xref: /dports/mail/archiveopteryx/archiveopteryx-3.2.0/core/estring.cpp

// Copyright 2009 The Archiveopteryx Developers <info@aox.org>

#include "estring.h"

#include "allocator.h"

// stderr, fprintf
#include <stdio.h>
// strlen
#include <string.h>


/*! \class EStringData estring.h

    This private helper class contains the actual string data. It has
    three fields, all accessible only to string. The only noteworthy
    field is max, which is 0 in the case of a shared/read-only string,
    and nonzero in the case of a string which can be modified.
*/


/*! \fn EStringData::EStringData()

    Creates a zero-length string. This is naturally read-only.
*/

/*! Creates a new EString with \a bytes capacity. */

EStringData::EStringData( int bytes )
    : str( 0 ), len( 0 ), max( bytes )
{
    if ( str )
        str = (char*)Allocator::alloc( max, 0 );
}


void * EStringData::operator new( size_t ownSize, uint extra )
{
    return Allocator::alloc( ownSize + extra, 1 );
}


/*! \class EString estring.h
    An email-oriented 8-bit string class.

    The string data are counted, so null bytes are allowed, and most
    operations are very fast.

    The data structure uses a simplified variant of reference counting,
    where only "one" and "many" are possible. The detach() function
    ensures that the count is "one" afterwards. Many functions leave
    the count on "many", even ones such as mid().

    The usual string functions are implemented, along with a variety
    of email-specific operations such as eQP(), deQP(), needsQP(),
    e64(). boring() returns true if the string can be used unquoted in
    e.g. MIME, quoted() quotes it. upper() and lower() have a third
    sibling, headerCased(). simplified() and trimmed() remove white
    space in ways email often needs.

    Several people have asked why we use this class instead of
    std::string, and UString instead of std::basic_string<uint>. At
    least one told us we were wrong and wheel reinvention and so on.

    The reason is that working with email using EString is fairly
    pleasant. Doing the same kind of work with std::string is a pain.

    Both wheels are round. EString carries its load and moves it
    smoothly along.
*/

/*! Creates an empty EString */

EString::EString()
    : d( 0 )
{
}

/*! Creates a EString from the NUL-terminated string \a s.
    The NUL is not copied. */

EString::EString( const char *s )
    : d( 0 )
{
    *this = s;
}

/*! Creates a EString from the first \a n bytes of \a s, which may
    contain NULs. */

EString::EString( const char *s, uint n )
    : d( 0 )
{
    append( s, n );
}


/*! Creates a copy of \a s. */

EString::EString( const EString &s )
    : Garbage(), d( 0 )
{
    *this = s;
}


/*! \fn void EString::detach()

    Ensures that the string is modifiable. All EString functions call
    this prior to modifying the string. */


/*! Destroys the string.

    Because EString is used so much, and can eat up such vast amounts
    of memory so quickly, this destructor does something: If the
    string is the sole owner of its data, it frees them.

    As of April 2005, the return values of data() or cstr() are NO
    LONGER valid after a string has gone out of scope or otherwise been
    lost.
*/

EString::~EString()
{
    if ( d && d->max )
        Allocator::dealloc( d );
    d = 0;
}


/*! Deletes \a p. (This function exists only so that gcc -O3 doesn't
    decide that EString objects don't need destruction.)
*/

void EString::operator delete( void *p )
{
    EStringData * & d = ((EString *)p)->d;
    if ( d && d->max )
        Allocator::dealloc( d );
    d = 0;
}


/*! Copies \a other to this string and returns a reference to this
    string. */

EString & EString::operator=( const EString & other )
{
    d = other.d;
    if ( d )
        d->max = 0;
    return *this;
}


/*! Copies \a s to this string and returns a reference to this
    string. If \a s is a null pointer, the result is an empty string. */

EString & EString::operator=( const char * s )
{
    if ( !s || !*s ) {
        d = 0;
        return *this;
    }

    uint len = strlen( s );
    if ( d && d->max )
        d->len = 0;
    reserve( len );
    d->len = len;
    memmove( d->str, s, d->len );
    return *this;
}


/*! \fn uint EString::length() const

    Returns the length of the string. The length does not include any
    terminator or padding. */


/*! \fn uint EString::capacity() const

    Returns the capacity of the string variable, that is, how long the
    string can be before it has to allocate memory.
*/


/*! \fn const char *EString::data() const

    Returns a pointer to the string's byte representation, which is
    NOT necessarily zero-terminated. */


/*! Returns the zero-terminated byte representation of the
    string. Note that even though the return value is zero-terminated,
    it can also contain null bytes in the middle.

    Even though this function modifies memory, it doesn't detach(),
    since it doesn't modify the string. However, in most cases its
    call to reserve() causes a detach().
*/

const char * EString::cstr()
{
    reserve( length()+1 );
    d->str[d->len] = '\0';
    return data();
}


/*! This const version of cstr() is the same as the non-const version
    above. The only difference is that it can be called on a const
    object, and that it may cause some memory allocation elsewhere.
*/

const char * EString::cstr() const
{
    if ( d && d->max > d->len ) {
        d->str[d->len] = '\0';
        return data();
    }
    EString tmp;
    tmp.reserve( length() + 1 );
    tmp = *this;
    return tmp.cstr();
}


/*! Returns a copy of this string where all upper-case letters (A-Z -
    this is ASCII only) have been changed to lower case. */

EString EString::lower() const
{
    EString result( *this );
    uint i = 0;
    while ( i < result.length() ) {
        if ( result[i] >= 'A' && result[i] <= 'Z' ) {
            result.detach();
            result.d->str[i] = result.d->str[i] + 32;
        }
        i++;
    }
    return result;
}


/*! Returns a copy of this string where all lower-case letters (a-z -
    this is ASCII only) have been changed to upper case. */

EString EString::upper() const
{
    EString result( *this );
    uint i = 0;
    while ( i < result.length() ) {
        if ( result[i] >= 'a' && result[i] <= 'z' ) {
            result.detach();
            result.d->str[i] = result.d->str[i] - 32;
        }
        i++;
    }
    return result;
}


/*! Returns a copy of this string where all letters have been changed
  to conform to typical mail header practice: Letters following digits
  and other letters are lower-cased. Other letters are upper-cased
  (notably including the very first character). */

