Bug#45803 Inaccurate estimates for partial key values with IBMDB2I

Some collations were causing IBMDB2I to report
inaccurate key range estimations to the optimizer
for LIKE clauses that select substrings. This can
be seen by running EXPLAIN. This problem primarily
affects multi-byte and unicode character sets.

This patch involves substantial changes to several
modules. There are a number of problems with the
character set and collation handling. These problems
have been or are being fixed,  and a comprehensive
test has been included which should provide much
better coverage than there was before. This test
is enabled only for IBM i 6.1, because that version
has support for the greatest number of collations.

mysql-test/suite/ibmdb2i/r/ibmdb2i_collations.result:
  Bug#45803 Inaccurate estimates for partial key values with IBMDB2I
  
  result file for test case.
mysql-test/suite/ibmdb2i/t/ibmdb2i_collations.test:
  Bug#45803 Inaccurate estimates for partial key values with IBMDB2I
  
  Tests for character sets and collations. This test
  is enabled only for IBM i 6.1, because that version
  has support for the greatest number of collations.
storage/ibmdb2i/db2i_conversion.cc:
  Bug#45803 Inaccurate estimates for partial key values with IBMDB2I
  
  - Added support in convertFieldChars to enable records_in_range
    to determine how many substitute characters were inserted and
    to suppress conversion warnings.
  
  - Fixed bug which was causing all multi-byte and Unicode fields
    to be created as UTF16 (CCSID 1200) fields in DB2. The corrected
    code will now create UCS2 fields as UCS2 (CCSID 13488), UTF8
    fields (except for utf8_general_ci) as UTF8 (CCSID 1208), and
    all other multi-byte or Unicode fields as UTF16.  This will only
    affect tables that are newly created through the IBMDB2I storage
    engine. Existing IBMDB2I tables will retain the original CCSID
    until recreated. The existing behavior is believed to be
    functionally correct, but it may negatively impact performance
    by causing unnecessary character conversion. Additionally, users
    accessing IBMDB2I tables through DB2 should be aware that mixing 
    tables created before and after this change may require extra type
    casts or other workarounds.  For this reason, users who have
    existing IBMDB2I tables using a Unicode collation other than
    utf8_general_ci are encouraged to recreate their tables (e.g.
    ALTER TABLE t1 ENGINE=IBMDB2I) in order to get the updated CCSIDs
    associated with their DB2 tables.
  
  - Improved error reporting for unsupported character sets by forcing
    a check for the iconv conversion table at table creation time,
    rather than at data access time.
storage/ibmdb2i/db2i_myconv.h:
  Bug#45803 Inaccurate estimates for partial key values with IBMDB2I
  
  Fix to set errno when iconv fails.
storage/ibmdb2i/db2i_rir.cc:
  Bug#45803 Inaccurate estimates for partial key values with IBMDB2I
  
  Significant improvements were made to the records_in_range code
  that handles partial length string data in keys for optimizer plan
  estimation. Previously, to obtain an estimate for a partial key
  value, the implementation would perform any necessary character
  conversion and then attempt to determine the unpadded length of
  the partial key by searching for the minimum or maximum sort
  character. While this algorithm was sufficient for most single-byte
  character sets, it did not treat Unicode and multi-byte strings
  correctly. Furthermore, due to an operating system limitation,
  partial keys having UTF8 collations (ICU sort sequences in DB2)
  could not be estimated with this method.
  
  With this patch, the code no longer attempts to explicitly determine
  the unpadded length of the key. Instead, the entire key is converted
  (if necessary), including padding, and then passed to the operating
  system for estimation. Depending on the source and target character
  sets and collations, additional logic is required to correctly
  handle cases in which MySQL uses unconvertible or differently
  -weighted values to pad the key. The bulk of the patch exists
  to implement this additional logic.
storage/ibmdb2i/ha_ibmdb2i.h:
  Bug#45803 Inaccurate estimates for partial key values with IBMDB2I
  
  The convertFieldChars declaration was updated to support additional 
  optional behaviors.

