/*  
**********************************************************************
*   Copyright (C) 2000, International Business Machines
*   Corporation and others.  All Rights Reserved.
**********************************************************************
*   file name:  ucnv_lmb.cpp
*   encoding:   US-ASCII
*   tab size:   4 (not used)
*   indentation:4
*
*   created on: 2000feb09
*   created by: Brendan Murray
*   extensively hacked up by: Jim Snyder-Grant
*
* Modification History:
* 
*   Date        Name             Description
* 
*   06/20/2000  helena           OS/400 port changes; mostly typecast.
*   06/27/2000  Jim Snyder-Grant Deal with partial characters and small buffers.
*                                Add comments to document LMBCS format and implementation
*                                restructured order & breakdown of functions
*   06/28/2000  helena           Major rewrite for the callback API changes.
*/

#include "unicode/utypes.h"
#include "cmemory.h"
#include "unicode/ucnv_err.h"
#include "ucnv_bld.h"
#include "unicode/ucnv.h"
#include "ucnv_cnv.h"

/*
  LMBCS

  (Lotus Multi-Byte Character Set)

  LMBCS was invented in the late 1980's and is primarily used in Lotus Notes 
  databases and in Lotus 1-2-3 files. Programmers who work with the APIs 
  into these products will sometimes need to deal with strings in this format.

  The code in this file provides an implementation for an ICU converter of 
  LMBCS to and from Unicode. 

  Since the LMBCS character set is only sparsely documented in existing 
  printed or online material, we have added  extensive annotation to this 
  file to serve as a guide to understanding LMBCS. 

  LMBCS was originally designed with these four sometimes-competing design goals:

  -Provide encodings for the characters in 12 existing national standards
   (plus a few other characters)
  -Minimal memory footprint
  -Maximal speed of conversion into the existing national character sets
  -No need to track a changing state as you interpret a string.


  All of the national character sets LMBCS was trying to encode are 'ANSI'
  based, in that the bytes from 0x20 - 0x7F are almost exactly the 
  same common Latin unaccented characters and symbols in all character sets. 

  So, in order to help meet the speed & memory design goals, the common ANSI 
  bytes from 0x20-0x7F are represented by the same single-byte values in LMBCS. 

  The general LMBCS code unit is from 1-3 bytes. We can describe the 3 bytes as
  follows:

  [G] D1 [D2]

  That is, a sometimes-optional 'group' byte, followed by 1 and sometimes 2
  data bytes. The maximum size of a LMBCS chjaracter is 3 bytes:
*/
#define ULMBCS_CHARSIZE_MAX      3
/*
  The single-byte values from 0x20 to 0x7F are examples of single D1 bytes.
  We often have to figure out if byte values are below or above this, so we 
  use the ANSI nomenclature 'C0' and 'C1' to refer to the range of control 
  characters just above & below the common lower-ANSI  range */
#define ULMBCS_C0END           0x1F   
#define ULMBCS_C1START         0x80   
/*
  Since LMBCS is always dealing in byte units. we create a local type here for 
  dealing with these units of LMBCS code units:

*/  
typedef uint8_t ulmbcs_byte_t;

/* 
   Most of the values less than 0x20 are reserved in LMBCS to announce 
   which national  character standard is being used for the 'D' bytes. 
   In the comments we show the common name and the IBM character-set ID
   for these character-set announcers:
*/

#define ULMBCS_GRP_L1         0x01   /* Latin-1    :ibm-850  */
#define ULMBCS_GRP_GR         0x02   /* Greek      :ibm-851  */
#define ULMBCS_GRP_HE         0x03   /* Hebrew     :ibm-1255 */
#define ULMBCS_GRP_AR         0x04   /* Arabic     :ibm-1256 */
#define ULMBCS_GRP_RU         0x05   /* Cyrillic   :ibm-1251 */
#define ULMBCS_GRP_L2         0x06   /* Latin-2    :ibm-852  */
#define ULMBCS_GRP_TR         0x08   /* Turkish    :ibm-1254 */
#define ULMBCS_GRP_TH         0x0B   /* Thai       :ibm-874  */
#define ULMBCS_GRP_JA         0x10   /* Japanese   :ibm-943  */
#define ULMBCS_GRP_KO         0x11   /* Korean     :ibm-1261 */
#define ULMBCS_GRP_CN         0x12   /* Chinese SC :ibm-950  */
#define ULMBCS_GRP_TW         0x13   /* Chinese TC :ibm-1386 */

/*
   So, the beginning of understanding LMBCS is that IF the first byte of a LMBCS 
   character is one of those 12 values, you can interpret the remaining bytes of 
   that character as coming from one of those character sets. Since the lower 
   ANSI bytes already are represented in single bytes, using one of the character 
   set announcers is used to announce a character that starts with a byte of 
   0x80 or greater.

   The character sets are  arranged so that the single byte sets all appear 
   before the multi-byte character sets. When we need to tell whether a 
   group byte is for a single byte char set or not we use this define: */

#define ULMBCS_DOUBLEOPTGROUP_START  0x10   

/* 
However, to fully understand LMBCS, you must also understand a series of 
exceptions & optimizations made in service of the design goals. 

First, those of you who are character set mavens may have noticed that
the 'double-byte' character sets are actually multi-byte character sets 
that can have 1 or two bytes, even in the upper-ascii range. To force
each group byte to introduce a fixed-width encoding (to make it faster to 
count characters), we use a convention of doubling up on the group byte 
to introduce any single-byte character > 0x80 in an otherwise double-byte
character set. So, for example, the LMBCS sequence x10 x10 xAE is the 
same as '0xAE' in the Japanese code page 943.

Next, you will notice that the list of group bytes has some gaps. 
These are used in various ways.