EString EString::headerCased() const
{
    EString result( *this );
    uint i = 0;
    bool u = true;
    while ( i < length() ) {
        if ( u && result[i] >= 'a' && result[i] <= 'z' ) {
            result.detach();
            result.d->str[i] = result[i] - 32;
        }
        else if ( !u && result[i] >= 'A' && result[i] <= 'Z' ) {
            result.detach();
            result.d->str[i] = result[i] + 32;
        }
        if ( ( result[i] >= 'A' && result[i] <= 'Z' ) ||
             ( result[i] >= 'a' && result[i] <= 'z' ) ||
             ( result[i] >= '0' && result[i] <= '9' ) )
            u = false;
        else
            u = true;
        i++;
    }
    return result;
}





/*! Returns the position of the first occurence of \a c on or after \a i
    in this string, or -1 if there is none.
*/

int EString::find( char c, int i ) const
{
    while ( i < (int)length() && at( i ) != c )
        i++;
    if ( i < (int)length() )
        return i;
    return -1;
}


/*! Returns the position of the first occurence of \a s on or after \a i
    in this string, or -1 if there is none.
*/

int EString::find( const EString & s, int i ) const
{
    uint j = 0;
    while ( j < s.length() && i+j < length() ) {
        if ( d->str[i+j] == s.d->str[j] ) {
            j++;
        }
        else {
            j = 0;
            i++;
        }
    }
    if ( j == s.length() )
        return i;
    return -1;
}


/*! Returns section \a n of this string, where a section is defined as
    a run of sequences separated by \a s. If \a s is the empty string
    or \a n is 0, section() returns this entire string. If this string
    contains fewer instances of \a s than \a n (ie. section \a n is
    after the end of the string), section returns an empty string.
*/

EString EString::section( const EString & s, uint n ) const
{
    if ( s.isEmpty() || n == 0 )
        return *this;

    int b = 0;
    while ( n && b <= (int)length() ) {
        int e = find( s, b );
        if ( e < 0 )
            e = length();
        if ( n == 1 )
            return mid( b, e - b );
        n--;
        b = e + s.length();
    }
    return "";
}


/*! Appends \a other to this string. */

void EString::append( const EString & other )
{
    if ( !other.length() )
        return;
    if ( !length() && ( !modifiable() || d->max < other.length() ) ) {
        // if this isn't modifiable, we just make a copy of the other
        // string. only sensible thing to do. if it's modifiable, but
        // we don't have enough bytes, we also just glue ourselves
        // onto the other. maybe we'll need to copy later, but maybe
        // not.
        *this = other;
        return;
    }
    reserve( length() + other.length() );
    memmove( d->str+d->len, other.d->str, other.d->len );
    d->len += other.d->len;
}


/*! \overload
    This version of append() appends \a num raw bytes from memory
    \a base. If \a base is null, this function does nothing.
*/

void EString::append( const char * base, uint num )
{
    if ( !base || !num )
        return;

    reserve( length() + num );
    memmove( d->str + d->len, base, num );
    d->len += num;
}


/*! \overload

    This version of append() appends the null-terminated string \a s,
    or does nothing if \a s is null.
*/

void EString::append( const char * s )
{
    if ( s )
        append( s, strlen( s ) );
}


/*! \overload
    This version of append() appends the single character \a c.
*/

void EString::append( char c )
{
    reserve( length() + 1 );
    d->str[d->len] = c;
    d->len++;
}


/*! Prepends \a other to this string. */

void EString::prepend( const EString & other )
{
    if ( other.isEmpty() )
        return;
    reserve( length() + other.length() );
    memmove( d->str + other.length(), d->str, length() );
    memmove( d->str, other.d->str, other.length() );
    setLength( length() + other.length() );
}


void EString::appendNumber( int n, int base )
{
    if ( n < 0 ) {
        append( '-' );
        n = -n;
    }
    appendNumber( (uint)n, base );
}

void EString::appendNumber( uint n, int base )
{
    EString s( fromNumber( n, base ) );

    if ( n > 0 )
        append( s );
    else
        append( "0" );
}

/*! Ensures that there is at least \a num bytes available in this
    string. This implicitly causes the string to become modifiable and
    have a nonzero number of available bytes.

    After calling reserve(), capacity() is at least as large as \a
    num, while length() has not changed.
*/

void EString::reserve( uint num )
{
    if ( num < length() )
        num = length();
    if ( !num )
        num = 1;
    if ( !d || d->max < num )
        reserve2( num );
}

/*! Equivalent to reserve(). reserve( \a num ) calls this function to
    do the heavy lifting. This function is not inline, while reserve()
    is, and calls to this function should be interesting wrt. memory
    allocation statistics.

    Noone except reserve() should call reserve2().
*/

void EString::reserve2( uint num )
{
    num = Allocator::rounded( num + sizeof( EStringData ) ) - sizeof( EStringData );

    EStringData * freeable = 0;
    if ( d && d->max )
        freeable = d;

    EStringData * nd = new( num ) EStringData( 0 );
    nd->max = num;
    nd->str = sizeof( EStringData ) + (char*)nd;
    if ( d )
        nd->len = d->len;
    if ( nd->len > num )
        nd->len = num;
    if ( d && d->len )
        memmove( nd->str, d->str, nd->len );
    d = nd;

    if ( freeable )
        Allocator::dealloc( freeable );
}


/*! Ensures that the string's length is either \a l or length(),
    whichever is smaller. If \a l is 0 (the default), the string will be
    empty after the function is called.
*/

void EString::truncate( uint l )
{
    if ( !l ) {
        if ( d && d->max )
            Allocator::dealloc( d );
        d = 0;
    }
    else if ( l < length() ) {
        detach();
        d->len = l;
    }
}


/*! Ensures that the string's length is \a l. If \a l is 0, the string
    will be empty after the function is called. If \a l is longer than
    the string used to be, the new part is uninitialised.
*/

void EString::setLength( uint l )
{
    reserve( l );
    if ( l )
        d->len = d->max;
    truncate( l );
}


/*! Returns a string containing the data starting at position \a start
    of this string, extending for \a num bytes. \a num may be left out,
    in which case the rest of the string is returned.

    If \a start is too large, an empty string is returned.
*/

EString EString::mid( uint start, uint num ) const
{
    if ( !d )
        num = 0;
    else if ( num > d->len || start + num > d->len )
        num = d->len - start;

    EString result;
    if ( !num || start >= length() )
        return result;

    d->max = 0;
    result.d = new EStringData;
    result.d->str = d->str + start;
    result.d->len = num;
    return result;
}


/*! Returns true is the string is quoted with \a c (default '"') as
    quote character and \a q (default '\') as escape character. \a c
    and \a q may be the same. */

bool EString::isQuoted( char c, char q ) const
{
    if ( length() < 2 || at( 0 ) != c || at( length() - 1 ) != c )
        return false;
    // skip past double escapes
    uint i = length() - 2;
    while ( i > 1 && d->str[i] == q && d->str[i-1] == q )
        i = i - 2;
    // empty string left?
    if ( i == 0 )
        return true;
    // trailing quote escaped?
    if ( d->str[i] == q )
        return false;
    return true;
}


/*! Returns the unquoted representation of the string if it isQuoted()
    and the string itself else.