4 files changed, 346 insertions, 110 deletions

diff --git a/storage/ibmdb2i/db2i_conversion.cc b/storage/ibmdb2i/db2i_conversion.cc
index 0acde10a4cd..9a85eb01c9b 100644
--- a/storage/ibmdb2i/db2i_conversion.cc
+++ b/storage/ibmdb2i/db2i_conversion.cc
@@ -137,7 +137,9 @@ int ha_ibmdb2i::convertFieldChars(enum_conversionDirection direction,
                                   char* output, 
                                   size_t ilen, 
                                   size_t olen, 
-                                  size_t* outDataLen)
+                                  size_t* outDataLen,
+                                  bool tacitErrors,
+                                  size_t* substChars)
 {
   DBUG_PRINT("ha_ibmdb2i::convertFieldChars",("Direction: %d; length = %d", direction, ilen));
   
@@ -157,26 +159,26 @@ int ha_ibmdb2i::convertFieldChars(enum_conversionDirection direction,
   size_t initOLen= olen;
   size_t substitutedChars = 0;
   int rc = iconv(conversion, (char**)&input, &ilen, &output, &olen, &substitutedChars );
+  if (outDataLen) *outDataLen = initOLen - olen;
+  if (substChars) *substChars = substitutedChars;
   if (unlikely(rc < 0))
   {
     int er = errno;
     if (er == EILSEQ)
     {
-      getErrTxt(DB2I_ERR_ILL_CHAR, table->field[fieldID]->field_name);
+      if (!tacitErrors) getErrTxt(DB2I_ERR_ILL_CHAR, table->field[fieldID]->field_name);
       return (DB2I_ERR_ILL_CHAR);
     }
     else
     {
-      getErrTxt(DB2I_ERR_ICONV,er);
+      if (!tacitErrors) getErrTxt(DB2I_ERR_ICONV,er);
       return (DB2I_ERR_ICONV);
     }
   }
-  if (unlikely(substitutedChars))
+  if (unlikely(substitutedChars) && (!tacitErrors))
   {
     warning(ha_thd(), DB2I_ERR_SUB_CHARS, table->field[fieldID]->field_name);
   }
-    
-  if (outDataLen) *outDataLen = initOLen - olen;
 
   return (0);
 }
@@ -555,12 +557,12 @@ int ha_ibmdb2i::getFieldTypeMapping(Field* field,
                 return 1;
               if (fieldCharSet->mbmaxlen > 1)
               {
-                if (strncmp(fieldCharSet->name, "ucs2_", sizeof("ucs2_")) == 0 ) // UCS2
+                if (memcmp(fieldCharSet->name, "ucs2_", sizeof("ucs2_")-1) == 0 ) // UCS2
                 {
                   sprintf(stringBuildBuffer, "GRAPHIC(%d)", max(fieldLength / fieldCharSet->mbmaxlen, 1)); // Number of characters
                   db2Ccsid = 13488;
                 }
-                else if (strncmp(fieldCharSet->name, "utf8_", sizeof("utf8_")) == 0 &&
+                else if (memcmp(fieldCharSet->name, "utf8_", sizeof("utf8_")-1) == 0 &&
                          strcmp(fieldCharSet->name, "utf8_general_ci") != 0) 
                 {
                   sprintf(stringBuildBuffer, "CHAR(%d)", max(fieldLength, 1)); // Number of bytes
@@ -584,12 +586,12 @@ int ha_ibmdb2i::getFieldTypeMapping(Field* field,
               {
                 if (fieldCharSet->mbmaxlen > 1)
                 {
-                  if (strncmp(fieldCharSet->name, "ucs2_", sizeof("ucs2_")) == 0 ) // UCS2
+                  if (memcmp(fieldCharSet->name, "ucs2_", sizeof("ucs2_")-1) == 0 ) // UCS2
                   {
                     sprintf(stringBuildBuffer, "VARGRAPHIC(%d)", max(fieldLength / fieldCharSet->mbmaxlen, 1)); // Number of characters
                     db2Ccsid = 13488;
                   }
-                  else if (strncmp(fieldCharSet->name, "utf8_", sizeof("utf8_")) == 0 &&
+                  else if (memcmp(fieldCharSet->name, "utf8_", sizeof("utf8_")-1) == 0 &&
                            strcmp(fieldCharSet->name, "utf8_general_ci") != 0) 
                   {
                     sprintf(stringBuildBuffer, "VARCHAR(%d)", max(fieldLength, 1)); // Number of bytes
@@ -611,12 +613,12 @@ int ha_ibmdb2i::getFieldTypeMapping(Field* field,
               {
                 if (fieldCharSet->mbmaxlen > 1)
                 {
-                  if (strncmp(fieldCharSet->name, "ucs2_", sizeof("ucs2_")) == 0 ) // UCS2
+                  if (memcmp(fieldCharSet->name, "ucs2_", sizeof("ucs2_")-1) == 0 ) // UCS2
                   {
                     sprintf(stringBuildBuffer, "LONG VARGRAPHIC ");
                     db2Ccsid = 13488;
                   }
-                  else if (strncmp(fieldCharSet->name, "utf8_", sizeof("utf8_")) == 0 &&
+                  else if (memcmp(fieldCharSet->name, "utf8_", sizeof("utf8_")-1) == 0 &&
                            strcmp(fieldCharSet->name, "utf8_general_ci") != 0) 
                   {
                     sprintf(stringBuildBuffer, "LONG VARCHAR ");
@@ -639,12 +641,12 @@ int ha_ibmdb2i::getFieldTypeMapping(Field* field,
 