We reserve a few special single byte values for common control 
characters. These are in the same place as their ANSI eqivalents for speed.
*/
                     
#define ULMBCS_HT    0x09   /* Fixed control char - Horizontal Tab */
#define ULMBCS_LF    0x0A   /* Fixed control char - Line Feed */
#define ULMBCS_CR    0x0D   /* Fixed control char - Carriage Return */

/* Then, 1-2-3 reserved a special single-byte character to put at the 
beginning of internal 'system' range names: */

#define ULMBCS_123SYSTEMRANGE  0x19   

/* Then we needed a place to put all the other ansi control characters 
that must be moved to different values because LMBCS reserves those 
values for other purposes. To represent the control characters, we start 
with a first byte of 0xF & add the control chaarcter value as the 
second byte */
#define ULMBCS_GRP_CTRL       0x0F   

/* For the C0 controls (less than 0x20), we add 0x20 to preserve the 
useful doctrine that any byte less than 0x20 in a LMBCS char must be 
the first byte of a character:*/
#define ULMBCS_CTRLOFFSET      0x20   

/* 
Where to put the characters that aren't part of any of the 12 national 
character sets? The first thing that was done, in the earlier years of 
LMBCS, was to use up the spaces of the form

  [G] D1, 
  
 where  'G' was one of the single-byte character groups, and
 D1 was less than 0x80. These sequences are gathered together 
 into a Lotus-invented doublebyte character set to represent a 
 lot of stray values. Internally, in this implementation, we track this 
 as group '0', as a place to tuck this exceptions list.*/

#define ULMBCS_GRP_EXCEPT     0x00    
/*
 Finally, as the durability and usefulness of UNICODE became clear, 
 LOTUS added a new group 0x14 to hold Unicode values not otherwise 
 represented in LMBCS: */
#define ULMBCS_GRP_UNICODE    0x14   
/* The two bytes appearing after a 0x14 are intrepreted as UFT-16 BE
(Big-Endian) characters. The exception comes when the UTF16 
representation would have a zero as the second byte. In that case,
'F6' is used in its place, and the bytes are swapped. (This prevents 
LMBCS from encoding any Unicode values of the form U+F6xx, but that's OK:
0xF6xx is in the middle of the Private Use Area.)*/
#define ULMBCS_UNICOMPATZERO   0xF6   

/* It is also useful in our code to have a constant for the size of 
a LMBCS char that holds a literal Unicode value */
#define ULMBCS_UNICODE_SIZE      3    

/* 
To squish the LMBCS representations down even further, and to make 
translations even faster,sometimes the optimization group byte can be dropped 
from a LMBCS character. This is decided on a process-by-process basis. The 
group byte that is dropped is called the 'optimization group'.

For Notes, the optimzation group is always 0x1.*/
#define ULMBCS_DEFAULTOPTGROUP 0x1    
/* For 1-2-3 files, the optimzation group is stored in the header of the 1-2-3 
file. 

 In any case, when using ICU, you either pass in the 
optimization group as part of the name of the converter (LMBCS-1, LMBCS-2, 
etc.). Using plain 'LMBCS' as the name of the converter will give you 
LMBCS-1.


*** Implementation strategy ***


Because of the extensive use of other character sets, the LMBCS converter
keeps a mapping between optimization groups and IBM character sets, so that
ICU converters can be created and used as needed. */

static const char * OptGroupByteToCPName[ULMBCS_CTRLOFFSET] = {
   /* 0x0000 */ "lmb-excp", /* internal home for the LOTUS exceptions list */
   /* 0x0001 */ "ibm-850",
   /* 0x0002 */ "ibm-851",
   /* 0x0003 */ "ibm-1255",
   /* 0x0004 */ "ibm-1256",
   /* 0x0005 */ "ibm-1251",
   /* 0x0006 */ "ibm-852",
   /* 0x0007 */ NULL,      /* Unused */
   /* 0x0008 */ "ibm-1254",
   /* 0x0009 */ NULL,      /* Control char HT */
   /* 0x000A */ NULL,      /* Control char LF */
   /* 0x000B */ "ibm-874",
   /* 0x000C */ NULL,      /* Unused */
   /* 0x000D */ NULL,      /* Control char CR */
   /* 0x000E */ NULL,      /* Unused */
   /* 0x000F */ NULL,      /* Control chars: 0x0F20 + C0/C1 character: algorithmic */
   /* 0x0010 */ "ibm-943",
   /* 0x0011 */ "ibm-1363",
   /* 0x0012 */ "ibm-950",
   /* 0x0013 */ "ibm-1386"

   /* The rest are null, including the 0x0014 Unicode compatibility region
   and 0x0019, the 1-2-3 system range control char */      
};

/* As you can see, even though any byte below 0x20 could be an optimization 
byte, only those at 0x13 or below can map to an actual converter. To limit
some loops and searches, we define a value for that last group converter:*/

#define ULMBCS_GRP_LAST       0x13   /* last LMBCS group that has a converter */


/* That's approximately all the data that's needed for translating 
  LMBCS to Unicode. 


However, to translate Unicode to LMBCS, we need some more support.

That's because there are often more than one possible mappings from a Unicode
code point back into LMBCS. The first thing we do is look up into a table
to figure out if there are more than one possible mappings. This table,
arranged by Unicode values (including ranges) either lists which group 
to use, or says that it could go into one or more of the SBCS sets, or
into one or more of the DBCS sets.  (If the character exists in both DBCS & 
SBCS, the table will place it in the SBCS sets, to make the LMBCS code point 
length as small as possible. Here's the two special markers we use to indicate
ambiguous mappings: */

#define ULMBCS_AMBIGUOUS_SBCS   0x80   /* could fit in more than one 
                                          LMBCS sbcs native encoding 
                                          (example: most accented latin) */
#define ULMBCS_AMBIGUOUS_MBCS   0x81   /* could fit in more than one 
                                          LMBCS mbcs native encoding 
                                          (example: Unihan) */

/* And here's a simple way to see if a group falls in an appropriate range */
#define ULMBCS_AMBIGUOUS_MATCH(agroup, xgroup) \
                  ((((agroup) == ULMBCS_AMBIGUOUS_SBCS) && \
                  (xgroup) < ULMBCS_DOUBLEOPTGROUP_START) || \
                  (((agroup) == ULMBCS_AMBIGUOUS_MBCS) && \
                  (xgroup) >= ULMBCS_DOUBLEOPTGROUP_START))


/* The table & some code to use it: */


struct _UniLMBCSGrpMap  
{
   UChar uniStartRange;
   UChar uniEndRange;
   ulmbcs_byte_t  GrpType;
} UniLMBCSGrpMap[]
=
{

