--- parser3/src/main/pa_string.C	2002/08/07 13:44:00	1.166
+++ parser3/src/main/pa_string.C	2012/06/20 21:01:20	1.243
@@ -1,828 +1,1002 @@
 /** @file
-	Parser: string class. @see untasize_t.C.
+	Parser: string class. @see untalength_t.C.
 
-	Copyright (c) 2001, 2002 ArtLebedev Group (http://www.artlebedev.com)
+	Copyright (c) 2001-2012 Art. Lebedev Studio (http://www.artlebedev.com)
 	Author: Alexandr Petrosian <paf@design.ru> (http://paf.design.ru)
 */
 
-static const char* IDENT_STRING_C="$Date: 2002/08/07 13:44:00 $";
-
-#include "pcre.h"
-
-#include "pa_pool.h"
 #include "pa_string.h"
-#include "pa_hash.h"
 #include "pa_exception.h"
-#include "pa_common.h"
-#include "pa_array.h"
-#include "pa_globals.h"
 #include "pa_table.h"
 #include "pa_dictionary.h"
 #include "pa_charset.h"
+#include "pa_vregex.h"
 
-#define DEBUG_STRING_APPENDS_VS_EXPANDS
+volatile const char * IDENT_PA_STRING_C="$Id: pa_string.C,v 1.243 2012/06/20 21:01:20 moko Exp $" IDENT_PA_STRING_H;
 
+const String String::Empty;
 
-#ifdef DEBUG_STRING_APPENDS_VS_EXPANDS
-ulong string_piece_appends=0;
-#endif
 
-String& String::OnPool(Pool& apool, const char *local_src, size_t src_size, bool tainted) {
-	if(local_src && *local_src) {
-		if(src_size==0)
-			src_size=strlen(local_src);
-		
-		char *pooled_src=(char *)apool.malloc(src_size);
-		memcpy(pooled_src, local_src, src_size);
-		return *new(apool) String(apool, pooled_src, src_size, tainted);
-	} else
-		return *new(apool) String(apool);
-}
-String::String(Pool& apool, const char *src, size_t src_size, bool tainted) :
-	Pooled(apool) {
-	last_chunk=&head.chunk;
-	head.chunk.count=CR_PREALLOCATED_COUNT;
-	append_here=head.chunk.rows;
+// pa_atoui is based on Manuel Novoa III _strto_l for uClibc
 
-	if(src)
-		if(tainted)
-			APPEND_TAINTED(src, src_size, 0, 0);
-		else
-			APPEND_CLEAN(src, src_size, 0, 0);
-}
+unsigned int pa_atoui(const char *str, int base, const String* problem_source){
+	unsigned int result = 0;
+	const char *pos = str;
 
-String::String(const String& src) :	
-	Pooled(src.pool()) {
-	last_chunk=&head.chunk;
-	head.chunk.count=CR_PREALLOCATED_COUNT;
-	append_here=head.chunk.rows;
+	while (isspace(*pos)) /* skip leading whitespace */
+		++pos;
 
-	append(src, UL_UNSPECIFIED);
-}
+	if (base == 16 && *pos == '0') { /* handle option prefix */
+		++pos;
+		if (*pos == 'x' || *pos == 'X') {
+			++pos;
+		}
+	}
 
-size_t  String::size() const {
-	size_t result=0;
-	STRING_FOREACH_ROW(
-			result+=row->item.size;
-	);
-	return result;
-}
+	if (base == 0) { /* dynamic base */
+		base = 10; /* default is 10 */
+		if (*pos == '0') {
+			++pos;
+			if (*pos == 'x' || *pos == 'X')
+				++pos;
+				base=16;
+		}
+	}
+
+	if (base < 2 || base > 16) { /* illegal base */
+		throw Exception(PARSER_RUNTIME, 0, "base to must be an integer from 2 to 16");
+	}
+
+	unsigned int cutoff = UINT_MAX / base;
+	int cutoff_digit = UINT_MAX - cutoff * base;
+
+	while(true) {
+		int digit;
+		
+		if ((*pos >= '0') && (*pos <= '9')) {
+			digit = (*pos - '0');
+		} else if (*pos >= 'a') {
+			digit = (*pos - 'a' + 10);
+		} else if (*pos >= 'A') {
+			digit = (*pos - 'A' + 10);
+		} else break;
+
+		if (digit >= base) {
+			break;
+		}
+		
+		++pos;
+		
+		/* adjust number, with overflow check */
+		if ((result > cutoff) || ((result == cutoff) && (digit > cutoff_digit))) {
+			throw Exception("number.format", problem_source, problem_source ? "out of range (int)" : "'%s' is out of range (int)", str);
+		} else {
+			result  = result * base + digit;
+		}
+	}
+
+	while(char c=*pos++)
+		if(!isspace(c))
+			throw Exception("number.format", problem_source, problem_source ? "invalid number (int)" : "'%s' is invalid number (int)", str);
 
-/// @todo not very optimal
-uint String::used_rows() const {
-	uint result=0;
-	STRING_FOREACH_ROW(
-		result++;
-	);
 	return result;
 }
-void String::expand() {
-	uint new_chunk_count=last_chunk->count+CR_GROW_COUNT;
-	if(new_chunk_count>max_integral(Chunk::count_type))
-		new_chunk_count=max_integral(Chunk::count_type);
-
-	Chunk *new_chunk=static_cast<Chunk *>(malloc(
-			sizeof(Chunk)// count+interpadding(?)+rows[CR_PREALLOCATED_COUNT]+tailpadding(??)
-			-sizeof(Chunk::rows_type) // PREALLOCATED rows
-			+sizeof(Chunk::Row)*new_chunk_count // neaded rows
-			+sizeof(Chunk *) // link size
-		, 10));
-	new_chunk->rows[new_chunk->count=new_chunk_count].link=0;
-	last_chunk->rows[last_chunk->count].link=new_chunk;
+
+int pa_atoi(const char* str, const String* problem_source) {
+	if(!str)
+		return 0;
+
+	while(isspace(*str))
+		str++;
+
+	if(!*str)
+		return 0;
+
+	bool negative=false;
+	if(str[0]=='-') {
+		negative=true;
+		str++;
+	} else if(str[0]=='+') {
+		str++;
+	}
+
+	unsigned int result=pa_atoui(str, 0, problem_source);
+
+	if(negative && result <= ((unsigned int)(-(1+INT_MIN)))+1)
+		return -(int)result;
+	
+	if(result<=INT_MAX)
+		return (int)result;
 	
-	last_chunk=new_chunk;
-	append_here=last_chunk->rows;
+	throw Exception("number.format", problem_source, problem_source ? "out of range (int)" : "'%s' is out of range (int)", str);
 }
 
-String& String::real_append(STRING_APPEND_PARAMS) {
-	if(!last_chunk) // growth stopped [we're appended as string to somebody]
-		throw Exception(0,
-			this,
-			"string growth stopped (append cstr)");
-
-	if(!src)
-		return *this;
-	if(!size)
-		size=strlen(src);
-	if(!size)
-		return *this;
+double pa_atod(const char* str, const String* problem_source) {
+	if(!str)
+		return 0;
 
-#ifdef DEBUG_STRING_APPENDS_VS_EXPANDS
-	string_piece_appends++;
-#endif
+	while(isspace(*str))
+		str++;
 