                 if (fieldCharSet->mbmaxlen > 1)
                 {
-                  if (strncmp(fieldCharSet->name, "ucs2_", sizeof("ucs2_")) == 0 ) // UCS2
+                  if (memcmp(fieldCharSet->name, "ucs2_", sizeof("ucs2_")-1) == 0 ) // UCS2
                   {
                     sprintf(stringBuildBuffer, "DBCLOB(%d)", max(fieldLength / fieldCharSet->mbmaxlen, 1)); // Number of characters
                     db2Ccsid = 13488;
                   }
-                  else if (strncmp(fieldCharSet->name, "utf8_", sizeof("utf8_")) == 0 &&
+                  else if (memcmp(fieldCharSet->name, "utf8_", sizeof("utf8_")-1) == 0 &&
                            strcmp(fieldCharSet->name, "utf8_general_ci") != 0) 
                   {
                     sprintf(stringBuildBuffer, "CLOB(%d)", max(fieldLength, 1)); // Number of bytes
@@ -671,11 +673,15 @@ int ha_ibmdb2i::getFieldTypeMapping(Field* field,
                 return rtnCode;
             }
             
-            // Check whether there is a character conversion available.
-            iconv_t temp;
-            int32 rc = getConversion(toDB2, fieldCharSet, db2Ccsid, temp);
-            if (unlikely(rc))
-              return rc;
+            if (db2Ccsid != 1208 &&
+                db2Ccsid != 13488)
+            {
+              // Check whether there is a character conversion available.
+              iconv_t temp;
+              int32 rc = getConversion(toDB2, fieldCharSet, db2Ccsid, temp);
+              if (unlikely(rc))
+                return rc;
+            }
             