   {0x0001, 0x001F,  ULMBCS_GRP_CTRL},
   {0x0080, 0x009F,  ULMBCS_GRP_CTRL},
   {0x00A0, 0x01CD,  ULMBCS_AMBIGUOUS_SBCS},
   {0x01CE, 0x01CE,  ULMBCS_GRP_TW }, 
   {0x01CF, 0x02B9,  ULMBCS_AMBIGUOUS_SBCS},
   {0x02BA, 0x02BA,  ULMBCS_GRP_CN},
   {0x02BC, 0x02C8,  ULMBCS_AMBIGUOUS_SBCS},
   {0x02C9, 0x02D0,  ULMBCS_AMBIGUOUS_MBCS},
   {0x02D8, 0x02DD,  ULMBCS_AMBIGUOUS_SBCS},
   {0x0384, 0x03CE,  ULMBCS_AMBIGUOUS_SBCS},
   {0x0400, 0x044E,  ULMBCS_GRP_RU},
   {0x044F, 0x044F,  ULMBCS_AMBIGUOUS_MBCS},
   {0x0450, 0x0491,  ULMBCS_GRP_RU},
   {0x05B0, 0x05F2,  ULMBCS_GRP_HE},
   {0x060C, 0x06AF,  ULMBCS_GRP_AR}, 
   {0x0E01, 0x0E5B,  ULMBCS_GRP_TH},
   {0x200C, 0x200F,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2010, 0x2010,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2013, 0x2015,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2016, 0x2016,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2017, 0x2024,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2025, 0x2025,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2026, 0x2026,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2027, 0x2027,  ULMBCS_GRP_CN},
   {0x2030, 0x2033,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2035, 0x2035,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2039, 0x203A,  ULMBCS_AMBIGUOUS_SBCS},
   {0x203B, 0x203B,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2074, 0x2074,  ULMBCS_GRP_KO},
   {0x207F, 0x207F,  ULMBCS_GRP_EXCEPT},
   {0x2081, 0x2084,  ULMBCS_GRP_KO},
   {0x20A4, 0x20AC,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2103, 0x2109,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2111, 0x2126,  ULMBCS_AMBIGUOUS_SBCS},
   {0x212B, 0x212B,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2135, 0x2135,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2153, 0x2154,  ULMBCS_GRP_KO},
   {0x215B, 0x215E,  ULMBCS_GRP_EXCEPT},
   {0x2160, 0x2179,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2190, 0x2195,  ULMBCS_GRP_EXCEPT},
   {0x2196, 0x2199,  ULMBCS_AMBIGUOUS_MBCS},
   {0x21A8, 0x21A8,  ULMBCS_GRP_EXCEPT},
   {0x21B8, 0x21B9,  ULMBCS_GRP_CN},
   {0x21D0, 0x21D5,  ULMBCS_GRP_EXCEPT},
   {0x21E7, 0x21E7,  ULMBCS_GRP_CN},
   {0x2200, 0x220B,  ULMBCS_GRP_EXCEPT},
   {0x220F, 0x2215,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2219, 0x2220,  ULMBCS_GRP_EXCEPT},
   {0x2223, 0x2228,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2229, 0x222B,  ULMBCS_GRP_EXCEPT},
   {0x222C, 0x223D,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2245, 0x2248,  ULMBCS_GRP_EXCEPT},
   {0x224C, 0x224C,  ULMBCS_GRP_TW},
   {0x2252, 0x2252,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2260, 0x2265,  ULMBCS_GRP_EXCEPT},
   {0x2266, 0x226F,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2282, 0x2297,  ULMBCS_GRP_EXCEPT},
   {0x2299, 0x22BF,  ULMBCS_AMBIGUOUS_MBCS},
   {0x22C0, 0x22C0,  ULMBCS_GRP_EXCEPT},
   {0x2310, 0x2310,  ULMBCS_GRP_EXCEPT},
   {0x2312, 0x2312,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2318, 0x2321,  ULMBCS_GRP_EXCEPT},
   {0x2318, 0x2321,  ULMBCS_GRP_CN},
   {0x2460, 0x24E9,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2500, 0x2500,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2501, 0x2501,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2502, 0x2502,  ULMBCS_AMBIGUOUS_SBCS},
   {0x2503, 0x2503,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2504, 0x2505,  ULMBCS_GRP_TW},
   {0x2506, 0x2665,  ULMBCS_AMBIGUOUS_MBCS},
   {0x2666, 0x2666,  ULMBCS_GRP_EXCEPT},
   {0x2667, 0xFFFE,  ULMBCS_AMBIGUOUS_MBCS},
   {0xFFFF, 0xFFFF,  ULMBCS_GRP_UNICODE}
};

static ulmbcs_byte_t 
FindLMBCSUniRange(UChar uniChar)
{
   struct _UniLMBCSGrpMap * pTable = UniLMBCSGrpMap;

   while (uniChar > pTable->uniEndRange) 
   {
      pTable++;
   }

   if (uniChar >= pTable->uniStartRange) 
   {
      return pTable->GrpType;
   }
   return ULMBCS_GRP_UNICODE;
}

/* 
We also ask the creator of a converter to send in a preferred locale 
that we can use in resolving ambiguous mappings. They send the locale
in as a string, and we map it, if possible, to one of the 
LMBCS groups. We use this table, and the associated code, to 
do the lookup: */

/**************************************************
  This table maps locale ID's to LMBCS opt groups.
  The default return is group 0x01. Note that for
  performance reasons, the table is sorted in
  increasing alphabetic order, with the notable
  exception of zh_TW. This is to force the check
  for Traditonal Chinese before dropping back to
  Simplified.

  Note too that the Latin-1 groups have been
  commented out because it's the default, and
  this shortens the table, allowing a serial
  search to go quickly.
 *************************************************/

struct _LocaleLMBCSGrpMap
{
   const char    *LocaleID;
   ulmbcs_byte_t OptGroup;
}  LocaleLMBCSGrpMap[] =
{
   {"ar", ULMBCS_GRP_AR},
   {"be", ULMBCS_GRP_RU},
   {"bg", ULMBCS_GRP_L2},
   /* {"ca", ULMBCS_GRP_L1}, */
   {"cs", ULMBCS_GRP_L2},
   /* {"da", ULMBCS_GRP_L1}, */
   /* {"de", ULMBCS_GRP_L1}, */
   {"el", ULMBCS_GRP_GR},
   /* {"en", ULMBCS_GRP_L1}, */
   /* {"es", ULMBCS_GRP_L1}, */
   /* {"et", ULMBCS_GRP_L1}, */
   /* {"fi", ULMBCS_GRP_L1}, */
   /* {"fr", ULMBCS_GRP_L1}, */
   {"he", ULMBCS_GRP_HE},
   {"hu", ULMBCS_GRP_L2},
   /* {"is", ULMBCS_GRP_L1}, */
   /* {"it", ULMBCS_GRP_L1}, */
   {"iw", ULMBCS_GRP_HE},
   {"ja", ULMBCS_GRP_JA},
   {"ko", ULMBCS_GRP_KO},
   /* {"lt", ULMBCS_GRP_L1}, */
   /* {"lv", ULMBCS_GRP_L1}, */
   {"mk", ULMBCS_GRP_RU},
   /* {"nl", ULMBCS_GRP_L1}, */
   /* {"no", ULMBCS_GRP_L1}, */
   {"pl", ULMBCS_GRP_L2},
   /* {"pt", ULMBCS_GRP_L1}, */
   {"ro", ULMBCS_GRP_L2},
   {"ru", ULMBCS_GRP_RU},
   {"sh", ULMBCS_GRP_L2},
   {"sk", ULMBCS_GRP_L2},
   {"sl", ULMBCS_GRP_L2},
   {"sq", ULMBCS_GRP_L2},
   {"sr", ULMBCS_GRP_RU},
   /* {"sv", ULMBCS_GRP_L1}, */
   {"th", ULMBCS_GRP_TH},
   {"tr", ULMBCS_GRP_TR},
   {"uk", ULMBCS_GRP_RU},
   /* {"vi", ULMBCS_GRP_L1}, */
   {"zh_TW", ULMBCS_GRP_TW},
   {"zh", ULMBCS_GRP_CN},
   {NULL, ULMBCS_GRP_L1}
};