-	// manually unrolled to avoid extra check
-	while(size>max_integral(Chunk::Row::item_size_type)) {
-		if(chunk_is_full())
-			expand();
-
-		append_here->item.ptr=src;
-		append_here->item.size=max_integral(Chunk::Row::item_size_type);
-		append_here->item.lang=lang;
-#ifndef NO_STRING_ORIGIN
-		append_here->item.origin.file=file;
-		append_here->item.origin.line=line;
-#endif
-		append_here++;
+	if(!*str)
+		return 0;
 
-		src+=max_integral(Chunk::Row::item_size_type);
-		size-=max_integral(Chunk::Row::item_size_type);
+	bool negative=false;
+	if(str[0]=='-') {
+		negative=true;
+		str++;
+	} else if(str[0]=='+') {
+		str++;
 	}
 
-	if(chunk_is_full())
-		expand();
+	double result;
+	if(str[0]=='0')
+		if(str[1]=='x' || str[1]=='X'){
+			// 0xABC
+			result=(double)pa_atoui(str, 0, problem_source);
+			return negative ? -result : result;
+		} else {
+			 // skip leading 0000, to disable octal interpretation
+			do str++; while(*str=='0');
+		}
 
-	append_here->item.ptr=src;
-	append_here->item.size=size;
-	append_here->item.lang=lang;
-#ifndef NO_STRING_ORIGIN
-	append_here->item.origin.file=file;
-	append_here->item.origin.line=line;
-#endif
-	append_here++;
+	char *error_pos;
+	result=strtod(str, &error_pos);
 
-	return *this;
+	while(char c=*error_pos++)
+		if(!isspace((unsigned char)c))
+			throw Exception("number.format", problem_source, problem_source ? "invalid number (double)" : "'%s' is invalid number (double)", str);
+
+	return negative ? -result : result;
 }
 
-char String::first_char() const {
-	if(is_empty())
-		throw Exception(0,
-			this,
-			"getting first char of empty string");
+// cord lib extension
 
-	return *head.chunk.rows[0].item.ptr;
+#ifndef DOXYGEN
+typedef struct {
+    ssize_t countdown;
+    int target;	/* Character we're looking for	*/
+} chr_data;
+#endif
+static int CORD_range_contains_chr_greater_then_proc(char c, size_t size, void* client_data)
+{
+    register chr_data * d = (chr_data *)client_data;
+    
+    if (d -> countdown<=0) return(2);
+    d -> countdown -= size;
+    if (c > d -> target) return(1);
+    return(0);
+}
+int CORD_range_contains_chr_greater_then(CORD x, size_t i, size_t n, int c)
+{
+    chr_data d;
+
+    d.countdown = n;
+    d.target = c;
+    return(CORD_block_iter(x, i, CORD_range_contains_chr_greater_then_proc, &d) == 1/*alternatives: 0 normally ended, 2=struck 'n'*/);
+}
+
+static int CORD_block_count_proc(char /*c*/, size_t /*size*/, void* client_data)
+{
+    int* result=(int*)client_data;
+    (*result)++;
+    return(0); // 0=continue
+}
+size_t CORD_block_count(CORD x)
+{
+	size_t result=0;
+	CORD_block_iter(x, 0, CORD_block_count_proc, &result);
+    return result;
 }
 
-uint String::hash_code() const {
-	uint result=0;
-	STRING_FOREACH_ROW(
-			result=Hash::generic_code(result, row->item.ptr, row->item.size);
-	);
-	return result;
+// helpers
+
+/// String::match uses this as replace & global search table columns
+
+const int MAX_MATCH_GROUPS=100;
+
+class String_match_table_template_columns: public ArrayString {
+public:
+	String_match_table_template_columns() {
+		*this+=new String("prematch");
+		*this+=new String("match");
+		*this+=new String("postmatch");
+		for(int i=0; i<MAX_MATCH_GROUPS; i++) {
+			*this+=new String(String::Body::Format(1+i), String::L_CLEAN);
+		}
+	}
+};
+
+Table string_match_table_template(new String_match_table_template_columns);
+
+// String::Body methods
+
+String::Body String::Body::Format(int value) {
+	char local[MAX_NUMBER];
+	size_t length=snprintf(local, MAX_NUMBER, "%d", value);
+	return String::Body(pa_strdup(local, length));
 }
 
-/// @todo move 'lang' skipping to pos
-int String::cmp(int& partial, const String& src, 
-				size_t this_offset, Untaint_lang lang) const {
-	partial=-1;
-	size_t a_size=size();
-	this_offset=min(this_offset, a_size-1);
-
-	const Chunk *a_chunk=&head.chunk;
-	const Chunk *b_chunk=&src.head.chunk;
-	const Chunk::Row *a_row=a_chunk->rows;
-	const Chunk::Row *b_row=b_chunk->rows;
-	size_t a_offset=this_offset;
-	size_t b_offset=0;
-	Chunk::Row *a_end=append_here;
-	Chunk::Row *b_end=src.append_here;
-	uint a_countdown=a_chunk->count;
-	uint b_countdown=b_chunk->count;
-	int result;
-	size_t pos=0; 
-
-	bool a_break=a_size==0;
-	bool b_break=src.is_empty();
-	if(!(a_break || b_break)) while(true) {
-		if(pos+a_row->item.size > this_offset) {
-			if(lang!=UL_UNSPECIFIED && a_row->item.lang>lang) 
-				return -1; // wrong lang -- bail out
-
-			int size_diff=
-				(a_row->item.size-a_offset)-
-				(b_row->item.size-b_offset);
-			
-			if(size_diff==0) { // a has same size as b
-				result=memcmp(a_row->item.ptr+a_offset, b_row->item.ptr+b_offset, 
-					a_row->item.size-a_offset);
-				if(result)
-					return result;
-				pos+=a_row->item.size;
-				a_row++; a_countdown--; a_offset=0;
-				b_row++; b_countdown--; b_offset=0;
-			} else if (size_diff>0) { // a longer
-				result=memcmp(a_row->item.ptr+a_offset, b_row->item.ptr+b_offset, 
-					b_row->item.size-b_offset);
-				if(result)
-					return result;
-				a_offset+=b_row->item.size-b_offset;
-				b_row++; b_countdown--; b_offset=0;
-			} else { // b longer
-				result=memcmp(a_row->item.ptr+a_offset, b_row->item.ptr+b_offset, 
-					a_row->item.size-a_offset);
-				if(result)
-					return result;
-				b_offset+=a_row->item.size-a_offset;
-				pos+=a_row->item.size;
-				a_row++; a_countdown--; a_offset=0;
-			}
-			if(b_break=b_row==b_end) {
-				a_break=a_row==a_end;
-				break;			
+String::Body String::Body::trim(String::Trim_kind kind, const char* chars, 
+								size_t* out_start, size_t* out_length, Charset* source_charset) const {
+	size_t our_length=length();
+	if(!our_length)
+		return *this;
+
+	// check if any UTF-8 in chars
+	bool fast=true;
+	if(chars && source_charset && source_charset->isUTF8()){
+		const char* pos=chars;
+		while(unsigned char c=*pos++)
+			if(c>127){
+				fast=false;
+				break;
 			}
-			if(!b_countdown) {
-				b_chunk=b_row->link;
-				b_row=b_chunk->rows;
-				b_countdown=b_chunk->count;
+	}
+
+	size_t start=0;
+	size_t end=our_length;
+	if(!chars)
+		chars=" \t\n"; // white space
+
+	if(fast){
+		// from left...
+		if(kind!=TRIM_END) {
+			CORD_pos pos; set_pos(pos, 0);
+			while(true) {
+				char c=CORD_pos_fetch(pos);
+				if(strchr(chars, c)) {
+					if(++start==our_length)
+						return 0; // all chars are empty, just return empty string
+				} else
+					break;			
+
+				CORD_next(pos);
 			}
-		} else {
-			a_offset-=a_row->item.size;
-			pos+=a_row->item.size;
-			a_row++; a_countdown--; 
 		}
 