             sprintf(stringBuildBuffer, " CCSID %d ", db2Ccsid);
             mapping.append(stringBuildBuffer);
diff --git a/storage/ibmdb2i/db2i_myconv.h b/storage/ibmdb2i/db2i_myconv.h
index a9e87474505..98032748148 100644
--- a/storage/ibmdb2i/db2i_myconv.h
+++ b/storage/ibmdb2i/db2i_myconv.h
@@ -220,6 +220,7 @@ INTERN  size_t	        myconv_dmap(myconv_t    cd,
       } else {
         *pOut=dmapS2S[*pIn];
         if (*pOut == 0x00) {
+          errno=EILSEQ;  /* 116 */
           *outBytesLeft-=(*inBytesLeft-inLen);
           *inBytesLeft=inLen;
           *outBuf=pOut;
diff --git a/storage/ibmdb2i/db2i_rir.cc b/storage/ibmdb2i/db2i_rir.cc
index a80a181c9ac..091c4d98383 100644
--- a/storage/ibmdb2i/db2i_rir.cc
+++ b/storage/ibmdb2i/db2i_rir.cc
@@ -51,7 +51,6 @@ static inline int getKeyCntFromMap(key_part_map keypart_map)
   return (cnt); 
 }
 
-
 /**
   @brief
   Given a starting key and an ending key, estimate the number of rows that
@@ -270,81 +269,163 @@ ha_rows ha_ibmdb2i::records_in_range(uint inx,
         DB2Field& db2Field = db2Table->db2Field(field->field_index);
         litDefPtr->DataType = db2Field.getType();
         /*
-           Convert the literal to DB2 format.  
-                                                                                   */
-        rc = convertMySQLtoDB2(field,
-                 db2Field,
-                 literalPtr,
-                 (uchar*)minPtr+((curKey.key_part[partsInUse].null_bit)? 1 : 0));
-        if (rc != 0) break;
-        litDefPtr->Offset = (uint32_t)(literalPtr - literalsPtr);
-        litDefPtr->Length = db2Field.getByteLengthInRecord();
-        tempLen = litDefPtr->Length;
-        /*
-           Do additional conversion of a character or graphic value.
-                                                                                 */
-        CHARSET_INFO* fieldCharSet = field->charset();                                    
+           Convert the literal to DB2 format
+                                                                                               */
         if ((field->type() != MYSQL_TYPE_BIT) &&           // Don't do conversion on BIT data
             (field->charset() != &my_charset_bin) &&       // Don't do conversion on BINARY data
-            (litDefPtr->DataType == QMY_CHAR || litDefPtr->DataType == QMY_VARCHAR ||
-             litDefPtr->DataType == QMY_GRAPHIC || litDefPtr->DataType == QMY_VARGRAPHIC))
+            (litDefPtr->DataType == QMY_CHAR ||
+             litDefPtr->DataType == QMY_VARCHAR ||
+             litDefPtr->DataType == QMY_GRAPHIC ||
+             litDefPtr->DataType == QMY_VARGRAPHIC))
         {
-           if (litDefPtr->DataType == QMY_VARCHAR ||
-               litDefPtr->DataType == QMY_VARGRAPHIC) 
-             tempPtr = literalPtr + sizeof(uint16);
-           else
-             tempPtr = literalPtr;
-           /* The following code checks to determine if MySQL is passing a
-              partial key. DB2 will accept a partial field value, but only
-              in the last field position of the key composite (and only if
-              there is no ICU sort sequence on the index).                  */
-           tempMinPtr = (char*)minPtr+((curKey.key_part[partsInUse].null_bit)? 1 : 0);
-           if (field->type() == MYSQL_TYPE_VARCHAR)
-           {
-             /* MySQL always stores key lengths as 2 bytes, little-endian. */
-               tempLen = *(uint8*)tempMinPtr + ((*(uint8*)(tempMinPtr+1)) << 8);
-               tempMinPtr = (char*)((char*)tempMinPtr + 2);
-           }
-           else
-             tempLen = field->field_length;                         
-
-           /* Determine if we are dealing with a partial key and if so, find the end of the partial key. */
-           if (litDefPtr->DataType == QMY_CHAR || litDefPtr->DataType == QMY_VARCHAR )
-           { /* Char or varchar.  If UTF8, no conversion is done to DB2 graphic.) */
-               endOfMinPtr = (char*)memchr(tempMinPtr,field->charset()->min_sort_char,tempLen);
-               if (endOfMinPtr)
-                 endOfLiteralPtr = tempPtr + ((uint32_t)(endOfMinPtr - tempMinPtr));
-           }
-           else
-           {
