FreeTDS Development Group

["SourceForge.net" <noreply AT sourceforge.net>]

Patches item #832242, was opened at 2003-10-29 09:17
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=407808&aid=832242&group_id=33106

Category: odbc
Group: None
Status: Open
Resolution: None
Priority: 5
Submitted By: David Fraser (djwfraser)
Assigned to: Nobody/Anonymous (nobody)
Summary: Patch for FreeTDS without an ODBC driver manager

Initial Comment:
As promised here are details of what needs done to get
FreeTDS ODBC working without a driver manager.

I leave it to the core-developers to review/critique/
complain/modify and hopefully accept the patch.

What it provides is mainly the following:
* Ability to make an ODBC driver standalone (i.e. with no
   driver manager).
* Provides configure script option for this, which also 
   defines the existing TDS_NO_DM define, thus providing
   a means to enable this existing code (put together by
   Freddy & Steve Murphree) via configure.
* Allows unittests to build for this standalone driver 
   environment.
* Minor fixes to buglets that got in the way of the
   above.

Pre-requisites
* The files sql.h, sqlext.h and sqltypes.h need to be 
   available. These can be taken from either the iODBC 
   or UnixODBC distributions. Either set of files work
   equally well. I copy them into /usr/local/include, in my
   setup. This directory is configurable - see next.

Code changes - the patch for this is attached and is 
relatively trivial:
1) I changed configure.in, adding in a clause which
    defines a "--with-odbc-nodm" option, much like 
    "--with-iodbc" or "--with-unixodbc". Indeed it rips of
    the configure.in settings for these options, since I'm
    not that familiar with configure scripts. As with these
    options a directory is supplied as a parameter - such
    as /usr/local. The include files (sql.h, sqlext.h etc)
    are looked for in the "include" subdirectory below this
    point (as with the iodbc and unixodbc options).

    I'm open to renaming the option for aesthetic
    reasons, and if anyone wants to alter my patch to
    the configure script to allow the supplied parameter
    to be the full include directory
    - i.e. /usr/local/include and not just /usr/local, then
      that's fine with me. On the other hand it would be
      consistent with the --with-unixodbc and 
      --with-iodbc setup to leave it as I propose.

    This new clause also defines TDS_NO_DM, when
    --with-odbc-nodm is set. This is the define that 
    Freddy mailed me about, that adds extra driver
    manager functionality in a DM-less scenario.

2) A complimentary change is required to 
    src/odbc/unittests/Makefile.am, to ensure the right
    files are linked with the unittests. This fix also allows
    iODBC to be built with the unittests (if so
    configured): previously, -lodbc was hard-wired into
    the makefile, whereas -liodbc is required for iODBC.

3) Changes are required to several include files to ensure
    sql.h and sqlext.h are included for the DM-less case
    (ifdef'd on TDS_NO_DM). These files are
    include/tdsodbc.h, src/odbc/prepare_query.h and
    src/odbc/odbc_util.h

4) There was an outstanding buglet in the code in
    odbc.c at line 4601. Extra {}s were required around
    the if body in order to get the code to compile.

5) I added an implementation of SQLCloseCursor ifdef'd
    for TDS_NO_DM, to give the unittests a chance of
    linking for this case. My one question here was
    whether or not it is appropriate to change the 
    hstmt->cursor_state to closed. Not that I noticed
    any problems with the code as it is.

So with these changes in place, the procedure to install 
this is:

1) Ensure the sql header files are in place (see pre-
    requisites above)
2) Get the FreeTDS code with patch applied.
3) Run autogen.sh --with-odbc-nodm=/usr/local 
                         [--with-tdsver=7.0] (etc)
4) make
5) make install.

I also verified that the code still works when configured
with the iODBC and unixODBC driver managers,
using "configure" which was generated by
the above invocation of autogen.sh.

Unittest failure results are as follows for builds 
configured with each of the given driver managers (in
my libiconv-less environment): (x/y) indicates x failed
out of y tests - so (0/y) is ideal.

                TDS    CTLIB  DBLIB   ODBC
unixODBC:   0/21   1/21    0/25    5/20* (3/20)
iODBC:       0/21   1/21    0/25   14/20
NO-DM:      0/21   1/21    0/25    7/20

Yes - it was probably pretty pointless testing TDS,
CTLIB and DBLIB multiple times, but I wanted to prove
that my configure.in/automake changes hadn't broken
these somehow.

The unixODBC ODBC figure drops to 3/20 when
ENABLE_DEVELOPING is defined. This trick can't be tried
with the no-DM code as there is a bug in the ODBC code
in this case which I reported earlier).
The failures for unixODBC are what Freddy sees, so 
that's the baseline I'm working to.

I'm not sure what's up with iODBC, but that I got those 
results on a clean build as well, so I'm not taking the rap 
for that. :-) It's possibly down to my environment,
though the unittests had unixODBC specifics in the
makefile.

Ok - there are some failures for NO-DM over and above 
unixODBC, but that's hardly surprising. Two-thirds of the
tests work, i.e. basic useful functionality does work for
the No DM case. It's a good start.

The attached patch was generated using 'cvs diff -u'.

Enjoy!


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=407808&aid=832242&group_id=33106