-		if(a_break=a_row==a_end) {
-			b_break=b_row==b_end;
-			break;
-		}
-		if(!a_countdown) {
-			a_chunk=a_row->link;
-			a_row=a_chunk->rows;
-			a_countdown=a_chunk->count;
+		// from right..
+		if(kind!=TRIM_START) {
+			CORD_pos pos; set_pos(pos, end-1);
+			while(true) {
+				char c=CORD_pos_fetch(pos);
+				if(strchr(chars, c)) {
+					if(--end==0) // optimization: NO need to check for 'end>=start', that's(<) impossible
+						return 0; // all chars are empty, just return empty string
+				} else
+					break;			
+
+				CORD_prev(pos);
+			}
 		}
-	}
-	if(a_break==b_break) { // ended simultaneously
-		partial=0; return 0;
-	} else if(a_break) { // first bytes equal, but a ended before b
-		partial=1; return -1;
 	} else {
-		partial=2; return +1;
-	}
-}
+		const XMLByte* src_begin=(const XMLByte*)cstr();
+		const XMLByte* src_end=src_begin+our_length;
 
-/// @todo move 'lang' skipping to pos
-int String::cmp(int& partial, const char* b_ptr, size_t src_size, 
-				size_t this_offset, Untaint_lang lang) const {
-	partial=-1;
-	size_t a_size=size();
-	size_t b_size=src_size?src_size:b_ptr?strlen(b_ptr):0;
-	this_offset=min(this_offset, a_size-1);
-
-	const Chunk *a_chunk=&head.chunk;
-	const Chunk::Row *a_row=a_chunk->rows;
-	size_t a_offset=this_offset;
-	size_t b_offset=0;
-	Chunk::Row *a_end=append_here;
-	uint a_countdown=a_chunk->count;
-	size_t pos=0;
-
-	bool a_break=a_size==0;
-	bool b_break=b_size==0;
-	if(!(a_break || b_break)) while(true) {
-		if(pos+a_row->item.size > this_offset) {
-			if(lang!=UL_UNSPECIFIED && a_row->item.lang>lang) 
-				return -1; // wrong lang -- bail out
-
-			int size_diff=
-				(a_row->item.size-a_offset)-
-				(b_size-b_offset);
-			
-			if(size_diff==0) { // a has same size as b
-				if(int result=memcmp(a_row->item.ptr+a_offset, b_ptr+b_offset, 
-					a_row->item.size-a_offset)!=0)
-					return result;
-				pos+=a_row->item.size;
-				a_row++; a_countdown--; a_offset=0;
-				b_break=true;
-			} else if (size_diff>0) { // a longer
-				if(int result=memcmp(a_row->item.ptr+a_offset, b_ptr+b_offset, 
-					b_size-b_offset)!=0)
-					return result;
-				a_offset+=b_size-b_offset;
-				b_break=true;
-			} else { // b longer
-				if(int result=memcmp(a_row->item.ptr+a_offset, b_ptr+b_offset, 
-					a_row->item.size-a_offset)!=0)
-					return result;
-				b_offset+=a_row->item.size-a_offset;
-				pos+=a_row->item.size;
-				a_row++; a_countdown--; a_offset=0;
+		// from left...
+		if(kind!=TRIM_END) {
+			while(src_begin<src_end){
+				uint char_length=1;
+				const XMLByte* ptr=src_begin;
+				// searching first UTF-8 byte: http://tools.ietf.org/html/rfc3629#section-3
+				while(++src_begin<=src_end && (*src_begin>127 && *src_begin<0xC0))
+					char_length++;
+
+				bool found=false;
+				for(const char* chars_byte=chars; chars_byte=strchr(chars_byte, *ptr); chars_byte++)
+					if(strncmp(chars_byte, (const char*)ptr, char_length)==0){
+						found=true;
+						break;
+					}
+
+				if(found){
+					start+=char_length;
+					if(start==our_length)
+						return 0; // all chars are empty, just return empty string
+				} else
+					break;
 			}
-		} else {
-			a_offset-=a_row->item.size; 
-			pos+=a_row->item.size;
-			a_row++; a_countdown--; 
 		}
 
-		a_break=a_row==a_end;
-		if(a_break || b_break)
-			break;
-		if(!a_countdown) {
-			a_chunk=a_row->link;
-			a_row=a_chunk->rows;
-			a_countdown=a_chunk->count;
+		// from right..
+		if(kind!=TRIM_START) {
+			while(src_begin<src_end){
+				uint char_length=1;
+				// searching first UTF-8 byte: http://tools.ietf.org/html/rfc3629#section-3
+				while(src_begin<=--src_end && (*src_end>127 && *src_end<0xC0))
+					char_length++;
+
+				bool found=false;
+				for(const char* chars_byte=chars; chars_byte=strchr(chars_byte, *src_end); chars_byte++)
+					if(strncmp(chars_byte, (const char*)src_end, char_length)==0){
+						found=true;
+						break;
+					}
+
+				if(found){
+					end-=char_length;
+					if(end==0)
+						return 0; // all chars are empty, just return empty string
+				} else
+					break;
+			}
 		}
 	}
-	if(a_break==b_break) { // ended simultaneously
-		partial=0; return 0;
-	} else if(a_break) { // first bytes equal, but a ended before b
-		partial=1; return -1;
+
+	if(start==0 && end==our_length) // nobody moved a thing
+		return *this;
+
+	if(out_start)
+		*out_start=start;
+	size_t new_length=end-start;
+	if(out_length)
+		*out_length=new_length;
+
+	return mid(start, new_length);
+}
+
+static int CORD_batched_iter_fn_generic_hash_code(char c, void * client_data) {
+	uint& result=*static_cast<uint*>(client_data);
+	generic_hash_code(result, c);
+	return 0;
+}
+static int CORD_batched_iter_fn_generic_hash_code(const char*  s, void * client_data) {
+	uint& result=*static_cast<uint*>(client_data);
+	generic_hash_code(result, s);
+	return 0;
+};
+uint String::Body::get_hash_code() const {
+#ifdef HASH_CODE_CACHING
+	if(hash_code)
+		return hash_code;
+#else
+	uint hash_code=0;
+#endif
+	if (body && CORD_IS_STRING(body)){
+		generic_hash_code(hash_code, body);
 	} else {
-		partial=2; return +1;
+		CORD_iter5(body, 0,
+			CORD_batched_iter_fn_generic_hash_code, 
+			CORD_batched_iter_fn_generic_hash_code, &hash_code);
+	}
+	return hash_code;
+}
+
+struct CORD_pos_info {
+	const char* chars;
+	size_t left;
+	size_t pos;
+};
+
+// can be called only for IS_FUNCTION(CORD) which is used in String::Body::strrpbrk
+static int CORD_iter_fn_rpos(char c, CORD_pos_info* info) {
+	if(info->pos < info->left){
+		info->pos=STRING_NOT_FOUND;
+		return 1;
+	}
+	if(strchr(info->chars, c))
+		return 1;
+	--(info->pos);
+	return 0;
+}
+
+size_t String::Body::strrpbrk(const char* chars, size_t left, size_t right) const {
+	if(is_empty() || !chars || !strlen(chars))
+		return STRING_NOT_FOUND;
+	CORD_pos_info info={chars, left, right};
+	if(CORD_riter4(body, right, (CORD_iter_fn)CORD_iter_fn_rpos, &info))
+		return info.pos;
+	else
+		return STRING_NOT_FOUND;
+}
+
+
+// can be called only for IS_FUNCTION(CORD) which is used in String::Body::rskipchars
+static int CORD_iter_fn_rskip(char c, CORD_pos_info* info) {
+	if(info->pos < info->left) {
+		info->pos=STRING_NOT_FOUND;
+		return 1;
 	}
+	if(!strchr(info->chars, c))
+		return 1;
+	--(info->pos);
+	return 0;
 }
 
