint16 c)
{
    while (n >= 4)
    {
	if (s[0] == c)
	    return (Uint16*)s;
	if (s[1] == c)
	    return (Uint16*)&s[1];
	if (s[2] == c)
	    return (Uint16*)&s[2];
	if (s[3] == c)
	    return (Uint16*)&s[3];

	n -= 4;
	s += 4;
    }

    if (n)
    {
	if (*s == c)
	    return (Uint16*)s;
	s++;
	n--;
    }

    if (n)
    {
	if (*s == c)
	    return (Uint16*)s;
	s++;
	n--;
    }

    if (n && *s == c)
	return (Uint16*)s;

    // Not found!
    return 0;
}

static int _compare(const Uint16* s1, const Uint16* s2)
{
    while (*s1 && *s2)
    {
        int r = *s1++ - *s2++;

        if (r)
            return r;
    }

    if (*s2)
        return -1;
    else if (*s1)
        return 1;

    return 0;
}

static int _compare_no_utf8(const Uint16* s1, const char* s2)
{
    Uint16 c1;
    Uint16 c2;

    do
    {
	c1 = *s1++;
	c2 = *s2++;

	if (c1 == 0)
	    return c1 - c2;
    }
    while (c1 == c2);

    return c1 - c2;
}

static int _compare(const Uint16* s1, const Uint16* s2, size_t n)
{
    // This should only be called when s1 and s2 have the same length.

    while (n-- && (*s1++ - *s2++) == 0)
	;

    // 

    return s1[-1] - s2[-1];
}

static inline void _copy(Uint16* s1, const Uint16* s2, size_t n)
{
    memcpy(s1, s2, n * sizeof(Uint16));
}

void String_throw_out_of_bounds()
{
    throw IndexOutOfBoundsException();
}

#ifdef PEGASUS_STRING_NO_THROW
# define _check_null_pointer(ARG) /* empty */
#else
template<class T>
inline void _check_null_pointer(const T* ptr)
{
    if (!ptr)
	throw NullPointer();
}
#endif

static size_t _copy_from_utf8(Uint16* dest, const char* src, size_t n)
{
    Uint16* p = dest;
    const Uint8* q = (const Uint8*)src;

    // Process leading 7-bit ASCII characters (to avoid UTF8 overhead later).
    // Use loop-unrolling.

    while (n >=8 && ((q[0]|q[1]|q[2]|q[3]|q[4]|q[5]|q[6]|q[7]) & 0x80) == 0)
    {
	p[0] = q[0];
	p[1] = q[1];
	p[2] = q[2];
	p[3] = q[3];
	p[4] = q[4];
	p[5] = q[5];
	p[6] = q[6];
	p[7] = q[7];
	p += 8;
	q += 8;
	n -= 8;
    }

    while (n >=4 && ((q[0]|q[1]|q[2]|q[3]) & 0x80) == 0)
    {
	p[0] = q[0];
	p[1] = q[1];
	p[2] = q[2];
	p[3] = q[3];
	p += 4;
	q += 4;
	n -= 4;
    }

    switch (n)
    {
	case 0:
	    return p - dest;
	case 1:
	    if (q[0] < 128)
	    {
		p[0] = q[0];
		return p + 1 - dest;
	    }
	    break;
	case 2:
	    if (((q[0]|q[1]) & 0x80) == 0)
	    {
		p[0] = q[0];
		p[1] = q[1];
		return p + 2 - dest;
	    }
	    break;
	case 3:
	    if (((q[0]|q[1]|q[2]) & 0x80) == 0)
	    {
		p[0] = q[0];
		p[1] = q[1];
		p[2] = q[2];
		return p + 3 - dest;
	    }
	    break;
    }

    // Process remaining characters.

    while (n)
    {
	// Optimize for 7-bit ASCII case.

