FreeTDS Development Group

[Dirk Koopman <freetds AT tobit.co.uk>]

On Fri, 2005-02-11 at 16:46 +0100, Frediano Ziglio wrote:
> Il giorno ven, 11-02-2005 alle 10:10 -0500, James K. Lowden ha scritto:
> > James Vanns <jimv AT canterbury.ac.uk> wrote:
> > > 
> > > There is a default block size:
> > > 
> > > [global]
> > >   # TDS protocol version
> > >   tds version = 8.0
> > >   initial block size = 1024
> > > 
> > > I know nothing about dblib or perl but I do know I have had a similar
> > > problem in the past where the block size was to small for what was being
> > > retrieved.
> > 
> > "initial block size" will not have any effect on the reported problem.  
> > 
> > It affects the size of TDS packets, roughly similar to the MTU in tcp/ip. 
> > Larger packets mean fewer packets (and fewer round trips to the server),
> > but also a larger recovery unit in the event of a lost/damaged packet. 
> > 
> > I suspect the problem lies somewhere in ct-lib.  First step would be to
> > create a (broken) test case.  Well, step zero would be to test with
> > RC10....
> > 
> 
> Mmm... easier... if 0.62.4 contains rebindpar test (under
> src/odbc/unittests) it do not truncate... if not it truncate... Fixed
> many time ago in 0.63. 0.63 support much better DBD::ODBC. Current 0.64
> support even better (only a sub test fails... cursors...)

I have now fixed this problem. It was nothing to do with freetds. It is
an error in DBD::ODBC which I have notified the author about. Basically
it tries, and fails, to keep track of the maximum size of parameter to a
bind. It reverts to a maximum of 80 characters in all cases.

A quick fix is to change the '80' to '1000' in dbdimp.c but I suspect
that there is a more elegant solution ...

Sorry to have bothered you.

Dirk