FreeTDS Development Group

[christos AT zoulas.com (Christos Zoulas)]

On Dec 12,  2:43pm, LowdenJK AT bernstein.com ("Lowden, James K") wrote:
-- Subject: RE: [freetds] HP/UX and iconv

| > From: Marquess, Dustin [mailto:Dustin.Marquess AT allegiancetelecom.com]
| > Sent: December 12, 2003 1:45 PM
| > 
| > Sounds like some code that is not 64-bit clean.  If you 
| > compile FreeTDS with
| > the cc compiler with the -taso -xtaso options, it will 
| > probably work around
| > these problems.
| > 
| > You should be able to run gdb or ladebug against those memory 
| > locations to
| > find the culprit.
| 
| Thanks, Dustin.  You're absolutely right.  I just read uac(1); I haven't
| found the gcc switch, if it exists, to align things.  
| 
| Here's an example.  In write.c::tds_put_string:
| 
| tds_put_string(TDSSOCKET * tds, const char *s, int len)
| {
|       TDS_ENCODING *client, *server;
|       char outbuf[256], *poutbuf;
|       unsigned int inbytesleft, outbytesleft, bytes_out = 0;
| [...]
|       if (-1 == tds_iconv(tds, tds->iconv_info[client2ucs2], to_server,
| &s, &inbytesleft, &poutbuf, &outbytesleft))
| 
| 
| We take the address of outbytesleft.  Because it's declared behind an int,
| it's on a 4-byte boundary.  Inside the iconv() function (our replacement),
| there's this line:
| 
|       copybytes = (*inbytesleft < *outbytesleft)? *inbytesleft :
| *outbytesleft;
| 
| which generates a fixup message, see below.  What to do?  

Fix the bug in the code :-) tds_iconv expects "size_t *", and you
are passing it "unsigned int *". This will break on LP64 systems
where size_t is unsigned long and sizeof(long) != sizeof(int).

so s/unsigned int/size_t/ in the above snippet and you are all set.

christos