-              if (strncmp(fieldCharSet->csname, "utf8", sizeof("utf8")) == 0)
-              {  /* The MySQL charset is UTF8 but we are converting to graphic on DB2. Divide number of UTF8 bytes
-                    by 3 to get the number of characters, then multiple by 2 for double-byte graphic.*/
-               endOfMinPtr = (char*)memchr(tempMinPtr,field->charset()->min_sort_char,tempLen);
-               if (endOfMinPtr)
-                 endOfLiteralPtr = tempPtr + (((uint32_t)((endOfMinPtr - tempMinPtr)) / 3) * 2);
-              } 
-              else
-              { /* The DB2 data type is graphic or vargraphic, and we are not converting from UTF8 to graphic. */  
-                endOfMinPtr = (char*)wmemchr((wchar_t*)tempMinPtr,field->charset()->min_sort_char,tempLen/2);
-                if (endOfMinPtr)
-                  endOfLiteralPtr = tempPtr + (endOfMinPtr - tempMinPtr);
+          // Most of the code is required by the considerable wrangling needed
+          // to prepare partial keys for use by DB2
+          // 1. UTF8 (CCSID 1208) data can be copied across unmodified if it is
+          //    utf8_bin. Otherwise, we need to convert the min and max
+          //    characters into the min and max characters employed
+          //    by the DB2 sort sequence. This is complicated by the fact that
+          //    the character widths are not always equal.
+          //  2. Likewise, UCS2 (CCSID 13488) data can be copied across unmodified
+          //     if it is ucs2_bin or ucs2_general_ci. Otherwise, we need to
+          //     convert the min and max characters into the min and max characters
+          //     employed by the DB2 sort sequence.
+          //  3. All other data will use standard iconv conversions. If an
+          //     unconvertible character is encountered, we assume it is the min
+          //     char and fill the remainder of the DB2 key with 0s. This may not
+          //     always be accurate, but it is probably sufficient for range
+          //     estimations.
+          const char* keyData = minPtr+((curKey.key_part[partsInUse].null_bit)? 1 : 0);
+          char* db2Data = literalPtr;
+          uint16 outLen = db2Field.getByteLengthInRecord();
+          uint16 inLen;
+          if (litDefPtr->DataType == QMY_VARCHAR ||
+              litDefPtr->DataType == QMY_VARGRAPHIC)
+          {
+            inLen = *(uint8*)keyData + ((*(uint8*)(keyData+1)) << 8);
+            keyData += 2;
+            outLen -= sizeof(uint16);
+            db2Data += sizeof(uint16);
+          }
+          else
+          {
+            inLen = field->max_display_length();
+          }
+          
+          size_t convertedBytes = 0;
+          if (db2Field.getCCSID() == 1208)
+          {
+            DBUG_ASSERT(inLen <= outLen);
+            if (strcmp(field->charset()->name, "utf8_bin"))
+            {
+              const char* end = keyData+inLen;
+              const char* curKey = keyData;
+              char* curDB2 = db2Data;
+              uint32 min = field->charset()->min_sort_char;
+              while ((curKey < end) && (curDB2 < db2Data+outLen-3))
+              {
+                my_wc_t temp;
+                int len = field->charset()->cset->mb_wc(field->charset(),
+                                                        &temp, 
+                                                        (const uchar*)curKey, 
+                                                        (const uchar*)end);
+                if (temp != min)
+                {
+                  DBUG_ASSERT(len <= 3);
+                  switch (len)
+                  {
+                    case 3: *(curDB2+2) = *(curKey+2);
+                    case 2: *(curDB2+1) = *(curKey+1);
+                    case 1: *(curDB2) = *(curKey);
+                  }                      
+                  curDB2 += len;
+                }
+                else
+                {
+                  *(curDB2++) = 0xEF;
+                  *(curDB2++) = 0xBF;
+                  *(curDB2++) = 0xBF;
+                }
+                curKey += len;
+              }
+              convertedBytes = curDB2 - db2Data;
+            }
+            else