static ulmbcs_byte_t 
FindLMBCSLocale(const char *LocaleID)
{
   struct _LocaleLMBCSGrpMap *pTable = LocaleLMBCSGrpMap;

   if ((!LocaleID) || (!*LocaleID)) 
   {
      return 0;
   }

   while (pTable->LocaleID)
   {
      if (*pTable->LocaleID == *LocaleID) /* Check only first char for speed */
      {
         /* First char matches - check whole name, for entry-length */
         if (strncmp(pTable->LocaleID, LocaleID, strlen(pTable->LocaleID)) == 0)
            return pTable->OptGroup;
      }
      else
      if (*pTable->LocaleID > *LocaleID) /* Sorted alphabetically - exit */
         break;
      pTable++;
   }
   return ULMBCS_GRP_L1;
}


/* 
  Before we get to the main body of code, here's how we hook up to the rest 
  of ICU. ICU converters are required to define a structure that includes 
  some function pointers, and some common data, in the style of a C++
  vtable. There is also room in there for converter-specific data. LMBCS
  uses that converter-specific data to keep track of the 12 subconverters
  we use, the optimization group, and the group (if any) that matches the 
  locale. We have one structure instantiated for each of the 12 possible
  optimization groups. To avoid typos & to avoid boring the reader, we 
  put the declarations of these structures and functions into macros. To see 
  the definitions of these structures, see unicode\ucnv_bld.h
*/


#define DECLARE_LMBCS_DATA(n) \
 static const UConverterImpl _LMBCSImpl##n={\
    UCNV_LMBCS_##n,\
    NULL,NULL,\
    _LMBCSOpen##n,\
    _LMBCSClose,\
    NULL,\
    _LMBCSToUnicodeWithOffsets,\
    _LMBCSToUnicodeWithOffsets,\
    _LMBCSFromUnicode,\
    _LMBCSFromUnicode,\
    _LMBCSGetNextUChar,\
    NULL\
};\
const UConverterStaticData _LMBCSStaticData##n={\
  sizeof(UConverterStaticData),\
 "LMBCS-"  #n,\
    0, UCNV_IBM, UCNV_LMBCS_##n, 1, 1,\
    { 0x3f, 0, 0, 0 },1,FALSE,FALSE,0,{0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0} \
};\
const UConverterSharedData _LMBCSData##n={\
    sizeof(UConverterSharedData), ~((uint32_t) 0),\
    NULL, NULL, &_LMBCSStaticData##n, FALSE, &_LMBCSImpl##n, \
    0 \
};

 /* The only function we needed to duplicate 12 times was the 'open'
function, which will do basically the same thing except set a  different
optimization group. So, we put the common stuff into a worker function, 
and set up another macro to stamp out the 12 open functions:*/
#define DEFINE_LMBCS_OPEN(n) \
static void \
   _LMBCSOpen##n(UConverter*  _this,const char* name,const char* locale,uint32_t options,UErrorCode*  err) \
{ _LMBCSOpenWorker(_this, name,locale,options, err, n);} 


/* Here's the prototypes for the functions we will put into the ICU structures:
*/

void 
_LMBCSToUnicodeWithOffsets(UConverterToUnicodeArgs *args,
                           UErrorCode*    err);         /* Std ICU err code */


void 
_LMBCSFromUnicode(UConverterFromUnicodeArgs *args,
                  UErrorCode*     err);

UChar32 
_LMBCSGetNextUChar(UConverterToUnicodeArgs *args,
                   UErrorCode*   err);


/* Here's the open worker & the common close function */
static void 
_LMBCSOpenWorker(UConverter*  _this, 
                       const char*  name, 
                       const char*  locale,
                       uint32_t options,
                       UErrorCode*  err,
                       ulmbcs_byte_t OptGroup
                       )
{
   UConverterDataLMBCS * extraInfo = (UConverterDataLMBCS*)uprv_malloc (sizeof (UConverterDataLMBCS));
   if(extraInfo != NULL)
    {
       ulmbcs_byte_t i;
       ulmbcs_byte_t imax;
       imax = sizeof(extraInfo->OptGrpConverter)/sizeof(extraInfo->OptGrpConverter[0]);

       for (i=0; i < imax; i++)         
       {
            extraInfo->OptGrpConverter[i] =
               (OptGroupByteToCPName[i] != NULL) ? 
               ucnv_open(OptGroupByteToCPName[i], err) : NULL;
       }
       extraInfo->OptGroup = OptGroup;
       extraInfo->localeConverterIndex = FindLMBCSLocale(locale);
   } 
   else
   {
       *err = U_MEMORY_ALLOCATION_ERROR;
   }
   _this->extraInfo = extraInfo;
}

static void 
_LMBCSClose(UConverter *   _this) 
{
    if (_this->extraInfo != NULL)
    {
        ulmbcs_byte_t Ix;
        UConverterDataLMBCS * extraInfo = (UConverterDataLMBCS *) _this->extraInfo;

        for (Ix=0; Ix < ULMBCS_GRP_UNICODE; Ix++)
        {
           if (extraInfo->OptGrpConverter[Ix] != NULL)
              ucnv_close (extraInfo->OptGrpConverter[Ix]);
        }
        uprv_free (_this->extraInfo);
    }
}