	if (*q < 128)
	{
	    *p++ = *q++;
	    n--;
	}
	else
	{
	    Uint8 c = UTF_8_COUNT_TRAIL_BYTES(*q) + 1;

	    if (c > n || !isValid_U8(q, c) ||
		UTF8toUTF16(&q, q + c, &p, p + n) != 0)
	    {
            MessageLoaderParms parms("Common.String.BAD_UTF8",
                "The byte sequence starting at index $0 is not valid UTF-8 encoding.",
                 q - (const Uint8*)src);
            throw Exception(parms);
	    }

	    n -= c;
	}
    }

    return p - dest;
}

// Note: dest must be at least three times src (plus an extra byte for 
// terminator).
static inline size_t _copy_to_utf8(char* dest, const Uint16* src, size_t n)
{
    const Uint16* q = src;
    Uint8* p = (Uint8*)dest;

    while (n >= 4 && q[0] < 128 && q[1] < 128 && q[2] < 128 && q[3] < 128)
    {
	p[0] = q[0];
	p[1] = q[1];
	p[2] = q[2];
	p[3] = q[3];
	p += 4;
	q += 4;
	n -= 4;
    }

    switch (n)
    {
	case 0:
	    return p - (Uint8*)dest;
	case 1:
	    if (q[0] < 128)
	    {
		p[0] = q[0];
		return p + 1 - (Uint8*)dest;
	    }
	    break;
	case 2:
	    if (q[0] < 128 && q[1] < 128)
	    {
		p[0] = q[0];
		p[1] = q[1];
		return p + 2 - (Uint8*)dest;
	    }
	    break;
	case 3:
	    if (q[0] < 128 && q[1] < 128 && q[2] < 128)
	    {
		p[0] = q[0];
		p[1] = q[1];
		p[2] = q[2];
		return p + 3 - (Uint8*)dest;
	    }
	    break;
    }

    // If this line was reached, there must be characters greater than 128.

    UTF16toUTF8(&q, q + n, &p, p + 3 * n);

    return p - (Uint8*)dest;
}

static inline size_t _convert(Uint16* p, const char* q, size_t n)
{
#ifdef PEGASUS_STRING_NO_UTF8
    _copy(p, q, n);
    return n;
#else
    return _copy_from_utf8(p, q, n);
#endif
}

//==============================================================================
//
// class CString
//
//==============================================================================

CString::CString(const CString& cstr) : _rep(0)
{
    if (cstr._rep)
    {
	size_t n = strlen(cstr._rep) + 1;
        _rep = (char*)operator new(n);
	memcpy(_rep, cstr._rep, n);
    }
}

CString& CString::operator=(const CString& cstr)
{
    if (&cstr != this)
    {
        if (_rep)
        {
            operator delete(_rep);
            _rep = 0;
        }

        if (cstr._rep)
        {
	    size_t n = strlen(cstr._rep) + 1;
            _rep = (char*)operator new(n);
	    memcpy(_rep, cstr._rep, n);
        }
    }

    return *this;
}

//==============================================================================
//
// class StringRep
//
//==============================================================================

StringRep StringRep::_empty_rep;

inline StringRep* StringRep::alloc(size_t cap)
{
    StringRep* rep = (StringRep*)::operator new(
	sizeof(StringRep) + cap * sizeof(Uint16));
    rep->cap = cap;
    Atomic_create(&rep->refs, 1);

    return rep;
}

static inline void _reserve(StringRep*& rep, Uint32 cap)
{
    if (cap > rep->cap || Atomic_get(&rep->refs) != 1)
    {
	size_t n = _next_pow_2(cap);
	StringRep* new_rep = StringRep::alloc(n);
	new_rep->size = rep->size;
	_copy(new_rep->data, rep->data, rep->size + 1);
	StringRep::unref(rep);
	rep = new_rep;
    }
}

StringRep* StringRep::create(const Uint16* data, size_t size)
{
    StringRep* rep = StringRep::alloc(size);
    rep->size = size;
    _copy(rep->data, data, size);
    rep->data[size] = '\0';
    return rep;
}

StringRep* StringRep::copy_on_write(StringRep* rep)
{
    // Return a new copy of rep. Release rep.