-#ifndef NO_STRING_ORIGIN
-const Origin& String::origin() const { 
-	if(is_empty()) {
-		static const Origin empty_origin={"empty string"};
-		return empty_origin;
+size_t String::Body::rskipchars(const char* chars, size_t left, size_t right) const {
+	if(is_empty() || !chars || !strlen(chars))
+		return STRING_NOT_FOUND;
+	CORD_pos_info info={chars, left, right};
+	if(CORD_riter4(body, right, (CORD_iter_fn)CORD_iter_fn_rskip, &info))
+		return info.pos;
+	else
+		return STRING_NOT_FOUND;
+}
+
+// String methods
+
+String& String::append_know_length(const char* str, size_t known_length, Language lang) {
+	if(!known_length)
+		return *this;
+
+	// first: langs
+	langs.append(body, lang, known_length);
+	// next: letters themselves
+	body.append_know_length(str, known_length);
+
+	ASSERT_STRING_INVARIANT(*this);
+	return *this;
+}
+String& String::append_help_length(const char* str, size_t helper_length, Language lang) {
+	if(!str)
+		return *this;
+	size_t known_length=helper_length?helper_length:strlen(str);
+	if(!known_length)
+		return *this;
+
+	return append_know_length(str, known_length, lang);
+}
+String::String(int value, char *format) : langs(L_CLEAN){
+	char buf[MAX_NUMBER];
+	body.append_strdup_know_length(buf, snprintf(buf, MAX_NUMBER, format, value));
+}
+String& String::append_strdup(const char* str, size_t helper_length, Language lang) {
+	size_t known_length=helper_length?helper_length:strlen(str);
+	if(!known_length)
+		return *this;
+
+	// first: langs
+	langs.append(body, lang, known_length);
+	// next: letters themselves
+	body.append_strdup_know_length(str, known_length);
+
+	ASSERT_STRING_INVARIANT(*this);
+	return *this;
+}
+
+struct CORD_length_info {
+	size_t len;
+	size_t skip;
+};
+
+int CORD_batched_len(const char* s, CORD_length_info* info){
+	info->len += lengthUTF8( (const XMLByte *)s, (const XMLByte *)s+strlen(s));	return 0;
+}
+
+// can be called only for IS_FUNCTION(CORD) which are used in large String::Body::mid
+int CORD_batched_len(const char c, CORD_length_info* info){
+	if (info->skip==0){
+		info->len++;
+		info->skip = lengthUTF8Char(c)-1;
+	} else {
+		info->skip--;
 	}
-	
-	// determining origin by first piece or last appended piece
-	// because any of them can be constant=without origin: 
-	// ex: ^load[/file] "document_root" + "/file"
-	// when last peice is constant, 
-	// ex: parser_root_auto_path{dynamic} / auto.p{const}
-	// using first piece
-	Origin& first_origin=head.chunk.rows[0].item.origin;
-	return first_origin.file ? first_origin : append_here[-1].item.origin;
+	return 0;
 }
-#endif
 
-String& String::mid(size_t start, size_t finish) const {
-	String& result=*NEW String(pool());
+size_t String::length(Charset& charset) const {
+	if(charset.isUTF8()){
+		CORD_length_info info = {0, 0};
+		body.for_each<CORD_length_info *>(CORD_batched_len, CORD_batched_len, &info);
+		return info.len;
+	} else
+		return body.length();
+}
 
-	start=min(start, size());
-	finish=max(finish, start);
-	if(start==finish)
+/// @todo check in doc: whether it documents NOW bad situation "abc".mid(-1, 3) =were?="ab"
+String& String::mid(size_t substr_begin, size_t substr_end) const {
+	String& result=*new String;
+
+	size_t self_length=length();
+	substr_begin=min(substr_begin, self_length);
+	substr_end=min(max(substr_end, substr_begin), self_length);
+	size_t substr_length=substr_end-substr_begin;
+	if(!substr_length)
 		return result;
 
-	size_t pos=0;
-	STRING_FOREACH_ROW(
-		size_t item_finish=pos+row->item.size;
-		if(item_finish > start) { // started now or already?
-			bool started=result.is_empty(); // started now?
-			bool finished=finish <= item_finish; // finished now?
-			size_t offset=started?start-pos:0;
-			size_t size=finished?finish-pos:row->item.size;
-			result.APPEND(
-				row->item.ptr+offset, size-offset, 
-				row->item.lang,
-				row->item.origin.file, row->item.origin.line);
-			if(finished)
-				goto break2;
-		}
-		pos+=row->item.size;
-	);
-break2:
-//	SAPI::log(pool(), "piece of '%s' from %d to %d is '%s'",
-		//cstr(), start, finish, result.cstr());
+	// first: their langs
+	result.langs.append(result.body, langs, substr_begin, substr_length);
+	// next: letters themselves
+	result.body=body.mid(substr_begin, substr_length);
+
+	ASSERT_STRING_INVARIANT(result);
 	return result;
 }
 
-int String::pos(const String& substr, 
-				int result, Untaint_lang lang) const {
-	size_t self_size=size();
-	for(; size_t(result)<self_size; result++) {
-		int partial; cmp(partial, substr, result, lang);
-		if(
-			partial==0 || // full match
-			partial==2) // 'substr' starts 'this'+'result'
+// from, to and helper_length in characters, not in bytes (it's important for utf-8)
+String& String::mid(Charset& charset, size_t from, size_t to, size_t helper_length) const {
+	String& result=*new String;
+
+	size_t self_length=(helper_length)?helper_length:length(charset);
+
+	if(!self_length)
+		return result;
+
+	from=min(min(to, from), self_length);
+	to=min(max(to, from), self_length);
+
+	size_t substr_length=to-from;
+
+	if(!substr_length)
+		return result;
+
+	if(charset.isUTF8()){
+		const XMLByte* src_begin=(const XMLByte*)cstr();
+		const XMLByte* src_end=src_begin+body.length();
+
+		// convert 'from' and 'substr_length' from 'characters' to 'bytes'
+		from=getUTF8BytePos(src_begin, src_end, from);
+		substr_length=getUTF8BytePos(src_begin+from, src_end, substr_length);
+		if(!substr_length)
 			return result;
 	}
-	
-	return -1;
+
+	// first: their langs
+	result.langs.append(result.body, langs, from, substr_length);
+	// next: letters themselves
+	result.body=body.mid(from, substr_length);
+
+	ASSERT_STRING_INVARIANT(result);
+	return result;
 }
 
