FreeTDS Development Group

["Brian Bruns" <camber AT ais.org>]

Freddy,

The problem is row buffering.  you may store many rows in the row buffer
and expect there null values to reflect that row as you traverse through
them.  This applies only to dblib, and perhaps row buffering in dblib
needs to take on this responsibility in order to keep libtds as clean as
possible.  Any of you dblib folks working on row buffering? I think Craig
did the orignal implementation.

row buffering is also broken for blob types for the same reason.

Brian

> I don't understand the reason to store null bool bit into current_row...
> Perhaps past implementation do not store field information (like size of
> pointer to data) into TDSCOLINFO and store all in current_row.
> I think the fastest way to store null flag is to initialize
> column_cur_size to -1... not require extra space...
> 
> The problem came with params. When returning params we don't know how
> many params are present in result, so we extend current_row for every
> params arrived. Current implementation do not take into account null
> flags (not allocated or handled) while RPC can return null params...
> I'm implementing code to handle null bit but I think storing in
> TDSCOLINFO is simple...
> 
> freddy77