FreeTDS Development Group

[Bob Kline <bkline AT rksystems.com>]

On Sun, 4 Jun 2000, Bob Kline wrote:

> On Sat, 3 Jun 2000, Bob Kline wrote:
> 
> So I made a set of diffs against the 0.50 sources.  [snip ...]

I'm having a public conversation with myself. :->} 

The patches I sent in earlier this morning were flawed. Here are the
flaws:

1. I was mistaken in my observation that the gcc wcslen() is calculating
   lengths correctly for UTF-16 strings.  It doesn't, because in the
   gcc world, wide characters are four bytes wide, not two.
2. I had hoped that I would be able to avoid touching anything but the
   tds layer, but Microsoft isn't null-terminating the NTEXT data that
   it sends.  The tds layer is obtaining the correct length from the tds
   stream, but the dblib layer was discarding this information.  So the
   tds conversion code can't calculate the length and it can't get the
   length without some modifications to dblib.c.
3. I believe I guessed incorrectly about the return value from 
   tds_convert(), thinking that for string values it should represent
   the number of characters.  It seems more likely that it is supposed
   to represent the number of bytes converted for all types.  Haven't
   been able to find the comments documenting what the real answer is.
4. The upper layers don't provide any way to bind to a wide-character
   string variable, so the destination type comes through as -1 when
   I pass through the data type byte for NTEXT (99).  The real fix 
   for this, of course, is to extend all the upper layers to fill
   this gap, but I'm reluctant to disturb any layers above tds more
   than I have to, so as a stopgap solution, I've added -1 to the
   switch statement in the new tds_convert_ntext() function, preserving
   the 16-bit characters from the original transmission.

The replacement patches below fix all of these flaws.  Please use them
instead of those sent in earlier today.  Sorry for the premature
submission.


*** dblib.c.orig        Tue Nov 30 22:20:34 1999
--- dblib.c     Sun Jun  4 21:05:45 2000
***************
*** 325,332 ****
--- 325,334 ----
           }
           else
           {
+             DBINT srclen = -1;
        
              if (is_blob_type(curcol->column_type)) {
+                srclen = curcol->column_textsize;
                src =  curcol->column_textvalue;
              } else {
                 src = ((char*)buffer_row_address(buf, index)) 
***************
*** 339,345 ****
              dbconvert(dbproc,
                        srctype,                      /* srctype  */
                        src,                          /* src      */
!                       -1,                           /* srclen   */
                        desttype,                     /* desttype */
                        curcol->varaddr,              /* dest     */
                        curcol->column_bindlen);      /* destlen  */
--- 341,347 ----
              dbconvert(dbproc,
                        srctype,                      /* srctype  */
                        src,                          /* src      */
!                       srclen,                       /* srclen   */
                        desttype,                     /* desttype */
                        curcol->varaddr,              /* dest     */
                        curcol->column_bindlen);      /* destlen  */


*** tds.h.orig  Sun Jun  4 08:28:18 2000
--- tds.h       Sun Jun  4 08:24:39 2000
***************
*** 155,160 ****
--- 155,161 ----
  #define SYBDATETIME  61   /* 0x3D */
  #define SYBBIT       50   /* 0x32 */
  #define SYBTEXT      35   /* 0x23 */
+ #define SYBNTEXT     99   /* 0x63 */
  #define SYBIMAGE     34   /* 0x22 */
  #define SYBMONEY4    122  /* 0x7A */
  #define SYBMONEY     60   /* 0x3C */
***************
*** 196,203 ****
                        x==SYBVARBINARY || \
                        x==SYBMONEYN    || \
                        x==SYBTEXT      || \
!                       x==SYBIMAGE)
! #define is_blob_type(x) (x==SYBTEXT || x==SYBIMAGE)
  /* large type means it has a two byte size field */
  #define is_large_type(x) (x>128)
  #define is_numeric_type(x) (x==SYBNUMERIC || x==SYBDECIMAL)
--- 197,205 ----
                        x==SYBVARBINARY || \
                        x==SYBMONEYN    || \
                        x==SYBTEXT      || \
!                       x==SYBIMAGE || \
!                       x==SYBNTEXT)
! #define is_blob_type(x) (x==SYBTEXT || x==SYBIMAGE || x==SYBNTEXT)
  /* large type means it has a two byte size field */
  #define is_large_type(x) (x>128)
  #define is_numeric_type(x) (x==SYBNUMERIC || x==SYBDECIMAL)