/* And now, the macroized declarations of data & functions: */
DEFINE_LMBCS_OPEN(1)
DEFINE_LMBCS_OPEN(2)
DEFINE_LMBCS_OPEN(3)
DEFINE_LMBCS_OPEN(4)
DEFINE_LMBCS_OPEN(5)
DEFINE_LMBCS_OPEN(6)
DEFINE_LMBCS_OPEN(8)
DEFINE_LMBCS_OPEN(11)
DEFINE_LMBCS_OPEN(16)
DEFINE_LMBCS_OPEN(17)
DEFINE_LMBCS_OPEN(18)
DEFINE_LMBCS_OPEN(19)


DECLARE_LMBCS_DATA(1)
DECLARE_LMBCS_DATA(2)
DECLARE_LMBCS_DATA(3)
DECLARE_LMBCS_DATA(4)
DECLARE_LMBCS_DATA(5)
DECLARE_LMBCS_DATA(6)
DECLARE_LMBCS_DATA(8)
DECLARE_LMBCS_DATA(11)
DECLARE_LMBCS_DATA(16)
DECLARE_LMBCS_DATA(17)
DECLARE_LMBCS_DATA(18)
DECLARE_LMBCS_DATA(19)

/* 
Here's an all-crash stop for debugging, since ICU does not have asserts.
Turn this on by defining LMBCS_DEBUG, or by changing it to 
#if 1 
*/
#if LMBCS_DEBUG
#define MyAssert(b) {if (!(b)) {*(char *)0 = 1;}}
#else
#define MyAssert(b) 
#endif

/* 
   Here's the basic helper function that we use when converting from
   Unicode to LMBCS, and we suspect that a Unicode character will fit into 
   one of the 12 groups. The return value is the number of bytes written 
   starting at pStartLMBCS (if any).
*/

static size_t
LMBCSConversionWorker (
   UConverterDataLMBCS * extraInfo,    /* subconverters, opt & locale groups */
   ulmbcs_byte_t group,                /* The group to try */
   ulmbcs_byte_t  * pStartLMBCS,              /* where to put the results */
   UChar * pUniChar,                   /* The input unicode character */
   ulmbcs_byte_t * lastConverterIndex, /* output: track last successful group used */
   UBool * groups_tried                /* output: track any unsuccessful groups */
)   
{
   ulmbcs_byte_t  * pLMBCS = pStartLMBCS;
   UConverter * xcnv = extraInfo->OptGrpConverter[group];

   int bytesConverted;
   uint32_t value;
   ulmbcs_byte_t firstByte;

   MyAssert(xcnv);
   MyAssert(group<ULMBCS_GRP_UNICODE);

   bytesConverted = _MBCSFromUChar32(xcnv->sharedData, *pUniChar, &value, FALSE);

   /* get the first result byte */
   switch(bytesConverted)
   {
   case 4:
      firstByte = (ulmbcs_byte_t)(value >> 24);
      break;
   case 3:
      firstByte = (ulmbcs_byte_t)(value >> 16);
      break;
   case 2:
      firstByte = (ulmbcs_byte_t)(value >> 8);
      break;
   case 1:
      firstByte = (ulmbcs_byte_t)value;
      break;
   default:
      /* most common failure mode is an unassigned character */
      groups_tried[group] = TRUE;
      return 0;
   }

   *lastConverterIndex = group;

   /* All initial byte values in lower ascii range should have been caught by now,
      except with the exception group.
    */
   MyAssert((firstByte <= ULMBCS_C0END) || (firstByte >= ULMBCS_C1START) || (group == ULMBCS_GRP_EXCEPT));
   
   /* use converted data: first write 0, 1 or two group bytes */
   if (group != ULMBCS_GRP_EXCEPT && extraInfo->OptGroup != group)
   {
      *pLMBCS++ = group;
      if (bytesConverted == 1 && group >= ULMBCS_DOUBLEOPTGROUP_START)
      {
         *pLMBCS++ = group;
      }
   }
   /* then move over the converted data */
   switch(bytesConverted)
   {
   case 4:
      *pLMBCS++ = (ulmbcs_byte_t)(value >> 24);
   case 3:
      *pLMBCS++ = (ulmbcs_byte_t)(value >> 16);
   case 2:
      *pLMBCS++ = (ulmbcs_byte_t)(value >> 8);
   case 1:
      *pLMBCS++ = (ulmbcs_byte_t)value;
   default:
      /* will never occur */
      break;
   }

   return (pLMBCS - pStartLMBCS);
}


/* This is a much simpler version of above, when we 
know we are writing LMBCS using the Unicode group
*/
static size_t 
LMBCSConvertUni(ulmbcs_byte_t * pLMBCS, UChar uniChar)  
{
     /* encode into LMBCS Unicode range */
   uint8_t LowCh =   (uint8_t)(uniChar & 0x00FF);
   uint8_t HighCh  = (uint8_t)(uniChar >> 8);

   *pLMBCS++ = ULMBCS_GRP_UNICODE;

   if (LowCh == 0)
   {
      *pLMBCS++ = ULMBCS_UNICOMPATZERO;
      *pLMBCS++ = HighCh;
   }
   else
   {
      *pLMBCS++ = HighCh;
      *pLMBCS++ = LowCh;
   }
   return ULMBCS_UNICODE_SIZE;
}


/* The main Unicode to LMBCS conversion function */
void 
_LMBCSFromUnicode(UConverterFromUnicodeArgs*     args,
                  UErrorCode*     err)
{
   ulmbcs_byte_t lastConverterIndex = 0;
   UChar uniChar;
   ulmbcs_byte_t  LMBCS[ULMBCS_CHARSIZE_MAX];
   ulmbcs_byte_t  * pLMBCS;
   int bytes_written;
   UBool groups_tried[ULMBCS_GRP_LAST];
   UConverterDataLMBCS * extraInfo = (UConverterDataLMBCS *) args->converter->extraInfo;
   int sourceIndex = 0; 


   /* Basic strategy: attempt to fill in local LMBCS 1-char buffer.(LMBCS)
      If that succeeds, see if it will all fit into the target & copy it over 
      if it does.

      We try conversions in the following order:

      1. Single-byte ascii & special fixed control chars (&null)
      2. Look up group in table & try that (could be 
            A) Unicode group
            B) control group,
            C) national encoding, 
               or ambiguous SBCS or MBCS group (on to step 4...)
        