    \a c at the start and end are removed; any occurence of \a c
    within the string is left alone; an occurence of \a q followed by
    \a c is converted into just \a c.
*/

EString EString::unquoted( char c, char q ) const
{
    if ( !isQuoted( c, q ) )
        return *this;
    EString r;
    r.reserve( length() );
    uint i = 1;
    while ( i < length()-1 ) {
        if ( d->str[i] == q )
            i++;
        r.append( d->str[i] );
        i++;
    }
    return r;
}


/*! Returns a version of this string quited with \a c, and where any
    occurences of \a c or \a q are escaped with \a q.
*/

EString EString::quoted( char c, char q ) const
{
    EString r;
    r.reserve( length()+2 );
    r.append( c );
    uint i = 0;
    while ( i < length() ) {
        if ( d->str[i] == c || d->str[i] == q )
            r.append( q );
        r.append( d->str[i] );
        i++;
    }
    r.append( c );
    return r;
}


/*! Returns true if this string is really boring, and false if it's
    empty or contains at least one character that may warrant quoting
    in some context. So far RFC 822 atoms, 2822 atoms, IMAP atoms and
    MIME tokens are considered.

    This function considers the intersection of those character
    classes to be the Totally boring subset. If \a b is not its
    default value, it may include other characters.
*/

bool EString::boring( Boring b ) const
{
    if ( isEmpty() )
        return false; // empty strings aren't boring - they may need quoting
    uint i = 0;
    bool exciting = false;
    while ( i < length() && !exciting ) {
        switch ( d->str[i] ) {
        case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'G':
        case 'H': case 'I': case 'J': case 'K': case 'L': case 'M': case 'N':
        case 'O': case 'P': case 'Q': case 'R': case 'S': case 'T': case 'U':
        case 'V': case 'W': case 'X': case 'Y': case 'Z':

        case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': case 'g':
        case 'h': case 'i': case 'j': case 'k': case 'l': case 'm': case 'n':
        case 'o': case 'p': case 'q': case 'r': case 's': case 't': case 'u':
        case 'v': case 'w': case 'x': case 'y': case 'z':

        case '0': case '1': case '2': case '3': case '4':
        case '5': case '6': case '7': case '8': case '9':

        case '!':
        case '#':
        case '$':
        case '&':
        case '+':
        case '-':
            break;

        case '.':
            if ( b != MIME )
                exciting = true;
            break;

        default:
            exciting = true;
            break;
        }
        i++;
    }
    if ( exciting ) // if we saw an exiting character...
        return false;
    return true;
}


/*! Returns a copy of this string where each run of whitespace is
    compressed to a single ASCII 32, and where leading and trailing
    whitespace is removed altogether.
*/

EString EString::simplified() const
{
    // scan for the first nonwhitespace character
    uint i = 0;
    uint first = 0;
    while ( i < length() && first == i ) {
        char c = d->str[i];
        if ( c == 9 || c == 10 || c == 13 || c == 32 )
            first++;
        i++;
    }
    // scan on to find the last nonwhitespace character and detect any
    // sequences of two or more whitespace characters within the
    // string.
    uint last = first;
    uint spaces = 0;
    bool identity = true;
    while ( identity && i < length() ) {
        char c = d->str[i];
        if ( c == 9 || c == 10 || c == 13 || c == 32 ) {
            spaces++;
        }
        else {
            if ( spaces > 1 )
                identity = false;
            spaces = 0;
            last = i;
        }
        i++;
    }
    if ( identity )
        return mid( first, last+1-first );

    EString result;
    result.reserve( length() );
    i = 0;
    spaces = 0;
    while ( i < length() ) {
        char c = d->str[i];
        if ( c == 9 || c == 10 || c == 13 || c == 32 ) {
            spaces++;
        }
        else {
            if ( spaces && !result.isEmpty() )
                result.append( ' ' );
            spaces = 0;
            result.append( c );
        }
        i++;
    }
    return result;
}


/*! Returns a copy of this string where leading and trailing
    whitespace have been removed.
*/

EString EString::trimmed() const
{
    uint i = 0;
    uint first = length();
    uint last = 0;
    while ( i < length() ) {
        char c = d->str[i];
        if ( c != 9 && c != 10 && c != 13 && c != 32 ) {
            if ( i < first )
                first = i;
            if ( i > last )
                last = i;
        }
        i++;
    }

    if ( last >= first )
        return mid( first, last + 1 - first );

    EString empty;
    return empty;
}


/*! Returns a copy of this EString with at most one trailing LF or CRLF
    removed. If there's more than one LF or CRLF, the remainder are
    left.
*/

EString EString::stripCRLF() const
{
    uint n = 0;
    if ( endsWith( "\r\n" ) )
        n = 2;
    else if ( endsWith( "\n" ) )
        n = 1;

    return mid( 0, length() - n );
}


/*! Returns the lowercase-hexadecimal representation of the string. */

EString EString::hex() const
{
    EString s;
    s.reserve( length()*2 );

    uint i = 0;
    while ( i < length() ) {
        uint x = d->str[i];
        s.appendNumber( x/16, 16 );
        s.appendNumber( x&15, 16 );
        i++;
    }

    return s;
}

const EString operator+( const EString & a, const EString & b )
{
    EString result;
    result.reserve( a.length() + b.length() );
    result.append( a );
    result.append( b );
    return result;
}


/*! Returns true if this string starts with \a prefix, and false if it
    does not.
*/

bool EString::startsWith( const EString & prefix ) const
{
    return length() >= prefix.length() &&
        prefix == mid( 0, prefix.length() );
}


/*! Returns true if this string starts with \a prefix, and false if it
    does not.
*/

bool EString::startsWith( const char * prefix ) const
{
    if ( !prefix )
        return true;
    uint i = 0;
    while ( prefix[i] && prefix[i] == at( i ) )
        i++;
    if ( prefix[i] )
        return false;
    return true;
}


/*! Returns true if this string ends with \a suffix, and false if it
    does not.
*/

bool EString::endsWith( const EString & suffix ) const
{
    return length() >= suffix.length() &&
        suffix == mid( length()-suffix.length() );
}


/*! Returns true if this string ends with \a suffix, and false if it
    does not.
*/

bool EString::endsWith( const char * suffix ) const
{
    if ( !suffix )
        return true;
    uint l = strlen( suffix );
    if ( l > length() )
        return false;
    uint i = 0;
    while ( i < l && suffix[i] == d->str[d->len - l + i] )
        i++;
    if ( i < l )
        return false;
    return true;
}


/*! Returns the number encoded by this string, and sets \a *ok to true
    if that number is valid, or to false if the number is invalid. By
    default the number is encoded in base 10, if \a base is specified
    that base is used. \a base must be at least 2 and at most 36.