    StringRep* new_rep = StringRep::alloc(rep->size);
    new_rep->size = rep->size;
    _copy(new_rep->data, rep->data, rep->size);
    new_rep->data[new_rep->size] = '\0';
    StringRep::unref(rep);
    return new_rep;
}

StringRep* StringRep::create(const char* data, size_t size)
{
    StringRep* rep = StringRep::alloc(size);
    rep->size = _convert((Uint16*)rep->data, data, size);
    rep->data[rep->size] = '\0';

    return rep;
}

StringRep* StringRep::createASCII7(const char* data, size_t size)
{
    StringRep* rep = StringRep::alloc(size);
    _copy((Uint16*)rep->data, data, size);
    rep->data[rep->size = size] = '\0';
    return rep;
}

Uint32 StringRep::length(const Uint16* str)
{
    // Note: We could unroll this but it is rarely called.

    const Uint16* end = (Uint16*)str;

    while (*end++)
	;

    return end - str - 1;
}

//==============================================================================
//
// class String
//
//==============================================================================

const String String::EMPTY;

String::String(const String& str, Uint32 n)
{
    _check_bounds(n, str._rep->size);
    _rep = StringRep::create(str._rep->data, n);
}

String::String(const Char16* str)
{
    _check_null_pointer(str);
    _rep = StringRep::create((Uint16*)str, StringRep::length((Uint16*)str));
}

String::String(const Char16* str, Uint32 n)
{
    _check_null_pointer(str);
    _rep = StringRep::create((Uint16*)str, n);
}

String::String(const char* str)
{
    _check_null_pointer(str);
    _rep = StringRep::create(str, strlen(str));
}

String::String(const char* str, String::ASCII7Tag tag)
{
    _check_null_pointer(str);
    _rep = StringRep::createASCII7(str, strlen(str));
}

String::String(const char* str, Uint32 n)
{
    _check_null_pointer(str);
    _rep = StringRep::create(str, n);
}

String::String(const char* str, size_t n, String::ASCII7Tag tag)
{
    _check_null_pointer(str);
    _rep = StringRep::createASCII7(str, n);
}

String::String(const String& s1, const String& s2)
{
    size_t n1 = s1._rep->size;
    size_t n2 = s2._rep->size;
    size_t n = n1 + n2;
    _rep = StringRep::alloc(n);
    _copy(_rep->data, s1._rep->data, n1);
    _copy(_rep->data + n1, s2._rep->data, n2);
    _rep->size = n;
    _rep->data[n] = '\0';
}

String::String(const String& s1, const char* s2)
{
    _check_null_pointer(s2);
    size_t n1 = s1._rep->size;
    size_t n2 = strlen(s2);
    _rep = StringRep::alloc(n1 + n2);
    _copy(_rep->data, s1._rep->data, n1);
    _rep->size = n1 + _convert((Uint16*)_rep->data + n1, s2, n2);
    _rep->data[_rep->size] = '\0';
}

String::String(const char* s1, const String& s2)
{
    _check_null_pointer(s1);
    size_t n1 = strlen(s1);
    size_t n2 = s2._rep->size;
    _rep = StringRep::alloc(n1 + n2);
    _rep->size = n2 + _convert((Uint16*)_rep->data, s1, n1);
    _copy(_rep->data + n1, s2._rep->data, n2);
    _rep->data[_rep->size] = '\0';
}

String& String::assign(const String& str)
{
    if (_rep != str._rep)
    {
	StringRep::unref(_rep);
	StringRep::ref(_rep = str._rep);
    }

    return *this;
}

String& String::assign(const Char16* str, Uint32 n)
{
    _check_null_pointer(str);

    if (n > _rep->cap || Atomic_get(&_rep->refs) != 1)
    {
	StringRep::unref(_rep);
	_rep = StringRep::alloc(n);
    }