-int String::pos(const char *substr, size_t substr_size, 
-				int result, Untaint_lang lang) const {
-	size_t self_size=size();
-	for(; size_t(result)<self_size; result++) {
-		int partial; cmp(partial, substr, substr_size, result, lang);
-		if(
-			partial==0 || // full match
-			partial==2) // 'substr' starts 'this'+'result'
-			return result;
+size_t String::pos(const String::Body substr, size_t this_offset, Language lang) const {
+	size_t substr_length=substr.length();
+	while(true) {
+		size_t substr_begin=body.pos(substr, this_offset);
+		
+		if(substr_begin==CORD_NOT_FOUND)
+			return STRING_NOT_FOUND;
+
+		if(langs.check_lang(lang, substr_begin, substr_length))
+			return substr_begin;
+
+		this_offset=substr_begin+substr_length;
+	}
+}
+
+size_t String::pos(const String& substr, 
+				size_t this_offset, Language lang) const {
+	return pos(substr.body, this_offset, lang);
+}
+
+size_t String::pos(Charset& charset, const String& substr, 
+				size_t this_offset, Language lang) const {
+
+	if(charset.isUTF8()){
+		const XMLByte* srcPtr=(const XMLByte*)cstr();
+		const XMLByte* srcEnd=srcPtr+body.length();
+
+		// convert 'this_offset' from 'characters' to 'bytes'
+		this_offset=getUTF8BytePos(srcPtr, srcEnd, this_offset);
+
+		size_t result=pos(substr.body, this_offset, lang);
+		return (result==CORD_NOT_FOUND)
+			? STRING_NOT_FOUND
+			: getUTF8CharPos(srcPtr, srcEnd, result); // convert 'result' from 'bytes' to 'characters'
+	} else {
+		size_t result=pos(substr.body, this_offset, lang);
+		return (result==CORD_NOT_FOUND)
+			? STRING_NOT_FOUND
+			: result;
 	}
-	
-	return -1;
 }
 
-void String::split(Array& result, 
-				   size_t* pos_after_ref, 
-				   const char *delim, size_t delim_size, 
-				   Untaint_lang lang, int limit) const {
-	size_t self_size=size();
-	if(delim_size) {
-		size_t pos_after=pos_after_ref?*pos_after_ref:0;
-		int pos_before;
+void String::split(ArrayString& result, 
+		   size_t& pos_after, 
+		   const char* delim, 
+		   Language lang, int limit) const {
+	if(is_empty())
+		return;
+	size_t self_length=length();
+	if(size_t delim_length=strlen(delim)) {
+		size_t pos_before;
 		// while we have 'delim'...
-		for(; (pos_before=pos(delim, delim_size, pos_after, lang))>=0 && limit; limit--) {
+		for(; (pos_before=pos(delim, pos_after, lang))!=STRING_NOT_FOUND && limit; limit--) {
 			result+=&mid(pos_after, pos_before);
-			pos_after=pos_before+delim_size;
+			pos_after=pos_before+delim_length;
 		}
 		// last piece
-		if(pos_after<self_size && limit) {
-			result+=&mid(pos_after, self_size);
-			pos_after=self_size;
+		if(pos_after<self_length && limit) {
+			result+=&mid(pos_after, self_length);
+			pos_after=self_length;
 		}
-		if(pos_after_ref)
-			*pos_after_ref=pos_after;
 	} else { // empty delim
 		result+=this;
-		if(pos_after_ref)
-			*pos_after_ref+=self_size;
+		pos_after+=self_length;
 	}
 }
 
-void String::split(Array& result, 
-				   size_t* pos_after_ref, 
-				   const String& delim, Untaint_lang lang, 
-				   int limit) const {
+void String::split(ArrayString& result, 
+		   size_t& pos_after, 
+		   const String& delim, Language lang, 
+		   int limit) const {
+	if(is_empty())
+		return;
 	if(!delim.is_empty()) {
-		size_t pos_after=pos_after_ref?*pos_after_ref:0;
-		int pos_before;
+		size_t pos_before;
 		// while we have 'delim'...
-		for(; (pos_before=pos(delim, pos_after, lang))>=0 && limit; limit--) {
+		for(; (pos_before=pos(delim, pos_after, lang))!=STRING_NOT_FOUND && limit; limit--) {
 			result+=&mid(pos_after, pos_before);
-			pos_after=pos_before+delim.size();
+			pos_after=pos_before+delim.length();
 		}
 		// last piece
-		if(pos_after<size() && limit) {
-			result+=&mid(pos_after, size());
-			pos_after=size();
+		if(pos_after<length() && limit) {
+			result+=&mid(pos_after, length());
+			pos_after=length();
 		}
-		if(pos_after_ref)
-			*pos_after_ref=pos_after;
 	} else { // empty delim
 		result+=this;
-		if(pos_after_ref)
-			*pos_after_ref+=size();
+		pos_after+=length();
 	}
 }
 
-static void regex_options(const String *options, int *result, bool& need_pre_post_match){
-    struct Regex_option {
-		const char *keyL;
-		const char *keyU;
-		int clear, set;
-		int *result;
-		bool *flag;
-    } regex_option[]={
-		{"i", "I", 0, PCRE_CASELESS, result}, // a=A
-		{"s", "S", 0, PCRE_DOTALL, result}, // \n\n$ [default]
-		{"x", "U", 0, PCRE_EXTENDED, result}, // whitespace in regex ignored
-		{"m", "M", PCRE_DOTALL, PCRE_MULTILINE, result}, // ^aaa\n$^bbb\n$
-		{"g", "G", 0, true, result+1}, // many rows
-		{"'", 0, 0, 0, 0, &need_pre_post_match},
-		{0}
-    };
-	result[0]=PCRE_EXTRA | PCRE_DOTALL;
-	result[1]=0;
-
-    if(options) 
-		for(Regex_option *o=regex_option; o->keyL; o++) 
-			if(options->pos(o->keyL)>=0
-				|| (o->keyU && options->pos(o->keyU)>=0)) {
-				if(o->flag)
-					*o->flag=true;
-				else { // result
-					*o->result &= ~o->clear;
-					*o->result |= o->set;
-				}
-			}
-}
+Table* String::match(VRegex* vregex,
+			Row_action row_action, void *info,
+			int& matches_count) const { 
 
-/// @todo make replacement Table stacked
-bool String::match(
-				   const String *aorigin,
-				   const String& regexp, 
-				   const String *options,
-				   Table **table,
-				   Row_action row_action, void *info,
-				   bool *was_global) const { 
-
-	if(regexp.is_empty())
-		throw Exception(0,
-			aorigin,
-			"regexp is empty");
-
-	const char *pattern=regexp.cstr();
-	const char *errptr;
-	int erroffset;
-    bool need_pre_post_match=false;
-	int option_bits[2];  regex_options(options, option_bits, need_pre_post_match);
-	if(was_global)
-		*was_global=option_bits[1]!=0;
-	pcre *code=pcre_compile(pattern, option_bits[0], 
-		&errptr, &erroffset,
-		pool().get_source_charset().pcre_tables);
-
-	if(!code)
-		throw Exception(0,
-			&regexp.mid(erroffset, regexp.size()),
-			"regular expression syntax error - %s", errptr);
-	
-	int info_substrings=pcre_info(code, 0, 0);
-	if(info_substrings<0) {
-		pcre_free(code);
-		throw Exception(0,
-			aorigin,
-			"pcre_info error (%d)", 
-				info_substrings);
-	}
-
-	const char *subject=cstr();
-	int length=strlen(subject);
-	const int ovecsize=(1/*match*/+MAX_STRING_MATCH_TABLE_COLUMNS)*3;
-	int ovector[ovecsize];
+	// vregex->info(); // I have no idea what does it for?
+
+	bool need_pre_post_match=vregex->is_pre_post_match_needed();
+	bool global=vregex->is_global_search();
 