    If the number is invalid (e.g. negative), number() returns 0.

    If \a ok is a null pointer, it is not modified.
*/

uint EString::number( bool * ok, uint base ) const
{
    uint i = 0;
    uint n = 0;

    bool good = !isEmpty();
    while ( good && i < length() ) {
        if ( d->str[i] < '0' || d->str[i] > 'z' )
            good = false;

        uint digit = d->str[i] - '0';

        // hex or something?
        if ( digit > 9 ) {
            uint c = d->str[i];
            if ( c > 'Z' )
                c = c - 32;
            digit = c - 'A' + 10;
        }

        // is the digit too large?
        if ( digit >= base )
            good = false;

        // Would n overflow if we multiplied by 10 and added digit?
        if ( n > UINT_MAX/base )
            good = false;
        n *= base;
        if ( n >= (UINT_MAX - UINT_MAX % base) && digit > (UINT_MAX % base) )
            good = false;
        n += digit;

        i++;
    }

    if ( !good )
        n = 0;

    if ( ok )
        *ok = good;

    return n;
}


/*! Returns a string representing the number \a n in the \a base
    system, which is 10 (decimal) by default and must be in the range
    2-36.

    For 0, "0" is returned.

    For bases 11-36, lower-case letters are used for the digits beyond
    9.
*/

EString EString::fromNumber( int64 n, uint base )
{
    EString r;
    r.appendNumber( n, base );
    return r;
}



/*! Converts \a n to a number in the \a base system and appends the
    result to this string. If \a n is 0, "0" is appended.

    Uses lower-case for digits above 9.
*/

void EString::appendNumber( int64 n, uint base )
{
    int64 top = 1;
    while ( top * base <= n )
        top = base * top;
    while ( top ) {
        uint d = ( n / top ) % base;
        char  c = '0' + d;
        if ( d > 9 )
            c = 'a' + d - 10;
        append( c );
        top = top / base;
    }
}


/*! Returns an \a e encoded version of this EString. If \a e is Base64,
    then \a n specifies the maximum line length.
    The default is 0, i.e. no limit.

    This function does not support Uuencode. If \a e is Uuencode, it
    returns the input string.
*/

EString EString::encoded( Encoding e, uint n ) const
{
    if ( e == Base64 )
        return e64( n );
    else if ( e == QP )
        return eQP( false, n > 0 );
    return *this;
}


/*! Returns a \a e decoded version of this EString. */

EString EString::decoded( Encoding e ) const
{
    if ( e == Base64 )
        return de64();
    else if ( e == QP )
        return deQP();
    else if ( e == Uuencode )
        return deUue();
    return *this;
}


/*! Returns a version of this EString with absolutely nothing changed.
    (This function is eventually intended to percent-escape URIs, the
    opposite of deURI().)
*/

EString EString::eURI() const
{
    return *this;
}


/*! Returns a version of this EString with every %xx escape replaced with
    the corresponding character (as used to encode URIs). Invalid escape
    sequences are left unchanged, so this function cannot be used for
    input from potentially malevolent sources.
*/

EString EString::deURI() const
{
    uint l = length();

    EString s;
    s.reserve( l );

    uint p = 0;
    while ( p < l ) {
        char c = d->str[p];
        if ( c == '%' ) {
            bool ok;
            uint n = mid( p+1, 2 ).number( &ok, 16 );
            if ( ok && l > p + 2 ) {
                p += 2;
                c = (char)n;
            }
        }
        s.append( c );
        p++;
    }

    return s;
}


/*! An implementation of uudecode, sufficient to handle some
    occurences of "content-transfer-encoding: x-uuencode"
    seen. Possibly not correct according to POSIX 1003.2b, who knows.
*/

EString EString::deUue() const
{
    if ( isEmpty() )
        return *this;
    uint i = 0;
    if ( !startsWith( "begin" ) ) {
        int begin = find( "\nbegin" );
        if ( begin < 0 )
            begin = find( "\rbegin" );
        if ( begin < 0 )
            return *this;
        i = (uint)begin+1;
    }
    EString r;
    while ( i < d->len ) {
        // step 0. skip over nonspace until CR/LF
        while ( i < d->len && d->str[i] != 13 && d->str[i] != 10 )
            i++;
        // step 1. skip over whitespace to the next length marker.
        while ( i < d->len &&
                ( d->str[i] == 9 || d->str[i] == 10 ||
                  d->str[i] == 13 || d->str[i] == 32 ) )
            i++;
        // step 2. the length byte, or the end line.
        uint linelength = 0;
        if ( i < d->len ) {
            char c = d->str[i];
            if ( c == 'e' && i < d->len - 2 &&
                 d->str[i+1] == 'n' && d->str[i+2] == 'd' &&
                 ( i + 3 == d->len ||
                   d->str[i+3] == 13 || d->str[i+3] == 10 ||
                   d->str[i+3] == 9 || d->str[i+3] == 32 ) )
                return r;
            else if ( c < 32 )
                return *this;
            else
                linelength = (c - 32) & 63;
            i++;
        }
        // step 3. the line data. we assume it's in groups of 4 tokens.
        while ( linelength && i < d->len ) {
            char c0 = 0, c1 = 0, c2 = 0, c3 = 0;
            if ( i < d->len )
                c0 = 63 & ( d->str[i] - 32 );
            if ( i+1 < d->len )
                c1 = 63 & ( d->str[i+1] - 32 );
            if ( i+2 < d->len )
                c2 = 63 & ( d->str[i+2] - 32 );
            if ( i+3 < d->len )
                c3 = 63 & ( d->str[i+3] - 32 );
            i += 4;
            if ( linelength > 0 ) {
                r.append( ( (c0 << 2) | (c1 >> 4) ) & 255 );
                linelength--;
            }
            if ( linelength > 0 ) {
                r.append( ( (c1 << 4) | (c2 >> 2) ) & 255 );
                linelength--;
            }
            if ( linelength > 0 ) {
                r.append( ( (c2 << 6) | (c3     ) ) & 255 );
                linelength--;
            }
        }
    }
    // we ran off the end without seeing an end line. what to do?
    // return what we've seen so far?
    return r;
}



static char from64[128] =
{
    64, 99, 99, 99,  99, 99, 99, 99,
    65, 99, 65, 99,  99, 65, 99, 99,
    99, 99, 99, 99,  99, 99, 99, 99,
    99, 99, 99, 99,  99, 99, 99, 99,