    _rep->size = n;
    _copy(_rep->data, (Uint16*)str, n);
    _rep->data[n] = '\0';

    return *this;
}

String& String::assign(const char* str, Uint32 n)
{
    _check_null_pointer(str);

    if (n > _rep->cap || Atomic_get(&_rep->refs) != 1)
    {
	StringRep::unref(_rep);
	_rep = StringRep::alloc(n);
    }

    _rep->size = _convert(_rep->data, str, n);
    _rep->data[_rep->size] = 0;

    return *this;
}

String& String::assignASCII7(const char* str, Uint32 n)
{
    _check_null_pointer(str);

    if (n > _rep->cap || Atomic_get(&_rep->refs) != 1)
    {
	StringRep::unref(_rep);
	_rep = StringRep::alloc(n);
    }

    _copy(_rep->data, str, n);
    _rep->data[_rep->size = n] = 0;

    return *this;
}

void String::clear()
{
    if (_rep->size)
    {
	if (Atomic_get(&_rep->refs) == 1)
	    _rep->size = 0;
	else
	{
	    StringRep::unref(_rep);
	    _rep = &StringRep::_empty_rep;
	}
    }
}

void String::reserveCapacity(Uint32 cap)
{
    _reserve(_rep, cap);
}

CString String::getCString() const
{
#ifdef PEGASUS_STRING_NO_UTF8
    char* str = (char*)operator new(_rep->size + 1);
    _copy(str, _rep->data, _rep->size);
    str[_rep->size] = '\0';
    return CString(str);
#else
    Uint32 n = 3 * _rep->size;
    char* str = (char*)operator new(n + 1);
    size_t size = _copy_to_utf8(str, _rep->data, _rep->size);
    str[size] = '\0';
    return CString(str);
#endif
}

String& String::append(const Char16* str, Uint32 n)
{
    _check_null_pointer(str);

    size_t old_size = _rep->size;
    size_t new_size = old_size + n;
    _reserve(_rep, new_size);
    _copy(_rep->data + old_size, (Uint16*)str, n);
    _rep->size = new_size;
    _rep->data[new_size] = '\0';

    return *this;
}

String& String::append(const String& str)
{
    return append((Char16*)str._rep->data, str._rep->size);
}

String& String::append(const char* str, Uint32 size)
{
    _check_null_pointer(str);

    size_t old_size = _rep->size;
    size_t cap = old_size + size;

    _reserve(_rep, cap);
    _rep->size += _convert((Uint16*)_rep->data + old_size, str, size);
    _rep->data[_rep->size] = '\0';

    return *this;
}

void String::remove(Uint32 index, Uint32 n)
{
    if (n == PEG_NOT_FOUND)
        n = _rep->size - index;

    _check_bounds(index + n, _rep->size);

    if (Atomic_get(&_rep->refs) != 1)
	_rep = StringRep::copy_on_write(_rep);

    assert(index + n <= _rep->size);

    size_t rem = _rep->size - (index + n);
    Uint16* data = _rep->data;

    if (rem)
        memmove(data + index, data + index + n, rem * sizeof(Uint16));

    _rep->size -= n;
    data[_rep->size] = '\0';
}

String String::subString(Uint32 index, Uint32 n) const
{
    // Note: this implementation is very permissive but used for
    // backwards compatibility.

    if (index < _rep->size)
    {
	if (n == PEG_NOT_FOUND || n > _rep->size - index)
	    n = _rep->size - index;

	return String((Char16*)_rep->data + index, n);
    }

    return String();
}

Uint32 String::find(Char16 c) const
{
    Uint16* p = (Uint16*)_find(_rep->data, _rep->size, c);

    if (p)
	return p - _rep->data;

    return PEG_NOT_FOUND;
}

Uint32 String::find(Uint32 index, Char16 c) const
{
    _check_bounds(index, _rep->size);

    if (index >= _rep->size)
	return PEG_NOT_FOUND;