*** convert.c.orig      Fri Dec  3 20:06:07 1999
--- convert.c   Sun Jun  4 20:42:59 2000
***************
*** 23,28 ****
--- 23,29 ----
  #include <stdlib.h>
  #include <unistd.h>
  #include <time.h>
+ #include <assert.h>
  
  
  static char  software_version[]   = "$Id: convert.c,v 1.35 1999/12/04 
01:06:07 camber Exp $";
***************
*** 103,108 ****
--- 104,167 ----
           return cplen;
     }
  }
+ 
+ typedef unsigned short utf16_t;
+ 
+ static TDS_UINT utf16len(const utf16_t* s)
+ {
+       const utf16_t* p = s;
+       while (*p++)
+               ;
+       return p - s;
+ }
+ 
+ TDS_INT tds_convert_ntext(int srctype,unsigned char *src,TDS_UINT srclen,
+       int desttype,unsigned char *dest,TDS_UINT destlen)
+ {
+       /* 
+        * XXX Limit of 255 + 1 needs to be fixed by determining what the 
+        *     real limit of [N][VAR]CHAR columns is.
+        * XXX What about NCHAR?  Don't see a constant for it in tds.h.
+        * XXX Case for -1 in switch statement because upper levels don't
+        *     have a way to bind to wide-character types.
+        */
+       TDS_UINT i, cplen, char_limit = 255;
+       utf16_t* wsrc  = (utf16_t*)src;
+       utf16_t* wdest = (utf16_t*)dest;
+       assert(sizeof(utf16_t) == 2);
+       switch (desttype) {
+               case SYBNVARCHAR:
+                       if (destlen > char_limit * sizeof(utf16_t))
+                               destlen = char_limit * sizeof(utf16_t);
+                       /* Fall through ... */
+               case SYBNTEXT:
+               case -1:
+                       cplen = srclen > destlen ? destlen : srclen;
+                       memcpy(dest, src, cplen);
+                       if (destlen < srclen + sizeof(utf16_t)) {
+                               size_t term_pos  = destlen - sizeof(utf16_t);
+                               size_t odd_bytes = term_pos % sizeof(utf16_t);
+                               term_pos -= odd_bytes;
+                               wdest[term_pos / sizeof(utf16_t)] = 0;
+                       }
+             else
+                               wdest[cplen / sizeof(utf16_t)] = 0;
+                       return utf16len(wdest) * 2;
+               default:
+                       /* Assume caller wants conversion to narrow string. */
+                       if (destlen > char_limit && desttype != SYBTEXT)
+                               destlen = char_limit;
+                       cplen = srclen > destlen ? destlen : srclen;
+                       for (i = 0; i < cplen; ++i)
+                               dest[i] = (unsigned char)wsrc[i];
+                       if (destlen < srclen)
+                               dest[destlen - 1] = 0;
+                       else
+                               dest[cplen] = 0;
+                       return strlen(dest);
+       }
+ }
+ 
  TDS_INT tds_convert_binary(int srctype,unsigned char *src,TDS_INT srclen,
        int desttype,unsigned char *dest,TDS_INT destlen)
  {
***************
*** 678,683 ****
--- 737,746 ----
                        break;
                case SYBTEXT:
                        return tds_convert_text(srctype,src,srclen,
+                               desttype,dest,destlen);
+                       break;
+               case SYBNTEXT:
+                       return tds_convert_ntext(srctype,src,srclen,
                                desttype,dest,destlen);
                        break;
                default:


-- 
Bob Kline
mailto:bkline AT rksystems.com
http://www.rksystems.com