        // 32
    99, 99, 99, 99,  99, 99, 99, 99,
    99, 99, 99, 62,  99, 99, 99, 63,
    52, 53, 54, 55,  56, 57, 58, 59,
    60, 61, 99, 99,  99, 64, 99, 99,

        // 64
    99,  0,  1,  2,   3,  4,  5,  6,
     7,  8,  9, 10,  11, 12, 13, 14,
    15, 16, 17, 18,  19, 20, 21, 22,
    23, 24, 25, 99,  99, 99, 99, 99,

        // 96
    99, 26, 27, 28,  29, 30, 31, 32,
    33, 34, 35, 36,  37, 38, 39, 40,
    41, 42, 43, 44,  45, 46, 47, 48,
    49, 50, 51, 99,  99, 99, 99, 99
};



/*! Decodes this string using the base-64 algorithm and returns the result. */

EString EString::de64() const
{
    // this code comes from mailchen, adapted for EString.
    EString result;
    result.reserve( length() * 3 / 4 + 20 ); // 20 = fudge
    EString body;
    uint bp = 0;
    uint decoded = 0;
    int m = 0;
    uint p = 0;
    bool done = false;
    while ( p < length() && !done ) {
        uint c = d->str[p++];
        if ( c <= 'z' )
            c = from64[c];
        if ( c < 64 ) {
            switch ( m ) {
            case 0:
                decoded = c << 2;
                break;
            case 1:
                decoded += ( (c & 0xf0) >> 4 );
                result.d->str[bp++] = decoded;
                decoded = (c & 15) << 4;
                break;
            case 2:
                decoded += ( (c & 0xfc) >> 2 );
                result.d->str[bp++] = decoded;
                decoded = (c & 3) << 6;
                break;
            case 3:
                decoded += c;
                result.d->str[bp++] = decoded;
                break;
            }
            m = (m+1)&3;
        }
        else if ( c == 64 ) {
            done = true;
        }
        else if ( c == 65 ) {
            // white space; perfectly normal and may be ignored.
        }
        else {
            // we're supposed to ignore all other characters. so
            // that's what we do, even though it may not be ideal in
            // all cases... consider that later.
        }
    }
    result.d->len = bp;
    return result;
}


static char to64[65] =
    "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";

/*! Encodes this string using the base-64 algorithm and returns the
    result in lines of at most \a lineLength characters. If \a
    lineLength is not supplied, e64() returns a single line devoid of
    whitespace.
*/

EString EString::e64( uint lineLength ) const
{
    // this code comes from mailchen, adapted for EString
    int l = length();
    int i = 0;
    EString r;
    r.reserve( l*2 );
    int p = 0;
    uint c = 0;
    while ( i <= l-3 ) {
        r.d->str[p++] = to64[ ((d->str[i]>>2))&63 ];
        r.d->str[p++] = to64[ ((d->str[i]<<4)&48) + ((d->str[i+1]>>4)&15) ];
        r.d->str[p++] = to64[ ((d->str[i+1]<<2)&60) + ((d->str[i+2]>>6)&3) ];
        r.d->str[p++] = to64[ (d->str[i+2]&63) ];
        i += 3;
        c += 4;
        if ( lineLength > 0 && c >= lineLength ) {
            r.d->str[p++] = 13;
            r.d->str[p++] = 10;
            c = 0;
        }
    }
    if ( i < l ) {
        int i0, i1, i2;
        i0 = d->str[i];
        i1 = i+1 < l ? d->str[i+1] : 0;
        i2 = i+2 < l ? d->str[i+2] : 0;
        r.d->str[p++] = to64[ ((i0>>2))&63 ];
        r.d->str[p++] = to64[ ((i0<<4)&48) + ((i1>>4)&15) ];
        if ( i+1 < l )
            r.d->str[p++] = to64[ ((i1<<2)&60) + ((i2>>6)&3) ];
        else
            r.d->str[p++] = '=';
        if ( i+2 < l )
            r.d->str[p++] = to64[ (i2&63) ];
        else
            r.d->str[p++] = '=';
    }
    if ( lineLength > 0 && c > 0 ) {
        r.d->str[p++] = 13;
        r.d->str[p++] = 10;
    }
    r.d->len = p;
    return r;
}


/*! Decodes this string according to the quoted-printable algorithm,
    and returns the result. Errors are overlooked, to cope with all
    the mail-munging brokenware in the great big world.

    If \a underscore is true, underscores in the input are translated
    into spaces (as specified in RFC 2047).
*/

EString EString::deQP( bool underscore ) const
{
    uint i = 0;
    EString r;
    r.reserve( length() );
    while ( i < length() ) {
        if ( d->str[i] != '=' ) {
            char c = d->str[i++];
            if ( underscore && c == '_' )
                c = ' ';
            r.d->str[r.d->len++] = c;
        }
        else {
            // are we looking at = followed by end-of-line?
            bool ok = false;
            uint c = 0;
            bool eol = false;
            uint j = i+1;
            // skip possibly appended whitespace first
            while ( j < length() &&
                    ( d->str[j] == ' ' || d->str[j] == '\t' ) )
                j++;
            // there are two types of soft EOLs:
            if ( j < d->len && d->str[j] == 10 ) {
                eol = true;
                j++;
            }
            else if ( j < d->len-1 && d->str[j] == 13 && d->str[j+1] == 10 ) {
                eol = true;
                j = j + 2;
            }
            else if ( i + 2 < d->len ) {
                // ... and one common case: a two-digit hex number, not EOL
                c = mid( i+1, 2 ).number( &ok, 16 );
            }