+            {
+              memcpy(db2Data, keyData, inLen);
+              convertedBytes = inLen;
+            }
+            rc = 0;
+          }
+          else if (db2Field.getCCSID() == 13488)
+          {
+            DBUG_ASSERT(inLen <= outLen);
+            if (strcmp(field->charset()->name, "ucs2_bin") &&
+                strcmp(field->charset()->name, "ucs2_general_ci"))
+            {
+              const char* end = keyData+inLen;
+              const uint16* curKey = (uint16*)keyData;
+              uint16* curDB2 = (uint16*)db2Data;
+              uint16 min = field->charset()->min_sort_char;
+              while (curKey < (uint16*)end)
+              {
+                if (*curKey != min)
+                  *curDB2 = *curKey;
+                else
+                  *curDB2 = 0xFFFF;
+                ++curKey;
+                ++curDB2;
               }
-           }
-           /* Enforce here that a partial is only allowed on the last field position 
-              of the key composite                                                    */
-           if (endOfLiteralPtr)
-           {
-             if ((partsInUse + 1) < minKeyCnt)   
-             {
-               rc = HA_POS_ERROR;
-               break;
-             }
-             endByte = endOfLiteralPtr - tempPtr;
-             /* We're making an assumption that if MySQL gives us a partial key, 
-                 the length of the partial is the same for both the min_key and max_key.      */
-           }  
+            }
+            else
+            {
+              memcpy(db2Data, keyData, inLen);
+            }
+            convertedBytes = inLen;
+            rc = 0;
+          }
+          else
+          {
+            rc = convertFieldChars(toDB2, 
+                                   field->field_index, 
+                                   keyData,
+                                   db2Data,
+                                   inLen,
+                                   outLen,
+                                   &convertedBytes,
+                                   true);
+
+            if (rc == DB2I_ERR_ILL_CHAR)
+            {
+              // If an illegal character is encountered, we fill the remainder
+              // of the key with 0x00. This was implemented as a corollary to
+              // Bug#45012, though it should probably remain even after that
+              // bug is fixed.
+              memset(db2Data+convertedBytes, 0x00, outLen-convertedBytes);
+              convertedBytes = outLen;
+              rc = 0;
+            }
+          }
+          
+          if (!rc &&
+              (litDefPtr->DataType == QMY_VARGRAPHIC ||
+               litDefPtr->DataType == QMY_VARCHAR))
+          {
+            *(uint16*)(db2Data-sizeof(uint16)) = 
+                convertedBytes / (litDefPtr->DataType == QMY_VARGRAPHIC ? 2 : 1);
+          }
+
         }
+        else // Non-character fields
+        {
+          rc = convertMySQLtoDB2(field,
+                                 db2Field,
+                                 literalPtr,
+                                 (uchar*)minPtr+((curKey.key_part[partsInUse].null_bit)? 1 : 0));
+        }
+
+        if (rc != 0) break;
+        litDefPtr->Offset = (uint32_t)(literalPtr - literalsPtr);
+        litDefPtr->Length = db2Field.getByteLengthInRecord();
         literalPtr = literalPtr + litDefPtr->Length;  // Bump pointer for next literal
       }
       /* If there is a max_key value for this field, and if the max_key value is 
@@ -389,28 +470,168 @@ ha_rows ha_ibmdb2i::records_in_range(uint inx,
         /*
            Convert the literal to DB2 format
                                                                                                */
-          rc = convertMySQLtoDB2(field,
-                 db2Field,
-                 literalPtr,
-                 (uchar*)maxPtr+((curKey.key_part[partsInUse].null_bit)? 1 : 0));
+        if ((field->type() != MYSQL_TYPE_BIT) &&           // Don't do conversion on BIT data
+            (field->charset() != &my_charset_bin) &&       // Don't do conversion on BINARY data
+            (litDefPtr->DataType == QMY_CHAR ||
+             litDefPtr->DataType == QMY_VARCHAR ||
