Re: Patch for snprintf problem (bug #1000650)

Fri, Jun 09, 2006 at 02:13:23PM +0200, Ludek Finstrle napsal(a):
> > > > buf = "text";
> > > > snprintf(buf,size,"%s append",buf);
> > > >
> > > > buf = "text append" on Windows (MS VC compiler)
> > > > buf = " append" on linux (gcc compiler)
> > > >
> > > > I solve it this way (main idea):
> > > > snprintf(buf + strlen(buf), " append");
> > >
> > > Hmm bad news.
> > > If so, it may be better to use the sequence like the following for example.
> > >
> > >     char    *query_ptr;
> > >     size_t    bufsize_res;
> > >     int    slen;
> > >
> > >     /* Initialize */
> > >     query_ptr = columns_query;
> > >     bufsize_res = sizeof(columns_query);
> > >
> > >     if (..)
> > >     {
> > >         if ((slen = snprintf(query_ptr, bufsize_res, .., )) <= 0)
> > >         {
> > >             .. error_handling ..
> > >         }
> > >         query_ptr += slen;
> > >         bufsize_res -= slen;
> > >     }
> > >
> > >     ...
> >
> > I'm not sure. There are a lot of strcat, my_strcat, schema_strcat calls
> > between snprintf. There is only one place where snprintf could be called
> > after another snprintf.
>
> I change the snprintf_addlen to snprintf_len as you mentioned.

I take a look at all snprintf in psqlodbc an I find one more place
where it's problematic.
If we accept snprintf_add I change one call to it.

Additional patch attached, comments are welcome

Luf