      3. If its ambiguous, try this order:
         A) The optimization group
         B) The locale group
         C) The last group that succeeded with this string.
         D) every other group that's relevent (single or double)
         E) If its single-byte ambiguous, try the exceptions group

      4. And as a grand fallback: Unicode
   */

   while (args->source < args->sourceLimit && !U_FAILURE(*err))
   {
      if (args->target >= args->targetLimit)
      {
         *err = U_BUFFER_OVERFLOW_ERROR;
         break;
      }
      uniChar = *(args->source);
      bytes_written = 0;
      pLMBCS = LMBCS;

      /* check cases in rough order of how common they are, for speed */

      /* single byte matches: strategy 1 */

      if (((uniChar > ULMBCS_C0END) && (uniChar < ULMBCS_C1START)) ||
          uniChar == 0 || uniChar == ULMBCS_HT || uniChar == ULMBCS_CR || 
          uniChar == ULMBCS_LF || uniChar == ULMBCS_123SYSTEMRANGE 
          )
      {
         *pLMBCS++ = (ulmbcs_byte_t ) uniChar;
         bytes_written = 1;
      }


      if (!bytes_written) 
      {
         /* Check by UNICODE range (Strategy 2) */
         ulmbcs_byte_t group = FindLMBCSUniRange(uniChar);
         
         if (group == ULMBCS_GRP_UNICODE)  /* (Strategy 2A) */
         {
            pLMBCS += LMBCSConvertUni(pLMBCS,uniChar);
            
            bytes_written = pLMBCS - LMBCS;
         }
         else if (group == ULMBCS_GRP_CTRL)  /* (Strategy 2B) */
         {
            /* Handle control characters here */
            if (uniChar <= ULMBCS_C0END)
            {
               *pLMBCS++ = ULMBCS_GRP_CTRL;
               *pLMBCS++ = (ulmbcs_byte_t)(ULMBCS_CTRLOFFSET + uniChar);
            }
            else if (uniChar >= ULMBCS_C1START && uniChar <= ULMBCS_C1START + ULMBCS_CTRLOFFSET)
            {
               *pLMBCS++ = ULMBCS_GRP_CTRL;
               *pLMBCS++ = (ulmbcs_byte_t ) (uniChar & 0x00FF);
            }
            bytes_written = pLMBCS - LMBCS;
         }
         else if (group < ULMBCS_GRP_UNICODE)  /* (Strategy 2C) */
         {
            /* a specific converter has been identified - use it */
            bytes_written = LMBCSConversionWorker (
                              extraInfo, group, pLMBCS, &uniChar, 
                              &lastConverterIndex, groups_tried);
         }
         if (!bytes_written)    /* the ambiguous group cases  (Strategy 3) */
         {
            memset(groups_tried, 0, sizeof(groups_tried));

         /* check for non-default optimization group (Strategy 3A )*/
            if (extraInfo->OptGroup != 1 
                  && ULMBCS_AMBIGUOUS_MATCH(group, extraInfo->OptGroup)) 
            {
               bytes_written = LMBCSConversionWorker (extraInfo, 
                  extraInfo->OptGroup, pLMBCS, &uniChar, 
                  &lastConverterIndex, groups_tried);
            }
            /* check for locale optimization group (Strategy 3B) */
            if (!bytes_written 
               && (extraInfo->localeConverterIndex) 
               && (ULMBCS_AMBIGUOUS_MATCH(group, extraInfo->localeConverterIndex)))
               {
                  bytes_written = LMBCSConversionWorker (extraInfo, 
                     extraInfo->localeConverterIndex, pLMBCS, &uniChar, 
                     &lastConverterIndex, groups_tried);
               }
            /* check for last optimization group used for this string (Strategy 3C) */
            if (!bytes_written 
                && (lastConverterIndex) 
               && (ULMBCS_AMBIGUOUS_MATCH(group, lastConverterIndex)))
               {
                  bytes_written = LMBCSConversionWorker (extraInfo, 
                     lastConverterIndex, pLMBCS, &uniChar, 
                     &lastConverterIndex, groups_tried);
           
               }
            if (!bytes_written)
            {
               /* just check every possible matching converter (Strategy 3D) */ 
               ulmbcs_byte_t grp_start;
               ulmbcs_byte_t grp_end;  
               ulmbcs_byte_t grp_ix;
               grp_start = (ulmbcs_byte_t)((group == ULMBCS_AMBIGUOUS_MBCS) 
                        ? ULMBCS_DOUBLEOPTGROUP_START 
                        :  ULMBCS_GRP_L1);
               grp_end = (ulmbcs_byte_t)((group == ULMBCS_AMBIGUOUS_MBCS) 
                        ? ULMBCS_GRP_LAST 
                        :  ULMBCS_GRP_TH);
               for (grp_ix = grp_start;
                   grp_ix <= grp_end && !bytes_written; 
                    grp_ix++)
               {
                  if (extraInfo->OptGrpConverter [grp_ix] && !groups_tried [grp_ix])
                  {
                     bytes_written = LMBCSConversionWorker (extraInfo, 
                       grp_ix, pLMBCS, &uniChar, 
                       &lastConverterIndex, groups_tried);
                  }
               }
                /* a final conversion fallback to the exceptions group if its likely 
                     to be single byte  (Strategy 3E) */
               if (!bytes_written && grp_start == ULMBCS_GRP_L1)
               {
                  bytes_written = LMBCSConversionWorker (extraInfo, 
                     ULMBCS_GRP_EXCEPT, pLMBCS, &uniChar, 
                     &lastConverterIndex, groups_tried);
               }
            }
            /* all of our other strategies failed. Fallback to Unicode. (Strategy 4)*/
            if (!bytes_written)
            {

               pLMBCS += LMBCSConvertUni(pLMBCS, uniChar);
               bytes_written = pLMBCS - LMBCS;
            }
         }
      }
  
      /* we have a translation. increment source and write as much as posible to target */
      args->source++;
      pLMBCS = LMBCS;
      while (args->target < args->targetLimit && bytes_written--)
      {
         *(args->target)++ = *pLMBCS++;
         if (args->offsets)
         {
            *(args->offsets)++ = sourceIndex;
         }
      }
      sourceIndex++;
      if (bytes_written > 0)
      {
         /* write any bytes that didn't fit in target to the error buffer,
            common code will move this to target if we get called back with
            enough target room
         */
         uint8_t * pErrorBuffer = args->converter->charErrorBuffer;
         *err = U_BUFFER_OVERFLOW_ERROR;
         args->converter->charErrorBufferLength = (int8_t)bytes_written;
         while (bytes_written--)
         {
            *pErrorBuffer++ = *pLMBCS++;
         }
      }
   }     
}