-	// create table
-	*table=NEW Table(pool(), *string_match_table_template);
+	const char* subject=cstr();
+	size_t subject_length=length();
+	const int ovector_size=(1/*match*/+MAX_MATCH_GROUPS)*3;
+	int ovector[ovector_size];
+
+	Table::Action_options table_options;
+	Table& table=*new Table(string_match_table_template, table_options);
 
-	int exec_option_bits=0;
 	int prestart=0;
 	int poststart=0;
-	int postfinish=size();
+	int postfinish=length();
 	while(true) {
-		int exec_substrings=pcre_exec(code, 0,
-			subject, length, prestart,
-			exec_option_bits, ovector, ovecsize);
-		
-		if(exec_substrings==PCRE_ERROR_NOMATCH) {
-			pcre_free(code);
-			row_action(**table, 0/*last time, no row*/, 0, 0, poststart, postfinish, info);
-			return option_bits[1]!=0; // global=true+table, not global=false
-		}
+		int exec_result=vregex->exec(subject, subject_length, ovector, ovector_size, prestart);
 
-		if(exec_substrings<0) {
-			pcre_free(code);
-			throw Exception(0,
-				aorigin,
-				"regular expression execute error (%d)", 
-					exec_substrings);
-		}
+		if(exec_result<0) // only PCRE_ERROR_NOMATCH might be here, other negative results cause an exception
+			break;
 
 		int prefinish=ovector[0];
 		poststart=ovector[1];
-		Array& row=*NEW Array(pool());
-		row+=need_pre_post_match?&mid(0, prefinish):0; // .prematch column value
-		row+=need_pre_post_match?&mid(prefinish, poststart):0; // .match
-		row+=need_pre_post_match?&mid(poststart, postfinish):0; // .postmatch
+
+		if (prestart==poststart && subject[poststart]=='\n'){
+			prestart++;
+			continue;
+		}
+
+		ArrayString* row=new ArrayString(3);
+		if(need_pre_post_match) {
+			*row+=&mid(0, prefinish); // .prematch column value
+			*row+=&mid(prefinish, poststart); // .match
+			*row+=&mid(poststart, postfinish); // .postmatch
+		} else {
+			*row+=&Empty; // .prematch column value
+			*row+=&Empty; // .match
+			*row+=&Empty; // .postmatch
+		}
 		
-		for(int i=1; i<exec_substrings; i++) {
+		for(int i=1; i<exec_result; i++) {
 			// -1:-1 case handled peacefully by mid() itself
-			row+=&mid(ovector[i*2+0], ovector[i*2+1]); // .i column value
+			*row+=(ovector[i*2+0]>=0 && ovector[i*2+1]>0)?&mid(ovector[i*2+0], ovector[i*2+1]):new String; // .i column value
 		}
 		
-		row_action(**table, &row, prestart, prefinish, poststart, postfinish, info);
+		matches_count++;
+		row_action(table, row, prestart, prefinish, poststart, postfinish, info);
 
-		if(!option_bits[1] || prestart==poststart) { // not global | going to hang
-			pcre_free(code);
-			row_action(**table, 0/*last time, no row*/, 0, 0, poststart, postfinish, info);
-			return true;
-		}
-		prestart=poststart;
+		if(!global || prestart==poststart) // last step
+			break;
 
-/*
-		if(option_bits[0] & PCRE_MULTILINE)
-			exec_option_bits|=PCRE_NOTBOL; // start of subject+startoffset not BOL
-*/
+		prestart=poststart;
 	}
+
+	row_action(table, 0/*last time, no raw*/, 0, 0, poststart, postfinish, info);
+	return vregex->is_just_count() ? 0 : &table;
 }
 
-String& String::change_case(Pool& pool, 
-							Change_case_kind kind) const {
-	const unsigned char *tables=pool.get_source_charset().pcre_tables;
-	String& result=*new(pool) String(pool);
-
-	const unsigned char *a;
-	const unsigned char *b;
-	switch(kind) {
-	case CC_UPPER:
-		a=tables+lcc_offset;
-		b=tables+fcc_offset;
-		break;
-	case CC_LOWER:
-		a=tables+lcc_offset;
-		b=0;
-		break;
-	default:
-		throw Exception(0, 
-			this, 
-			"unknown change case kind #%d", 
-				static_cast<int>(kind)); // never
-		a=b=0; // calm, compiler
-		break; // never
-	}	
+String& String::change_case(Charset& source_charset, Change_case_kind kind) const {
+	String& result=*new String();
+	if(is_empty())
+		return result;
+
+	char* new_cstr=cstrm();
+
+	if(source_charset.isUTF8()) {
+		size_t new_cstr_len=length();
+		switch(kind) {
+		case CC_UPPER:
+			change_case_UTF8((const XMLByte*)new_cstr, new_cstr_len, (XMLByte*)new_cstr, new_cstr_len, UTF8CaseToUpper);
+			break;
+		case CC_LOWER:
+			change_case_UTF8((const XMLByte*)new_cstr, new_cstr_len, (XMLByte*)new_cstr, new_cstr_len, UTF8CaseToLower);
+			break;
+		default:
+			assert(!"unknown change case kind");
+			break; // never
+		}	
+		
+	} else {
+		const unsigned char *tables=source_charset.pcre_tables;
+
+		const unsigned char *a;
+		const unsigned char *b;
+		switch(kind) {
+		case CC_UPPER:
+			a=tables+lcc_offset;
+			b=tables+fcc_offset;
+			break;
+		case CC_LOWER:
+			a=tables+lcc_offset;
+			b=0;
+			break;
+		default:
+			assert(!"unknown change case kind");
+			a=b=0; // calm, compiler
+			break; // never
+		}	
 
-	STRING_FOREACH_ROW(
-		char *new_cstr=(char *)pool.malloc(row->item.size, 12);
 		char *dest=new_cstr;
-		const char *src=row->item.ptr; 
-		for(int size=row->item.size; size--; src++) {
-			unsigned char c=a[(unsigned char)*src];
+		unsigned char index;
+		for(const char* current=new_cstr; (index=(unsigned char)*current); current++) {
+			unsigned char c=a[index];
 			if(b)
 				c=b[c];
 
