FreeTDS Development Group

[Brian Bruns <camber AT ais.org>]

On 26 Jun 2002, Stan Liberman wrote:

> 
> 574400 - Several patches for thread safety in tds. Our apps consistently
> core dumped at attempt to login from several threads. Pretty much just
> replaced non thread-safe functions with their safe equivalents.

This gets into a sticky situation, older unices will not have the _r 
versions of the calls, so we should be checking via autoconf for their 
existence and using only then.  So, something like 

#if HAVE_LOCALTIME_R
        tm = localtime_r(&t, &res);
#else
        tm = localtime(&t);
#endif

and in configure.in adding:

AC_CHECK_FUNCS(localtime_r)

Otherwise, I'm all for this one going in.

> 574402 - If connection was dropped ( we created it by simply killing the
> database, and once disconnected the network cable ) the application
> using dblib was going into tight loop consuming 99% of CPU.
> There are two fixes: one for tds/token.c checking of the marker is 0 in
> tds_process_cancel, other one for dblib/dblib.c dbsqlok() to make sure
> that tds_process_default_tokens() returns success.
> I'm not sure if this is the proper way of fixing this particular
> problem, but we've been running with this patch for several weeks now
> without any problems.

Looks like you are returning TDS_ERROR from the dblib routines?  This is 
not a valid return, we should be using SUCCEED or FAIL always.  Otherwise, 
the patch looks ok, was it intended only to deal with tds_process_cancel?

It's definitely better what we have now, but I'm wondering how 
comprehensive it is?
> 
> I hope these patches will make it into release, or I'll have to create
> our own branch again (which is no fun). Just a note :)
>