/* Now, the Unicode from LMBCS section */


/* A function to call when we are looking at the Unicode group byte in LMBCS */
static UChar
GetUniFromLMBCSUni(char const ** ppLMBCSin)  /* Called with LMBCS-style Unicode byte stream */
{
   uint8_t  HighCh = *(*ppLMBCSin)++;  /* Big-endian Unicode in LMBCS compatibility group*/
   uint8_t  LowCh  = *(*ppLMBCSin)++;

   if (HighCh == ULMBCS_UNICOMPATZERO ) 
   {
      HighCh = LowCh;
      LowCh = 0; /* zero-byte in LSB special character */
   }
   return (UChar)((HighCh << 8) | LowCh);
}


/* CHECK_SOURCE_LIMIT: Helper macro to verify that there are at least'index' 
   bytes left in source up to  sourceLimit.Errors appropriately if not 
*/

#define CHECK_SOURCE_LIMIT(index) \
     if (args->source+index > args->sourceLimit){\
         *err = U_TRUNCATED_CHAR_FOUND;\
         args->source = saveSource;\
         return 0xffff;}


/* Return the Unicode representation for the current LMBCS character

   This worker function is used by both ucnv_getNextUChar() and ucnv_ToUnicode().  
   The last parameter says whether the return value should be treated as UTF-16 or
   UTF-32. The only difference is in surrogate handling
*/

static UChar32 
_LMBCSGetNextUCharWorker(UConverterToUnicodeArgs*   args,
                         UErrorCode*   err,
                         UBool         returnUTF32)
{
   ulmbcs_byte_t   CurByte; /* A byte from the input stream */
   UChar32 uniChar;    /* an output UNICODE char */
   const char * saveSource;
  
   /* error check */
   if (args->source >= args->sourceLimit)
   {
      *err = U_ILLEGAL_ARGUMENT_ERROR;
      return 0xffff;
   }
   /* Grab first byte & save address for error recovery */
   CurByte = *((ulmbcs_byte_t  *) (saveSource = args->source++));
   
   /*
    * at entry of each if clause:
    * 1. 'CurByte' points at the first byte of a LMBCS character
    * 2. '*source'points to the next byte of the source stream after 'CurByte' 
    *
    * the job of each if clause is:
    * 1. set '*source' to point at the beginning of next char (nop if LMBCS char is only 1 byte)
    * 2. set 'uniChar' up with the right Unicode value, or set 'err' appropriately
    */
   
   /* First lets check the simple fixed values. */

   if(((CurByte > ULMBCS_C0END) && (CurByte < ULMBCS_C1START)) /* ascii range */
      ||  (CurByte == 0) 
      ||  CurByte == ULMBCS_HT || CurByte == ULMBCS_CR 
      ||  CurByte == ULMBCS_LF || CurByte == ULMBCS_123SYSTEMRANGE)
   {
      uniChar = CurByte;
   }
   else  
   {
      UConverterDataLMBCS * extraInfo;
      ulmbcs_byte_t group; 
      UConverter* cnv; 
            
      if (CurByte == ULMBCS_GRP_CTRL)  /* Control character group - no opt group update */
      {
         ulmbcs_byte_t  C0C1byte;
         CHECK_SOURCE_LIMIT(1);
         C0C1byte = *(args->source)++;
         uniChar = (C0C1byte < ULMBCS_C1START) ? C0C1byte - ULMBCS_CTRLOFFSET : C0C1byte;
      }
      else 
      if (CurByte == ULMBCS_GRP_UNICODE) /* Unicode compatibility group: BigEndian UTF16 */
      {
         UChar second;
         CHECK_SOURCE_LIMIT(2);
         
         uniChar = GetUniFromLMBCSUni(&(args->source));
            
         /* at this point we are usually done, but we need to make sure we are not in 
         a situation where we can successfully put together a surrogate pair */

         if(returnUTF32 && UTF_IS_FIRST_SURROGATE(uniChar) && (args->source+3 <= args->sourceLimit)
            && *(args->source)++ == ULMBCS_GRP_UNICODE
            && UTF_IS_SECOND_SURROGATE(second = GetUniFromLMBCSUni(&(args->source))))
         {
               uniChar = UTF16_GET_PAIR_VALUE(uniChar, second);
         }
      }
      else if (CurByte <= ULMBCS_CTRLOFFSET)  
      {
         group = CurByte;                   /* group byte is in the source */
         extraInfo = (UConverterDataLMBCS *) args->converter->extraInfo;
         cnv = extraInfo->OptGrpConverter[group];
      
         if (!cnv)
         {
            /* this is not a valid group byte - no converter*/
            *err = U_INVALID_CHAR_FOUND;
         }
      
         else if (group >= ULMBCS_DOUBLEOPTGROUP_START)    /* double byte conversion */
         {

            CHECK_SOURCE_LIMIT(2);

            /* check for LMBCS doubled-group-byte case */
            if (*args->source == group) {
               /* single byte */
               ++args->source;
               uniChar = _MBCSSimpleGetNextUChar(cnv->sharedData, &args->source, args->source + 1, FALSE);
            } else {
               /* double byte */
               const char *newLimit = args->source + 2;
               uniChar = _MBCSSimpleGetNextUChar(cnv->sharedData, &args->source, newLimit, FALSE);
               args->source = newLimit; /* set the correct limit even in case of an error */
            }
         }
         else {                                  /* single byte conversion */
            CHECK_SOURCE_LIMIT(1);
            CurByte = *(args->source)++;
            
            if (CurByte >= ULMBCS_C1START)
            {
               uniChar = _MBCS_SINGLE_SIMPLE_GET_NEXT_BMP(cnv->sharedData, CurByte);
               /* SBCS was: uniChar = cnv->sharedData->table->sbcs.toUnicode[CurByte]; */
            }
            else
            {
            /* The non-optimizable oddballs where there is an explicit byte 
             * AND the second byte is not in the upper ascii range
            */
               const char *s;
               char bytes[2];

               extraInfo = (UConverterDataLMBCS *) args->converter->extraInfo;
               cnv = extraInfo->OptGrpConverter [ULMBCS_GRP_EXCEPT];  
            
            /* Lookup value must include opt group */
               bytes[0] = group;
               bytes[1] = CurByte;
               s = bytes;
               uniChar = _MBCSSimpleGetNextUChar(cnv->sharedData, &s, bytes + 2, FALSE);
            }
         }
      }
      else if (CurByte >= ULMBCS_C1START) /* group byte is implicit */
      {
         extraInfo = (UConverterDataLMBCS *) args->converter->extraInfo;
         group = extraInfo->OptGroup;
         cnv = extraInfo->OptGrpConverter[group];
         if (group >= ULMBCS_DOUBLEOPTGROUP_START)    /* double byte conversion */
         {
            if (!_MBCSIsLeadByte(cnv->sharedData, CurByte))
            {
               CHECK_SOURCE_LIMIT(0);