    Uint16* p = (Uint16*)_find(_rep->data + index, _rep->size - index, c);

    if (p)
	return p - _rep->data;

    return PEG_NOT_FOUND;
}

Uint32 String::_find_aux(const Char16* s, Uint32 n) const
{
    _check_null_pointer(s);

    const Uint16* data = _rep->data;
    size_t rem = _rep->size;

    while (n <= rem)
    {
	Uint16* p = (Uint16*)_find(data, rem, s[0]);

	if (!p)
	    break;

	if (memcmp(p, s, n * sizeof(Uint16)) == 0)
	    return p - _rep->data;

	p++;
	rem -= p - data;
	data = p;
    }

    return PEG_NOT_FOUND;
}

Uint32 String::find(const char* s) const
{
    _check_null_pointer(s);

    // Note: could optimize away creation of temporary, but this is rarely
    // called.
    return find(String(s));
}

Uint32 String::reverseFind(Char16 c) const
{
    Uint16 x = c;
    Uint16* p = _rep->data;
    Uint16* q = _rep->data + _rep->size;

    while (q != p)
    {
	if (*--q == x)
	    return q - p;
    }

    return PEG_NOT_FOUND;
}

void String::toLower()
{
#ifdef PEGASUS_STRING_ENABLE_ICU

    if (InitializeICU::initICUSuccessful())
    {
        if (Atomic_get(&_rep->refs) != 1)
	        _rep = StringRep::copy_on_write(_rep);

        // This will do a locale-insensitive, but context-sensitive convert.
        // Since context-sensitive casing looks at adjacent chars, this 
        // prevents optimizations where the us-ascii is converted before 
        // calling ICU.
        // The string may shrink or expand after the convert.

	//// First calculate size of resulting string. u_strToLower() returns
	//// only the size when zero is passed as the destination size argument.

        UErrorCode err = U_ZERO_ERROR;

        int32_t new_size = u_strToLower(
	    NULL, 0, (UChar*)_rep->data, _rep->size, NULL, &err);
        
        err = U_ZERO_ERROR;

	//// Reserve enough space for the result.

	if ((Uint32)new_size > _rep->cap)
	    _reserve(_rep, new_size);

	//// Perform the conversion (overlapping buffers are allowed).

        u_strToLower((UChar*)_rep->data, new_size,
	    (UChar*)_rep->data, _rep->size, NULL, &err);

	_rep->size = new_size;
    return;
    }

#endif /* PEGASUS_STRING_ENABLE_ICU */

    if (Atomic_get(&_rep->refs) != 1)
	_rep = StringRep::copy_on_write(_rep);

    Uint16* p = _rep->data;
    size_t n = _rep->size;

    for (; n--; p++)
    {
	if (!(*p & 0xFF00))
	    *p = _to_lower(*p);
    }
}

void String::toUpper()
{
#ifdef PEGASUS_STRING_ENABLE_ICU

    if (InitializeICU::initICUSuccessful())
    {
        if (Atomic_get(&_rep->refs) != 1)
	        _rep = StringRep::copy_on_write(_rep);

        // This will do a locale-insensitive, but context-sensitive convert.
        // Since context-sensitive casing looks at adjacent chars, this 
        // prevents optimizations where the us-ascii is converted before 
        // calling ICU.
        // The string may shrink or expand after the convert.

	//// First calculate size of resulting string. u_strToUpper() returns
	//// only the size when zero is passed as the destination size argument.