+             litDefPtr->DataType == QMY_GRAPHIC ||
+             litDefPtr->DataType == QMY_VARGRAPHIC))
+          {
+            // We need to handle char fields in a special way in order to account
+            // for partial keys. Refer to the note above for a description of the
+            // basic design.
+            char* keyData = maxPtr+((curKey.key_part[partsInUse].null_bit)? 1 : 0);
+            char* db2Data = literalPtr;
+            uint16 outLen = db2Field.getByteLengthInRecord();
+            uint16 inLen;
+            if (litDefPtr->DataType == QMY_VARCHAR ||
+                litDefPtr->DataType == QMY_VARGRAPHIC)
+            {
+              inLen = *(uint8*)keyData + ((*(uint8*)(keyData+1)) << 8);
+              keyData += 2;
+              outLen -= sizeof(uint16);
+              db2Data += sizeof(uint16);
+            }
+            else
+            {
+              inLen = field->max_display_length();
+            }
+            
+            size_t convertedBytes;
+            if (db2Field.getCCSID() == 1208)
+            {
+              if (strcmp(field->charset()->name, "utf8_bin"))
+              {
+                const char* end = keyData+inLen;
+                const char* curKey = keyData;
+                char* curDB2 = db2Data;
+                uint32 max = field->charset()->max_sort_char;
+                while (curKey < end && (curDB2 < db2Data+outLen-3))
+                {
+                  my_wc_t temp;
+                  int len = field->charset()->cset->mb_wc(field->charset(), &temp, (const uchar*)curKey, (const uchar*)end);
+                  if (temp != max)
+                  {
+                    DBUG_ASSERT(len <= 3);
+                    switch (len)
+                    {
+                      case 3: *(curDB2+2) = *(curKey+2);
+                      case 2: *(curDB2+1) = *(curKey+1);
+                      case 1: *(curDB2) = *(curKey);
+                    }                      
+                    curDB2 += len;
+                  }
+                  else
+                  {
+                    *(curDB2++) = 0xE4;
+                    *(curDB2++) = 0xB6;
+                    *(curDB2++) = 0xBF;
+                  }
+                  curKey += len;
+                }
+                convertedBytes = curDB2 - db2Data;
+              }
+              else
+              {
+                DBUG_ASSERT(inLen <= outLen);
+                memcpy(db2Data, keyData, inLen);
+                convertedBytes = inLen;
+              }
+              rc = 0;
+            }
+            else if (db2Field.getCCSID() == 13488)
+            {
+              if (strcmp(field->charset()->name, "ucs2_bin") &&
+                  strcmp(field->charset()->name, "ucs2_general_ci"))
+              {
+                char* end = keyData+inLen;
+                uint16* curKey = (uint16*)keyData;
+                uint16* curDB2 = (uint16*)db2Data;
+                uint16 max = field->charset()->max_sort_char;
+                while (curKey < (uint16*)end)
+                {
+                  if (*curKey != max)
+                    *curDB2 = *curKey;
+                  else
+                    *curDB2 = 0x4DBF;
+                  ++curKey;
+                  ++curDB2;
+                }
+              }
+              else
+              {
+                memcpy(db2Data, keyData, outLen);
+              }
+              rc = 0;
+            }
+            else
+            {
+              size_t substituteChars = 0;
+              rc = convertFieldChars(toDB2, 
+                                     field->field_index, 
+                                     keyData,
+                                     db2Data,
+                                     inLen,
+                                     outLen,
+                                     &convertedBytes,
+                                     true,
+                                     &substituteChars);
+
+              if (rc == DB2I_ERR_ILL_CHAR)
+              {
+                // If an illegal character is encountered, we fill the remainder
+                // of the key with 0xFF. This was implemented to work around