            // write the proper decoded string and increase i.
            if ( eol ) { // ... if it's a soft EOL
                i = j;
            }
            else if ( ok ) { // ... or if it's a two-digit hex number
                r.d->str[r.d->len++] = c;
                i = i + 3;
            }
            else { // ... or if it's an error... we overlook it
                r.d->str[r.d->len++] = d->str[i++];
            }
        }
    }
    return r;
}


static char qphexdigits[17] = "0123456789ABCDEF";


static bool maybeBoundary( const EString & s, uint i ) {
    if ( s.length() < i + 2 )
        return false;
    if ( s[i] != '-' || s[i+1] != '-' )
        return false;

    while ( i < s.length() && s[i] >= ' ' ) {
        //bchars := bcharsnospace / " "
        //bcharsnospace := DIGIT / ALPHA / "'" / "(" / ")" /
        //                 "+" / "_" / "," / "-" / "." /
        //                 "/" / ":" / "=" / "?"
        switch( s[i] ) {
        case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': case 'g':
        case 'h': case 'i': case 'j': case 'k': case 'l': case 'm': case 'n':
        case 'o': case 'p': case 'q': case 'r': case 's': case 't': case 'u':
        case 'v': case 'w': case 'x': case 'y': case 'z':
        case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'G':
        case 'H': case 'I': case 'J': case 'K': case 'L': case 'M': case 'N':
        case 'O': case 'P': case 'Q': case 'R': case 'S': case 'T': case 'U':
        case 'V': case 'W': case 'X': case 'Y': case 'Z':
        case '0': case '1': case '2': case '3': case '4': case '5': case '6':
        case '7': case '8': case '9':
        case '\'':
        case '(': case ')': case '+': case '_': case ',': case '-': case '.':
        case '/': case ':': case '=': case '?':
        case ' ':
            // ok
            break;
        default:
            return false;
        }
        ++i;
    }
    return true;
}


/*! Encodes this string using the quoted-printable algorithm and
    returns the encoded version. In the encoded version, all line
    feeds are CRLF, and soft line feeds are positioned so that the q-p
    looks as good as it can.

    Note that this function is slightly incompatible with RFC 2646: It
    encodes trailing spaces, as suggested in RFC 2045, but RFC 2646
    suggest that if trailing spaces are the only reason to q-p, then
    the message should not be encoded.

    If \a underscore is present and true, this function uses the variant
    of q-p specified by RFC 2047, where a space is encoded as an
    underscore and a few more characters need to be encoded.

    If \a from is present and true, this function also makes sure that
    no output line starts with "From " or looks like a MIME boundary.
*/

EString EString::eQP( bool underscore, bool from ) const
{
    if ( isEmpty() )
        return *this;
    uint i = 0;
    EString r;
    // no input character can use more than six output characters (=
    // CR LF = 3 D), so we allocate as much space as we could possibly
    // need.
    r.reserve( length()*6 );
    uint c = 0;
    while ( i < d->len ) {
        if ( d->str[i] == 10 ||
             ( i < d->len-1 && d->str[i] == 13 && d->str[i+1] == 10 ) ) {
            // we have a line feed. if the last character on the line
            // was a space, we need to quote that to protect it.
            if ( r.d->len > 0 && r.d->str[r.d->len-1] == ' ' ) {
                r.d->str[r.d->len-1] = '=';
                r.d->str[r.d->len++] = '2';
                r.d->str[r.d->len++] = '0';
            }
            c = 0;
            if ( d->str[i] == 13 )
                r.d->str[r.d->len++] = d->str[i++];
            r.d->str[r.d->len++] = 10;
            // worst case: five bytes
        }
        else {
            if ( c > 72 ) {
                uint j = 1;
                while ( j < 10 && r.d->str[r.d->len-j] != ' ' )
                    j++;
                if ( j >= 10 )
                    j = 0;
                else
                    j--;
                uint k = 1;
                while ( k <= j ) {
                    r.d->str[r.d->len - k + 3] = r.d->str[r.d->len - k];
                    k++;
                }
                // always CRLF for soft linefeed
                r.d->str[r.d->len++ - j] = '=';
                r.d->str[r.d->len++ - j] = 13;
                r.d->str[r.d->len++ - j] = 10;
                c = j;
            }

            if ( underscore && d->str[i] == ' ' ) {
                r.d->str[r.d->len++] = '_';
                c += 1;
            }
            else if ( underscore &&
                      ! ( ( d->str[i] >= '0' && d->str[i] <= '9' ) ||
                          ( d->str[i] >= 'a' && d->str[i] <= 'z' ) ||
                          ( d->str[i] >= 'A' && d->str[i] <= 'Z' ) ) ) {
                r.d->str[r.d->len++] = '=';
                r.d->str[r.d->len++] = qphexdigits[d->str[i]/16];
                r.d->str[r.d->len++] = qphexdigits[d->str[i]%16];
                c += 3;
            }
            else if ( from && c == 0 && maybeBoundary( *this, i ) ) {
                r.d->str[r.d->len++] = '=';
                r.d->str[r.d->len++] = qphexdigits[d->str[i]/16];
                r.d->str[r.d->len++] = qphexdigits[d->str[i]%16];
                c += 3;
            }
            else if ( from && c == 0 && d->len >= i + 4 &&
                      d->str[i] == 'F' && d->str[i+1] == 'r' &&
                      d->str[i+2] == 'o' && d->str[i+3] == 'm' &&
                      d->str[i+4] == ' ' ) {
                r.d->str[r.d->len++] = '=';
                r.d->str[r.d->len++] = qphexdigits[d->str[i]/16];
                r.d->str[r.d->len++] = qphexdigits[d->str[i]%16];
                c += 3;
            }
            else if ( ( d->str[i] >= ' ' && d->str[i] < 127 &&
                        d->str[i] != '=' ) ||
                      ( d->str[i] == '\t' ) ) {
                r.d->str[r.d->len++] = d->str[i];
                c++;
            }
            else {
                r.d->str[r.d->len++] = '=';
                r.d->str[r.d->len++] = qphexdigits[d->str[i]/16];
                r.d->str[r.d->len++] = qphexdigits[d->str[i]%16];
                c += 3;
            }
        }
        i++;
    }
    return r;
}


/*! This function returns true if the string would need to be encoded
    using quoted-printable. It is a greatly simplified copy of eQP(),
    with the changes made necessary by RFC 2646.
*/

bool EString::needsQP() const
{
    uint i = 0;
    uint c = 0;
    while ( i < length() ) {
        if ( c == 0 && maybeBoundary( *this, i ) )
            return true;
        if ( c == 0 && d->str[i] == 'F' && d->str[i+1] == 'r' )
            return true;
        if ( d->str[i] == 10 )
             c = 0;
        else if ( c > 78 )
            return true;
        else if ( ( d->str[i] >= ' ' && d->str[i] < 127 ) ||
                  ( d->str[i] == '\t' ) ||
                  ( d->str[i] == 13 && d->str[i+1] == 10 ) )
            c++;
        else
            return true;
        i++;
    }
    return false;

}


/*! Returns -1 if this string is lexicographically before \a other, 0
    if they are the same, and 1 if this string is lexicographically
    after \a other.