        UErrorCode err = U_ZERO_ERROR;

        int32_t new_size = u_strToUpper(
	    NULL, 0, (UChar*)_rep->data, _rep->size, NULL, &err);

        err = U_ZERO_ERROR;

	//// Reserve enough space for the result.

	if ((Uint32)new_size > _rep->cap)
	    _reserve(_rep, new_size);

	//// Perform the conversion (overlapping buffers are allowed).

        u_strToUpper((UChar*)_rep->data, new_size,
	    (UChar*)_rep->data, _rep->size, NULL, &err);

	_rep->size = new_size;

    return;
    }

#endif /* PEGASUS_STRING_ENABLE_ICU */

    if (Atomic_get(&_rep->refs) != 1)
	_rep = StringRep::copy_on_write(_rep);

    Uint16* p = _rep->data;
    size_t n = _rep->size;

    for (; n--; p++)
	*p = _to_upper(*p);
}

int String::compare(const String& s1, const String& s2, Uint32 n)
{
    assert(n <= s1._rep->size);
    assert(n <= s2._rep->size);

    // Ignoring error in which n is greater than s1.size() or s2.size()
    return _compare(s1._rep->data, s2._rep->data, n);
}

int String::compare(const String& s1, const String& s2)
{
    return _compare(s1._rep->data, s2._rep->data);
}

int String::compare(const String& s1, const char* s2)
{
    _check_null_pointer(s2);

#ifdef PEGASUS_STRING_NO_UTF8
    return _compare_no_utf8(s1._rep->data, s2);
#else
    // ATTN: optimize this!
    return String::compare(s1, String(s2));
#endif
}

int String::compareNoCase(const String& str1, const String& str2)
{
#ifdef PEGASUS_STRING_ENABLE_ICU

    if (InitializeICU::initICUSuccessful())
    {
        return  u_strcasecmp(
	    str1._rep->data, str2._rep->data, U_FOLD_CASE_DEFAULT);
    }

#endif /* PEGASUS_STRING_ENABLE_ICU */

    const Uint16* s1 = str1._rep->data;
    const Uint16* s2 = str2._rep->data;

    while (*s1 && *s2)
    {
        int r = _to_lower(*s1++) - _to_lower(*s2++);

        if (r)
            return r;
    }

    if (*s2)
        return -1;
    else if (*s1)
        return 1;

    return 0;
}

Boolean String::equalNoCase_aux(const String& s1, const String& s2)
{
#ifdef PEGASUS_STRING_ENABLE_ICU

    return String::compareNoCase(s1, s2) == 0;

#else /* PEGASUS_STRING_ENABLE_ICU */

    Uint16* p = (Uint16*)s1._rep->data;
    Uint16* q = (Uint16*)s2._rep->data;
    Uint32 n = s2._rep->size;

    while (n >= 8)
    {
	if (((p[0] - q[0]) && (_to_upper(p[0]) - _to_upper(q[0]))) ||
	    ((p[1] - q[1]) && (_to_upper(p[1]) - _to_upper(q[1]))) ||
	    ((p[2] - q[2]) && (_to_upper(p[2]) - _to_upper(q[2]))) ||
	    ((p[3] - q[3]) && (_to_upper(p[3]) - _to_upper(q[3]))) ||
	    ((p[4] - q[4]) && (_to_upper(p[4]) - _to_upper(q[4]))) ||
	    ((p[5] - q[5]) && (_to_upper(p[5]) - _to_upper(q[5]))) ||
	    ((p[6] - q[6]) && (_to_upper(p[6]) - _to_upper(q[6]))) ||
	    ((p[7] - q[7]) && (_to_upper(p[7]) - _to_upper(q[7]))))
	{
	    return false;
	}

	n -= 8;
	p += 8;
	q += 8;
    }

    while (n >= 4)
    {
	if (((p[0] - q[0]) && (_to_upper(p[0]) - _to_upper(q[0]))) ||
	    ((p[1] - q[1]) && (_to_upper(p[1]) - _to_upper(q[1]))) ||
	    ((p[2] - q[2]) && (_to_upper(p[2]) - _to_upper(q[2]))) ||
	    ((p[3] - q[3]) && (_to_upper(p[3]) - _to_upper(q[3]))))
	{
	    return false;
	}

	n -= 4;
	p += 4;
	q += 4;
    }

    while (n--)
    {
	if (((p[0] - q[0]) && (_to_upper(p[0]) - _to_upper(q[0]))))
	    return false;

	p++;
	q++;
    }

    return true;