+                // Bug#45012, though it should probably remain even after that
+                // bug is fixed.
+                memset(db2Data+convertedBytes, 0xFF, outLen-convertedBytes);
+                rc = 0;
+              }
+              else if ((substituteChars &&
+                        (litDefPtr->DataType == QMY_VARCHAR ||
+                         litDefPtr->DataType == QMY_CHAR)) ||
+                       strcmp(field->charset()->name, "cp1251_bulgarian_ci") == 0)
+              {
+                // When iconv translates the max_sort_char with a substitute 
+                // character, we have no way to know whether this affects
+                // the sort order of the key. Therefore, to be safe, when
+                // we know that substitute characters have been used in a
+                // single-byte string, we traverse the translated key
+                // in reverse, replacing substitue characters with 0xFF, which
+                // always sorts with the greatest weight in DB2 sort sequences.
+                // cp1251_bulgarian_ci is also handled this way because the
+                // max_sort_char is a control character which does not sort
+                // equivalently in DB2.
+                DBUG_ASSERT(inLen == outLen);
+                char* tmpKey = keyData + inLen - 1;
+                char* tmpDB2 = db2Data + outLen - 1;
+                while (*tmpKey == field->charset()->max_sort_char &&
+                       *tmpDB2 != 0xFF)
+                {
+                  *tmpDB2 = 0xFF;
+                  --tmpKey;
+                  --tmpDB2;
+                }                  
+              }
+            }
+            
+            if (!rc &&
+                (litDefPtr->DataType == QMY_VARGRAPHIC ||
+                 litDefPtr->DataType == QMY_VARCHAR))
+            {
+              *(uint16*)(db2Data-sizeof(uint16)) = 
+                  outLen / (litDefPtr->DataType == QMY_VARGRAPHIC ? 2 : 1);
+            }
+          }
+          else
+          {
+            rc = convertMySQLtoDB2(field,
+                                   db2Field,
+                                   literalPtr,
+                                   (uchar*)maxPtr+((curKey.key_part[partsInUse].null_bit)? 1 : 0));
+          }
           if (rc != 0) break;
           litDefPtr->Offset = (uint32_t)(literalPtr - literalsPtr);
           litDefPtr->Length = db2Field.getByteLengthInRecord();
-          tempLen = litDefPtr->Length;
-          /*
-             Now convert a character or graphic value.
-                                                                                 */
-          if ((field->type() != MYSQL_TYPE_BIT) &&     
-             (litDefPtr->DataType == QMY_CHAR || litDefPtr->DataType == QMY_VARCHAR ||
-              litDefPtr->DataType == QMY_GRAPHIC || litDefPtr->DataType == QMY_VARGRAPHIC))
-          {
-             if (litDefPtr->DataType == QMY_VARCHAR || litDefPtr->DataType == QMY_VARGRAPHIC)
-             {
-                tempPtr = literalPtr + sizeof(uint16);
-             }
-             else
-               tempPtr = literalPtr;
-          }
           literalPtr = literalPtr + litDefPtr->Length;   // Bump pointer for next literal
         }
         boundsPtr->HiBound.Position = literalCnt;
diff --git a/storage/ibmdb2i/ha_ibmdb2i.h b/storage/ibmdb2i/ha_ibmdb2i.h
index 2a8d65825bf..006b6b163b8 100644
--- a/storage/ibmdb2i/ha_ibmdb2i.h
+++ b/storage/ibmdb2i/ha_ibmdb2i.h
@@ -383,7 +383,15 @@ private:
   int32 prepareWriteBufferForLobs();
   uint32 adjustLobBuffersForRead();
   bool lobFieldsRequested();
-  int convertFieldChars(enum_conversionDirection direction, uint16 fieldID, const char* input, char* output, size_t ilen, size_t olen, size_t* outDataLen);
+  int convertFieldChars(enum_conversionDirection direction, 
+                        uint16 fieldID, 
+                        const char* input, 
+                        char* output, 
+                        size_t ilen, 
+                        size_t olen, 
+                        size_t* outDataLen,
+                        bool tacitErrors=FALSE,
+                        size_t* substChars=NULL);
 
   /**
     Fast integer log2 function