 			*dest++=(char)c;
 		}
-		
-		result.APPEND(new_cstr, row->item.size, 
-			row->item.lang,
-			row->item.origin.file, row->item.origin.line);
-	);
+	}
+	result.langs=langs;
+	result.body=new_cstr;
 
 	return result;
 }
 
-/// @test if in some piece were found no dict words, append it, not it's duplicate
-String& String::replace(Pool& pool, Dictionary& dict) const {
-//	return reconstruct(pool).replace_in_reconstructed(pool, dict);
-	String& result=*new(pool) String(pool);
-
-	STRING_FOREACH_ROW(
-		const char *src=row->item.ptr; 
-		size_t src_size=row->item.size;
-		char *new_cstr=(char *)pool.malloc((size_t)ceil(src_size*dict.max_ratio()), 14);
-		char *dest=new_cstr;
-		while(src_size) {
-			// there is a row where first column starts 'src'
-			if(Table::Item *item=dict.first_that_starts(src, src_size)) {
-				// get a=>b values
-				const String& a=*static_cast<Array *>(item)->get_string(0);
-				const String& b=*static_cast<Array *>(item)->get_string(1);
-				// skip 'a' in 'src' && reduce work size
-				src+=a.size();  src_size-=a.size();
-				// write 'b' to 'dest' && skip 'b' in 'dest'
-				b.store_to(dest);  dest+=b.size();
-			} else {
-				// write a char to b && reduce work size
-				*dest++=*src++;  src_size--;
+const String& String::escape(Charset& source_charset) const {
+	if(is_empty())
+		return *this;
+
+	return Charset::escape(*this, source_charset);
+}
+
+#define STRING_APPEND(result, from_cstr, langs, langs_offset, length) \
+			result.langs.append(result.body, langs, langs_offset, length); \
+			result.body.append_strdup_know_length(from_cstr, length);
+
+const String& String::replace(const Dictionary& dict) const {
+	if(!dict.count() || is_empty())
+		return *this;
+
+	String& result=*new String();
+	const char* old_cstr=cstr();
+	const char* prematch_begin=old_cstr;
+
+	if(dict.count()==1) {
+		// optimized simple case
+
+		Dictionary::Subst subst=dict.get(0);
+		while(const char* p=strstr(prematch_begin, subst.from)) {
+			// prematch
+			if(size_t prematch_length=p-prematch_begin) {
+				STRING_APPEND(result, prematch_begin, langs, prematch_begin-old_cstr, prematch_length)
 			}
+
+			// match
+			prematch_begin=p+subst.from_length;
+
+			if(const String* b=subst.to) // are there any b?
+				result<<*b;
 		}
 
-		result.APPEND(new_cstr, dest-new_cstr, row->item.lang,
-			row->item.origin.file, row->item.origin.line);
-	);
+	} else {
+
+		const char* current=old_cstr;
+		while(*current) {
+			if(Dictionary::Subst subst=dict.first_that_begins(current)) {
+				// prematch
+				if(size_t prematch_length=current-prematch_begin) {
+					STRING_APPEND(result, prematch_begin, langs, prematch_begin-old_cstr, prematch_length)
+				}
+
+				// match
+				// skip 'a' in 'current'; move prematch_begin
+				current+=subst.from_length; prematch_begin=current;
+
+				if(const String* b=subst.to) // are there any b?
+					result<<*b;
+			} else // simply advance
+				current++; 
+		}
+
+	}
+
+	if(prematch_begin==old_cstr) // not modified
+		return *this;
+
+	// postmatch
+	if(size_t postmatch_length=old_cstr+length()-prematch_begin) {
+		STRING_APPEND(result, prematch_begin, langs, prematch_begin-old_cstr, postmatch_length)
+	}
+
+	ASSERT_STRING_INVARIANT(result);
 	return result;
 }
 
-String& String::join_chains(Pool& pool, char** acstr) const {
-	char *lcstr=cstr();
-	const char *current=lcstr;
-
-	String& result=*new(pool) String(pool);
-	STRING_FOREACH_ROW(
-IFNDEF_NO_STRING_ORIGIN(
-		const char *joined_origin_file=row->item.origin.file;
-		const size_t joined_origin_line=row->item.origin.line;
-);
-		uchar joined_lang=row->item.lang;
-		const char *joined_ptr=current;
-		// calc size
-		size_t joined_size=0;
-		STRING_PREPARED_FOREACH_ROW(*this, 
-			if(row->item.lang==joined_lang)
-				joined_size+=row->item.size;
-			else
-				break; // before non-ours
-		);
-		current+=joined_size;
-
-		// pointers are after joined piece
-		// & one step back, see STRING_FOREACH_ROW
-		--row;  ++countdown;
-		
-		result.APPEND(joined_ptr, joined_size, joined_lang,
-			joined_origin_file, joined_origin_line);
-	);
+static int serialize_body_char(char c, char** cur) {
+	*((*cur)++)=c;
+	return 0; // 0=continue
+};
+static int serialize_body_piece(const char* s, char** cur) {
+	size_t length=strlen(s);
+	memcpy(*cur, s, length);  *cur+=length;
+	return 0; // 0=continue
+};
+static int serialize_lang_piece(char alang, size_t asize, char** cur) {
+	// lang
+	**cur=alang; (*cur)++;
+	// length [WARNING: not cast, addresses must be %4=0 on sparc]
+	memcpy(*cur, &asize, sizeof(asize));  *cur+=sizeof(asize);
+
+	return 0; // 0=continue
+}
+String::Cm String::serialize(size_t prolog_length) const {
+	size_t fragments_count=langs.count();
+	size_t body_length=body.length();
+	size_t buf_length=
+		prolog_length //1
+		+sizeof(size_t) //2
+		+body_length //3
+		+1 // 4 for zero terminator used in deserialize
+		+sizeof(size_t) //5
+		+fragments_count*(sizeof(char)+sizeof(size_t)); //6
+
+	String::Cm result(new(PointerFreeGC) char[buf_length], buf_length);
+
+	// 1: prolog
+	char *cur=result.str+prolog_length;
+	// 2: chars.count [WARNING: not cast, addresses must be %4=0 on sparc]
+	memcpy(cur, &body_length, sizeof(body_length));  cur+=sizeof(body_length);
+	// 3: letters
+	body.for_each(serialize_body_char, serialize_body_piece, &cur);
+	// 4: zero terminator
+	*cur++=0;
+	// 5: langs.count [WARNING: not cast, addresses must be %4=0 on sparc]
+	memcpy(cur, &fragments_count, sizeof(fragments_count));  cur+=sizeof(fragments_count);
+	// 6: lang info
+	langs.for_each(body, serialize_lang_piece, &cur);
 
-	if(acstr)
-		*acstr=lcstr;
 	return result;
 }
+bool String::deserialize(size_t prolog_size, void *buf, size_t buf_size) {
+	size_t in_buf=buf_size;
+	if(in_buf<=prolog_size)
+		return false;
+	in_buf-=prolog_size;
 
-double String::as_double() const { 
-	double result;
-	const char *cstr;
-	char buf[MAX_NUMBER];
-	if(head.chunk.rows+1==append_here) {
-		int size=min(head.chunk.rows[0].item.size, MAX_NUMBER-1);
-		memcpy(buf, head.chunk.rows[0].item.ptr, size);
-		buf[size]=0;
-		cstr=buf;
-	} else
-		cstr=this->cstr();
-	while(*cstr && isspace(*cstr))
-		cstr++;
-	if(!*cstr)
-		return 0;
+	// 1: prolog
+	const char* cur=(const char* )buf+prolog_size;
 
-	char *error_pos;
-	// 0xABC
-	if(cstr[0]=='0')
-		if(cstr[1]=='x' || cstr[1]=='X')
-			result=(double)(unsigned long)strtol(cstr, &error_pos, 0);
-		else
-			result=(double)strtod(cstr+1/*skip leading 0*/, &error_pos);
-	else
-		result=(double)strtod(cstr, &error_pos);
+	// 2: chars.count
+	size_t body_length;
+	if(in_buf<sizeof(body_length)) // body.length don't fit?
+		return false;
+	// [WARNING: not cast, addresses must be %4=0 on sparc]
+	memcpy(&body_length, cur, sizeof(body_length));  cur+=sizeof(body_length);
+	in_buf-=sizeof(body_length);
 