               /* let the MBCS conversion consume CurByte again */
               --args->source;
               uniChar = _MBCSSimpleGetNextUChar(cnv->sharedData, &args->source, args->source + 1, FALSE);
            }
            else
            {
               CHECK_SOURCE_LIMIT(1);

               /* let the MBCS conversion consume CurByte again */
               --args->source;

               /* since we know that we start at a lead byte, args->source _will_ be incremented by 2 */
               uniChar = _MBCSSimpleGetNextUChar(cnv->sharedData, &args->source, args->source + 2, FALSE);
            }
         }
         else                                   /* single byte conversion */
         {
            uniChar = _MBCS_SINGLE_SIMPLE_GET_NEXT_BMP(cnv->sharedData, CurByte);
            /* SBCS was: uniChar = cnv->sharedData->table->sbcs.toUnicode[CurByte]; */
         }
      }
   }
   if (((uint32_t)uniChar - 0xfffe) <= 1) /* 0xfffe<=uniChar<=0xffff */
   {
       /*It is very likely that the ErrorFunctor will write to the
       *internal buffers */

      /* This code needs updating when new error callbacks are installed */
      UConverterToUnicodeArgs cbArgs = *args;
      UChar * pUniChar = (UChar *)&uniChar;
      UConverterCallbackReason reason;

      if (uniChar == 0xfffe)
      {
        reason = UCNV_UNASSIGNED;
        *err = U_INVALID_CHAR_FOUND;
      }
      else
      {
        reason = UCNV_ILLEGAL;
        *err = U_ILLEGAL_CHAR_FOUND;
      }

      cbArgs.target = pUniChar;
      cbArgs.targetLimit = pUniChar + 1;
      cbArgs.converter->fromCharErrorBehaviour(cbArgs.converter->toUContext,
                                    &cbArgs,
                                    saveSource,
                                    args->sourceLimit - saveSource,
                                    reason,
                                    err);
   }
   return uniChar;
}


/* The exported function that gets one UTF32 character from a LMBCS stream
*/
UChar32 
_LMBCSGetNextUChar(UConverterToUnicodeArgs*   args,
                   UErrorCode*   err)
{
   return _LMBCSGetNextUCharWorker(args, err, TRUE);
}

/* The exported function that converts lmbcs to one or more
   UChars - currently UTF-16
*/
void 
_LMBCSToUnicodeWithOffsets(UConverterToUnicodeArgs*    args,
                     UErrorCode*    err)
{
   UChar uniChar;    /* one output UNICODE char */
   const char * saveSource = args->source;
   const char * pStartLMBCS = args->source;  /* beginning of whole string */

   if (args->targetLimit == args->target)         /* error check may belong in common code */
   {
      *err = U_BUFFER_OVERFLOW_ERROR;
      return;
   }
   
   /* Process from source to limit, or until error */
   while (!*err && args->sourceLimit > args->source && args->targetLimit > args->target)
   {
      saveSource = args->source; /* beginning of current code point */

      if (args->converter->invalidCharLength) /* reassemble char from previous call */
      {
         char LMBCS [ULMBCS_CHARSIZE_MAX];
         const char *pLMBCS = LMBCS, /* *saveSource,*/ *saveSourceLimit; 
         size_t size_old = args->converter->invalidCharLength;

         /* limit from source is either reminder of temp buffer, or user limit on source */
         size_t size_new_maybe_1 = sizeof(LMBCS) - size_old;
         size_t size_new_maybe_2 = args->sourceLimit - args->source;
         size_t size_new = (size_new_maybe_1 < size_new_maybe_2) ? size_new_maybe_1 : size_new_maybe_2;
         
      
         uprv_memcpy(LMBCS, args->converter->invalidCharBuffer, size_old);
         uprv_memcpy(LMBCS + size_old, args->source, size_new);
         /* saveSource = (char*)args->source; */ /* Already done */
         saveSourceLimit = args->sourceLimit;
         args->source = pLMBCS;
         args->sourceLimit = pLMBCS+size_old+size_new;
         uniChar = (UChar) _LMBCSGetNextUCharWorker(args, err, FALSE);
         pLMBCS = args->source;
         args->source =saveSource;
         args->sourceLimit = saveSourceLimit;
         args->source += (pLMBCS - LMBCS - size_old);

         if (*err == U_TRUNCATED_CHAR_FOUND && !args->flush)
         {
            /* evil special case: source buffers so small a char spans more than 2 buffers */
            int8_t savebytes = (int8_t)(size_old+size_new);
            args->converter->invalidCharLength = savebytes;
            uprv_memcpy(args->converter->invalidCharBuffer, LMBCS, savebytes);
            args->source = args->sourceLimit;
            *err = U_ZERO_ERROR;
            return;
         }
         else
         {
            /* clear the partial-char marker */
            args->converter->invalidCharLength = 0;
         }
      }
      else
      {
         uniChar = (UChar) _LMBCSGetNextUCharWorker(args, err, FALSE);
      }
      if (U_SUCCESS(*err))
      {
         if (uniChar < 0xfffe)
         {
            *(args->target)++ = uniChar;
            if(args->offsets)
            {
               *(args->offsets)++ = saveSource - pStartLMBCS;
            }
         }
         else if (uniChar == 0xfffe)
         {
            *err = U_INVALID_CHAR_FOUND;
         }
         else /* if (uniChar == 0xffff) */
         {
            *err = U_ILLEGAL_CHAR_FOUND;
         }
      }
   }
   /* if target ran out before source, return U_BUFFER_OVERFLOW_ERROR */
   if (U_SUCCESS(*err) && args->sourceLimit > args->source && args->targetLimit <= args->target)
   {
      *err = U_BUFFER_OVERFLOW_ERROR;
   }

   /* If character incomplete, store away partial char if more to come */
   if ((*err == U_TRUNCATED_CHAR_FOUND) && !args->flush )
   {
      int8_t savebytes = (int8_t)(args->sourceLimit - saveSource);
      args->converter->invalidCharLength = (int8_t)savebytes;
      uprv_memcpy(args->converter->invalidCharBuffer, saveSource, savebytes);
      args->source = args->sourceLimit;
      *err = U_ZERO_ERROR;
   }
}