    The comparison is case sensitive - just a byte comparison.
*/

int EString::compare( const EString & other ) const
{
    if ( d == other.d )
        return 0;
    uint i = 0;
    while ( i < length() && i < other.length() &&
            d->str[i] == other.d->str[i] )
        i++;
    if ( i >= length() && i >= other.length() )
        return 0;
    if ( i >= length() )
        return -1;
    if ( i >= other.length() )
        return 1;
    if ( d->str[i] < other.d->str[i] )
        return -1;
    return 1;
}


bool EString::operator<( const EString & other ) const
{
    return compare( other ) < 0;
}


bool EString::operator>( const EString & other ) const
{
    return compare( other ) > 0;
}


bool EString::operator<=( const EString & other ) const
{
    return compare( other ) <= 0;
}


bool EString::operator>=( const EString & other ) const
{
    return compare( other ) >= 0;
}


bool EString::operator<( const char * other ) const
{
    if ( !other )
        return false;

    uint l = length();
    uint i = 0;
    while ( i < l && d->str[i] == other[i] && other[i] )
        i++;
    // four cases: we ran out of characters, other did, both did,
    // there's a difference
    if ( i == l && !other[i] )
        return false;
    if ( !other[i] )
        return false;
    if ( i == l )
        return true;
    return d->str[i] < other[i];
}


/*! This function is a debugging aid. It prints the contents of the
    string within single quotes followed by a trailing newline to
    stderr.
*/

void EString::print() const
{
    uint i = 0;

    fprintf( stderr, "'" );
    while ( i < length() )
        fprintf( stderr, "%c", d->str[i++] );
    fprintf( stderr, "'\n" );
}


/*! Returns \a n as a string representing that number in a
    human-readable fashion optionally suffixed by K, M, G or T.

    The number is rounded more or less correctly.
*/

EString EString::humanNumber( int64 n )
{
    if ( n < 1024 )
        return fromNumber( n );

    int64 f = 1024;
    char s = 'K';
    if ( n < 1024 * 1024 ) {
        // ok
    }
    else if ( n < 1024 * 1024 * 1024 ) {
        f = 1024 * 1024;
        s = 'M';
    }
    else if ( n < 1024LL * 1024 * 1024 * 1024 ) {
        f = 1024 * 1024 * 1024;
        s = 'G';
    }
    else {
        // terabytes. we don't use petabytes or exabytes since people
        // don't know their abbreviations by heart.
        f = 1024LL * 1024 * 1024 * 1024;
        s = 'T';
    }

    EString r;
    // if it's single-digit, we add a decimal point. since we only go
    // to TB, not petabyte or exabyte, we don't need to check for
    // INT64_MAX/10. (actually we'd only need that check for exabytes.)
    if ( n < f * 10 ) {
        n += f/20-1;
        r = fromNumber( n/f );
        uint m = (n%f)/(f/10);
        r.append( '.' );
        r.append( '0' + m );
    }
    else {
        n += f/2-1;
        r = fromNumber( n/f );
    }
    r.append( s );
    return r;
}


// all the keywords we know about, found by grepping through message/*.cpp
static const char * keywords[] = {
    "7bit", "8bit", "alternative", "aug", "binary", "bcc", "cc", "comments",
    "content-description", "content-disposition", "content-id",
    "content-language", "content-location", "content-md5",
    "content-transfer-encoding", "content-type", "date", "fri", "from",
    "in-reply-to", "jun", "jul", "keywords", "may", "message-id",
    "mime-version", "mon", "orig-date",
    "received", "references", "reply-to", "resent-bcc", "resent-cc",
    "resent-date", "resent-from", "resent-message-id", "resent-sender",
    "resent-to", "return-path", "sender", "sep", "subject", "to",
    "us-ascii", "adt", "akdt", "akst", "apr", "ast", "attachment",
    "base64", "body", "boundary", "brt", "bst", "bytes", "cadt", "cast",
    "cct", "cdt", "ces", "cest", "cet", "charset", "cst", "cut", "data",
    "dec", "deleted", "digest", "eadt", "east", "edt", "eet", "est",
    "feb", "flag", "fri", "gmt", "grnlnddt", "grnlndst", "hadt", "hast",
    "helo", "hkt", "hst", "html", "id", "idate", "inline", "jan", "jst",
    "kdt", "kst", "lhlo", "lines", "lockuidnext", "mar", "mdt", "message",
    "mest", "mesz", "met", "metdst", "mez", "mezt", "mon", "msd", "msk",
    "mst", "multipart", "name", "ndt", "nov", "nst", "nzdt", "nzst", "oct",
    "part", "plain", "pdt", "pst", "quit", "quoted-printable", "rawbytes",
    "rfc822", "rfc822size", "root", "sast", "sat", "seen", "sep",
    "supplied", "text", "tue", "uid", "us-ascii", "ut", "utc", "value",
    "wadt", "wast", "wed", "wet", "ydt", "yst",
    "mixed",
    0
};

// helper for EString::anonymised()
static inline bool isMungableChar( char c ) {
    if ( ( c >= 'a' && c <= 'z' ) ||
         ( c >= 'A' && c <= 'Z' ) ||
         ( c >= '0' && c <= '9' ) ||
         ( c == '=' ||
           c == '"' ||
           c == ':' ||
           c == '?' ||
           c == '-' ||
           c == '(' ||
           c == ')' ||
           c == '_' ) )
        return true;
    return false;
}


/*! Returns a copy of this string where most/all content has been
    replaced with the letter 'x' or the digit '4', but if the message
    was an RFC 822 message, it keeps the same parse tree.

    Specifically, most ASCII words are changed to xxxx, while most/all
    syntax elements are kept.