-	while(char c=*error_pos++)
-		if(!isspace(c))
-			throw Exception("number.format",
-				this,
-				"invalid number (double)");
+	if(in_buf<body_length+1) // letters+terminator don't fit?
+		return false;
+	// 4: zero terminator
+	if(cur[body_length] != 0) // in place?
+		return false;
+	// 3: letters
+	body=String::Body(*cur?cur:0);
+#ifdef STRING_LENGTH_CACHING
+	body.set_length(body_length);
+#endif
+	cur+=body_length+1;
+	in_buf-=body_length+1;
 
-	return result;
-}
-int String::as_int() const { 
-	int result;
-	const char *cstr;
-	char buf[MAX_NUMBER];
-	if(head.chunk.rows+1==append_here) {
-		size_t size=min(head.chunk.rows[0].item.size, MAX_NUMBER-1);
-		memcpy(buf, head.chunk.rows[0].item.ptr, size);
-		buf[size]=0;
-		cstr=buf;
-	} else
-		cstr=this->cstr();
-	while(*cstr && isspace(*cstr))
-		cstr++;
-	if(!*cstr)
-		return 0;
+	// 5: langs.count
+	size_t fragments_count;
+	if(in_buf<sizeof(fragments_count)) // langs.count don't fit?
+		return false;
+	// [WARNING: not cast, addresses must be %4=0 on sparc]
+	memcpy(&fragments_count, cur, sizeof(fragments_count));  cur+=sizeof(fragments_count);
+	in_buf-=sizeof(fragments_count);
+	
+	if(fragments_count) {
+		// 6: lang info
+		size_t total_length=0;
+		for(size_t f=0; f<fragments_count; f++) {
+			char lang;
+			size_t fragment_length;
+			size_t piece_length=sizeof(lang)+sizeof(fragment_length);
+			if(in_buf<piece_length) // lang+length
+				return false;
+
+			// lang
+			lang=*cur++;
+			// length [WARNING: not cast, addresses must be %4=0 on sparc]
+			memcpy(&fragment_length, cur, sizeof(fragment_length));  cur+=sizeof(fragment_length);
+
+			size_t combined_length=total_length+fragment_length;
+			if(combined_length>body_length)
+				return false; // file curruption
+			// uchar needed to prevent propagating 0x80 bit to upper bytes
+			langs.append(total_length, (String::Language)(uchar)lang, fragment_length);
+			total_length=combined_length;
+			in_buf-=piece_length;
+		}
 
-	char *error_pos;
-	// 0xABC
-	if(cstr[0]=='0')
-		if(cstr[1]=='x' || cstr[1]=='X')
-			result=(int)(unsigned long)strtol(cstr, &error_pos, 0);
-		else
-			result=(int)strtol(cstr+1/*skip leading 0*/, &error_pos, 0);
-	else
-		result=(int)strtol(cstr, &error_pos, 0);
+		if(total_length!=body_length) // length(all language fragments) vs length(letters)
+			return false;
+	}
+	if(in_buf!=0) // some strange extra bytes
+		return false;
 
-	while(char c=*error_pos++)
-		if(!isspace(c))
-			throw Exception("number.format",
-				this,
-				"invalid number (int)");
+	ASSERT_STRING_INVARIANT(*this);
+	return true;
+}
 
-	return result;
+const char* String::Body::v() const {
+	return CORD_to_const_char_star(body, length());
+}
+void String::Body::dump() const {
+	CORD_dump(body);
 }
 
-inline void ushort2uchars(ushort word, uchar& byte1, uchar& byte2) {
-	byte1=word&0xFF;
-	byte2=word>>8;
-}
-inline ushort uchars2ushort(uchar byte1, uchar byte2) {
-	return (byte2<<8) | byte1;
-}
-/* @todo maybe network order worth spending some effort?
-	don't bothering myself with network byte order,
-	am not planning to be able to move resulting file across platforms
-	for now
-*/
-void String::serialize(size_t prolog_size, void *& buf, size_t& buf_size) const {
-	buf_size=
-		prolog_size
-		+used_rows()*(sizeof(uchar)+sizeof(ushort))
-		+size();
-	buf=malloc(buf_size,15);
-	char *cur=(char *)buf+prolog_size;
-
-	STRING_FOREACH_ROW(
-		// lang
-		memcpy(cur, &row->item.lang, sizeof(uchar));
-		cur+=sizeof(uchar);
-		// size
-		uchar byte1; uchar byte2;
-		ushort2uchars(row->item.size, byte1, byte2);
-		memcpy(cur, &byte1, sizeof(uchar)); cur+=sizeof(uchar);
-		memcpy(cur, &byte2, sizeof(uchar)); cur+=sizeof(uchar);
-		// bytes
-		memcpy(cur, row->item.ptr, row->item.size);
-		cur+=row->item.size;
+const char* String::Languages::v() const {
+	if(opt.is_not_just_lang)
+		return CORD_to_const_char_star(langs, 0);
+	else
+		return (const char*)&langs;
+}
+void String::Languages::dump() const {
+	if(opt.is_not_just_lang)
+		CORD_dump(langs);
+	else
+		puts((const char*)&langs);
+}
+const char* String::v() const {
+	const uint LIMIT_VIEW=20;
+	char* buf=(char*)malloc(MAX_STRING);
+	const char*body_view=body.v();
+	const char*langs_view=langs.v();
+	snprintf(buf, MAX_STRING, 
+		"%d:%.*s%s}   "
+		"{%d:%s",
+		langs.count(), LIMIT_VIEW, langs_view, strlen(langs_view)>LIMIT_VIEW?"...":"",
+		strlen(body_view), body_view
 	);
+
+	return buf;
 }
-bool String::deserialize(size_t prolog_size, void *buf, size_t buf_size, const char *file) {
-	if(buf_size<=prolog_size)
-		return false;
 
-	char *cur=(char *)buf+prolog_size;
-	buf_size-=prolog_size;
+void String::dump() const {
+	body.dump();
+	langs.dump();
+}
 
-	while(buf_size) {
-		if(sizeof(uchar)+sizeof(ushort)>buf_size) // lang+size
-			return false;
+const String& String::trim(String::Trim_kind kind, const char* chars, Charset* source_charset) const {
+	if(is_empty())
+		return *this;
 
-		uchar lang=*(uchar *)(cur);		
-		ushort size=uchars2ushort(
-			*(uchar*)(cur+sizeof(uchar)*1),
-			*(uchar*)(cur+sizeof(uchar)*2)
-		);
+	size_t substr_begin, substr_length;
+	Body new_body=body.trim(kind, chars, &substr_begin, &substr_length, source_charset);
+	if(new_body==body) // we received unchanged pointer, do likewise
+		return *this;
+	// new_body differs from body, adjust langs along
 
-		size_t piece_size=sizeof(uchar)+sizeof(ushort)+size;
-		if(piece_size>buf_size) // buffer overrun, can be on incomplete cache files
-			return false;
+	String& result=*new String;
+	if(!new_body) // body.trim produced empty result
+		return result;
+	// body.trim produced nonempty result
 
-		const char *ptr=(const char*)(cur+sizeof(uchar)*3); 
-		APPEND(ptr, size, lang, file, 0);
+	// first: their langs
+	result.langs.append(result.body, langs, substr_begin, substr_length);
+	// next: letters themselves
+	result.body=new_body;
 
-		cur+=piece_size;
-		buf_size-=piece_size;
-	}
-	return true;
+	ASSERT_STRING_INVARIANT(result);
+	return result;
 }