#endif /* PEGASUS_STRING_ENABLE_ICU */
}

Boolean String::equalNoCase(const String& s1, const char* s2)
{
    _check_null_pointer(s2);

#if defined(PEGASUS_STRING_ENABLE_ICU)

    return String::equalNoCase(s1, String(s2));

#elif defined(PEGASUS_STRING_NO_UTF8)

    const Uint16* p1 = (Uint16*)s1._rep->data;
    const char* p2 = s2;
    size_t n = s1._rep->size;

    while (n--)
    {
	if (!*p2)
	    return false;

	if (_to_upper(*p1++) != CharSet::to_upper(int(*p2++)))
	    return false;
    }
    
    return true;

#else /* PEGASUS_STRING_ENABLE_ICU */

    // ATTN: optimize this!
    return String::equalNoCase(s1, String(s2));

#endif /* PEGASUS_STRING_ENABLE_ICU */
}

Boolean String::equal(const String& s1, const String& s2)
{
    return s1._rep->size == s2._rep->size && memcmp(s1._rep->data, 
	s2._rep->data, s1._rep->size * sizeof(Uint16)) == 0;
}

Boolean String::equal(const String& s1, const char* s2)
{
#ifdef PEGASUS_STRING_NO_UTF8

    _check_null_pointer(s2);

    const Uint16* p = (Uint16*)s1._rep->data;
    const char* q = s2;

    while (*p && *q)
    {
	if (*p++ != Uint16(*q++))
	    return false;
    }

    return !(*p || *q);

#else /* PEGASUS_STRING_NO_UTF8 */

    return String::equal(s1, String(s2));

#endif /* PEGASUS_STRING_NO_UTF8 */
}

PEGASUS_STD(ostream)& operator<<(PEGASUS_STD(ostream)& os, const String& str)
{
#if defined(PEGASUS_OS_OS400)

    CString cstr = str.getCString();
    const char* utf8str = cstr;
    os << utf8str;
    return os;
#else    

#if defined(PEGASUS_STRING_ENABLE_ICU)

    if (InitializeICU::initICUSuccessful())
    {
        char *buf = NULL;
        const int size = str.size() * 6;
        UnicodeString UniStr(
	    (const UChar *)str.getChar16Data(), (int32_t)str.size());
        Uint32 bufsize = UniStr.extract(0,size,buf);
        buf = new char[bufsize+1];
        UniStr.extract(0,bufsize,buf);
        os << buf;
        os.flush();
        delete [] buf;
        return os;       
    }

#endif  // PEGASUS_STRING_ENABLE_ICU 

    for (Uint32 i = 0, n = str.size(); i < n; i++)
    {
        Uint16 code = str[i];

       	if (code > 0 && !(code & 0xFF00))
   	        os << char(code);
        else
   	    {
            // Print in hex format:
            char buffer[8];
            sprintf(buffer, "\\x%04X", code);
            os << buffer;
        }
    }

    return os;
#endif // PEGASUS_OS_OS400
}

void String::_append_char_aux()
{
    StringRep* tmp;

    if (_rep->cap)
    {
	tmp = StringRep::alloc(2 * _rep->cap);
	tmp->size = _rep->size;
	_copy(tmp->data, _rep->data, _rep->size);
    }
    else
    {
	tmp = StringRep::alloc(8);
	tmp->size = 0;
    }

    StringRep::unref(_rep);
    _rep = tmp;
}

PEGASUS_NAMESPACE_END

/*
================================================================================

String optimizations:

    1.  Added mechanism allowing certain functions to be inlined only when
	used by internal Pegasus modules. External modules (i.e., providers)
	link to a non-inline version, which allows for binary compatibility.