    This function is very, very slow. That's okay since it's only used
    for sending bug reports to us, and we all know, that's not a common
    case.
*/

EString EString::anonymised() const
{
    uint b = 0;
    EString r;
    while ( b < length() ) {
        uint e = b;
        while ( e < d->len && ( d->str[e] > 127 ||
                                isMungableChar( d->str[e] ) ) )
            e++;
        // we have a word.
        bool munge = true;
        if ( e == b )
            munge = false;

        if ( munge && d->str[e-1] == ':' ) // header field names
            munge = false;

        if ( munge ) { // mime parameters
            uint i = b;
            while ( i < e && d->str[i] != '"' && d->str[i] != '=' )
                i++;
            if ( i < e )
                munge = false;
        }

        if ( munge && // boundary lines
             b + 2 <= e &&
             d->str[b] == '-' && d->str[b+1] == '-' ) {
            munge = false;
        }

        if ( munge ) { // any keyword
            EString m = mid( b, e-b ).lower();
            uint i = 0;
            while ( keywords[i] && m != keywords[i] )
                i++;
            if ( keywords[i] )
                munge = false;
        }

        if ( munge ) { // any word containing non-ascii
            uint i = b;
            while ( i < e && d->str[i] < 128 )
                i++;
            if ( i < e )
                munge = false;
        }

        if ( munge ) {
            uint i = 0;
            while ( b + i < e ) {
                char c = d->str[b+i];
                if ( c >= 'a' && c <= 'z' )
                    r.append( 'a' + (i%26) );
                else if ( c >= 'A' && c <= 'Z' )
                    r.append( 'a' + (i%26) );
                else
                    r.append( c );
                i++;
            }
        }
        else {
            r.append( mid( b, e-b ) );
        }
        b = e;

        while ( b < d->len && !isMungableChar( d->str[b] ) ) {
            r.append( d->str[b] );
            b++;
        }
    }

    return r;
}



/*! Returns a copy of this string where every linefeed is CRLF, and
    where the last two characters are CRLF.
*/

EString EString::crlf() const
{
    bool copy = true;
    if ( length() < 2 ||
         d->str[d->len-1] != 10 ||
         d->str[d->len-2] != 13 )
        copy = false;
    uint i = 0;
    while ( copy && i < d->len ) {
        if ( d->str[i] == 13 && i < d->len && d->str[i+1] == 10 )
            i += 2;
        else if ( d->str[i] == 13 || d->str[i] == 10 )
            copy = false;
        else
            i++;
    }
    if ( copy )
        return *this;

    EString r;
    r.reserve( length() );
    r.append( mid( 0, i ) );
    bool lf = false;
    uint len = 0;
    if ( d )
        len = d->len;
    while ( i < len ) {
        lf = false;
        char c = d->str[i++];

        if ( c == 10 ) {
            lf = true;
        }
        else if ( c == 13 ) {
            lf = true;
            if ( i < d->len && d->str[i] == 10 )
                i++;
            else if ( i < d->len-1 &&
                      d->str[i] == 13 && d->str[i+1] == 10 )
                i += 2;
        }

        if ( lf )
            r.append( "\r\n" );
        else
            r.append( c );
    }
    if ( !lf )
        r.append( "\r\n" );

    return r;
}


/*! Returns true if this string contains at least one instance of \a s. */

bool EString::contains( const EString & s ) const
{
    if ( find( s ) >= 0 )
        return true;
    return false;
}


/*! Returns true if this string contains at least one instance of \a c. */

bool EString::contains( const char c ) const
{
    if ( find( c ) >= 0 )
        return true;
    return false;
}


/*! Returns true if this string contains at least one instance of \a
    s, and the characters before and after the occurence aren't
    letters.
*/

bool EString::containsWord( const EString & s ) const
{
    int i = find( s );
    while ( i >= 0 ) {
        bool before = false;
        bool after = false;
        if ( i == 0 ) {
            before = true;
        }
        else {
            char c = d->str[i-1];
            if ( c < 'A' || ( c > 'Z' && c < 'a' ) || c > 'z' )
                before = true;
        }
        if ( i + s.length() == length() ) {
            after = true;
        }
        else {
            char c = d->str[i+s.length()];
            if ( c < 'A' || ( c > 'Z' && c < 'a' ) || c > 'z' )
                after = true;
        }
        if ( before && after )
            return true;
        i = find( s, i+1 );
    }
    return false;
}


/*! Returns a copy of this string wrapped so that each line contains
    at most \a linelength characters. The first line is prefixed by \a
    firstPrefix, subsequent lines by \a otherPrefix. If \a spaceAtEOL
    is true, all lines except the last end with a space.

    The prefixes are counted towards line length, but the optional
    trailing space is not.

    Only space (ASCII 32) is a line-break opportunity. If there are
    multiple spaces where a line is broken, all the spaces are
    replaced by a single CRLF. Linefeeds added use CRLF.
*/

EString EString::wrapped( uint linelength,
                        const EString & firstPrefix, const EString & otherPrefix,
                        bool spaceAtEOL ) const
{
    // result must be modifiable() at all times, otherwise we allocate
    // all the RAM.

    // working:
    EString result;
    result.reserve( length() );
    result.append( firstPrefix );
    // broken but should work. needs investigation.
    // EString result = firstPrefix;
    // result.reserve( length() );

    // move is where we keep the text that has to be moved to the next
    // line. it too should be modifiable() all the time.
    EString move;
    uint i = 0;
    uint linestart = 0;
    uint space = 0;
    while ( i < length() ) {
        char c = at( i );
        if ( c == ' ' )
            space = result.length();
        else if ( c == '\n' )
            linestart = result.length() + 1;
        result.append( c );
        i++;
        // add a soft linebreak?
        if ( result.length() > linestart + linelength && space > linestart ) {
            while ( space > 0 && result[space-1] == ' ' )
                space--;
            linestart = space + 1;
            while ( result[linestart] == ' ' )
                linestart++;
            move.truncate();
            if ( result.length() > linestart )
                move.append( result.cstr() + linestart );
            if ( spaceAtEOL )
                result.truncate( space + 1 );
            else
                result.truncate( space );
            result.append( "\r\n" );
            result.append( otherPrefix );
            result.append( move );
        }
    }
    return result;
}


/*! Replaces all occurences of \a a in this string with \a b. Rather
    slow and allocates much memory. Could be optimised if it ever
    shows up on the performance graphs.

    \a a must not be empty.

    Replaced sections are not considered when looking for the next
    match.
*/

void EString::replace( const EString & a, const EString & b )
{
    if ( a == b)
        return; // noop
    if ( a.isEmpty() )
        return; // infinite loop

    int i = find( a );
    while ( i >= 0 ) {
        EString r = mid( i+a.length() );
        truncate( i );
        append( b );
        append( r );
        i = find( a, i + b.length() );
    }
}