    2.  Implemented copy-on-write with atomic increment/decrement. This
	yieled a 10% improvement for the 'gc' benchmark and a 11% improvment
	for the 'ni1000' benchmark.

    3.	Employed loop unrolling in several places. For example, see:

	    static Uint16* _find(const Uint16* s, size_t n, Uint16 c);

    4.  Used the "empty-rep" optimization (described in whitepaper from the
	GCC Developers Summit). This reduced default construction to a simple
	pointer assignment.

	    inline String::String() : _rep(&_empty_rep) { }

    5.  Implemented Uint16 versions of toupper() and tolower() using tables.
	For example:

	    static const char _upper[] =
	    {
		0,1,2,...255
	    };

	    inline Uint16 _to_upper(Uint16 x)
	    {
		return (x & 0xFF00) ? x : _upper[x];
	    }

	This outperforms the system implementation by avoiding an anding 
	operation.

    6.  Implemented char* version of the following member functions to 
	eliminate unecessary creation of anonymous string objects 
	(temporaries).

	    String(const String& s1, const char* s2);
	    String(const char* s1, const String& s2);
	    String& String::operator=(const char* str);
	    Uint32 String::find(const char* s) const;
	    bool String::equal(const String& s1, const char* s2);
	    static int String::compare(const String& s1, const char* s2);
	    String& String::append(const char* str);
	    String& String::append(const char* str, Uint32 size);
	    static bool String::equalNoCase(const String& s1, const char* s2);
	    String& operator=(const char* str)
	    String& String::assign(const char* str)
	    String& String::append(const char* str)
	    Boolean operator==(const String& s1, const char* s2)
	    Boolean operator==(const char* s1, const String& s2)
	    Boolean operator!=(const String& s1, const char* s2)
	    Boolean operator!=(const char* s1, const String& s2)
	    Boolean operator<(const String& s1, const char* s2)
	    Boolean operator<(const char* s1, const String& s2)
	    Boolean operator>(const String& s1, const char* s2)
	    Boolean operator>(const char* s1, const String& s2)
	    Boolean operator<=(const String& s1, const char* s2)
	    Boolean operator<=(const char* s1, const String& s2)
	    Boolean operator>=(const String& s1, const char* s2)
	    Boolean operator>=(const char* s1, const String& s2)
	    String operator+(const String& s1, const char* s2)
	    String operator+(const char* s1, const String& s2)

    7.  Optimized _next_pow_2(), used in rounding the capacity to the next 
        power of two (algorithm from the book "Hacker's Delight").

	    static Uint32 _next_pow_2(Uint32 x)
	    {
		if (x < 8)
		    return 8;

		x--;
		x |= (x >> 1);
		x |= (x >> 2);
		x |= (x >> 4);
		x |= (x >> 8);
		x |= (x >> 16);
		x++;

		return x;
	    }

    8.  Implemented "concatenating constructors" to eliminate temporaries
	created by operator+(). This scheme employs the "return-value 
	optimization" described by Stan Lippman.

	    inline String operator+(const String& s1, const String& s2)
	    {
		return String(s1, s2, 0);
	    }

    9.  Experimented to find the optimial initial size for a short string.
	Eight seems to offer the best tradoff between space and time.

    10. Inlined all members of the Char16 class.

    11. Used Uint16 internally in the String class. This showed no improvememnt
	since Char16 was already fully inlined and was essentially reduced to
	Uint16 in any case.

    12. Implemented conditional logic (#if) allowing error checking logic to
	be excluded to better performance. Examples include bounds checking 
	and null-pointer checking.

    13. Used memcpy() and memcmp() where possible. These are implemented using
	the rep family of intructions under Intel and are much faster.

    14. Used loop unrolling, jump-tables, and short-circuiting to reduce UTF8 
	copy routine overhead.

    15. Added ASCII7 form of the constructor and assign().

	    String s("hello world", String::ASCII7);

	    s.assignASCII7("hello world");

	This avoids slower UTF8 processing when not needed.

================================================================================
*/