pgsql-odbc and iodbc (on OSX, at least)

Hi,

I've been chasing down a confusing problem with pgsql-odbc and iodbc.
I've downloaded the latest version from git, configured with
--with-iodbc, compiled, and tried running this simple test program with
surprising results:

#include <sql.h>
#include <sqlext.h>

int main(int argc, char** argv)
{
    SQLHENV henv;
    SQLHDBC hdbc;
    SQLCHAR connection_out[1025];
    SQLSMALLINT connection_out_len;
    SQLAllocEnv(&henv);
    SQLAllocConnect(henv, &hdbc);
    printf("Connect: %d\n", SQLDriverConnect(hdbc, NULL, DRIVER_STRING,
SQL_NTS, connection_out, 1024, &connection_out_len, SQL_DRIVER_NOPROMPT));
    printf("Status: %d\n", SQLSetConnectOption(hdbc, SQL_AUTOCOMMIT, 0));
    return 0;
}

I get Connect: 0, Status -2. (I should say this is also true of the
prebuilt binaries for OSX last time I tried them). Time for some
digging. pgsql-odbc does not implement SQLSetConnectOption, so I
expected iodbc's call to dlsym(..., "SQLSetConnectOption[W]") to return
NULL. It doesn't. Digging deeper, I realised that this is defined in
iodbc, and because iodbc-config --libs returns "-L/usr/lib  -liodbc
-liodbcinst", psqlodbc[w].so is linked to libiodbc.x.dylib. This
explains which function it's finding: the driver manager is getting its
/own/ SQLSetConnectOption, and thinking it's the driver's one. When it
then calls the function, predictably the result is that the handle is
not recognised, and the call is rejected. (Yes, I know
SQLSetConnectOption is deprecated and has been for a long time now. This
sample code above is really a test case from a much larger application
which I don't control the source for.)

I edited the Makefile and removed the -liodbc, relinked, and presto -
everything worked. I don't know what the right answer is here: Is this
the fault of iodbc for returning -liodbc in --libs? (I'm not entirely
sure what --libs is intended for. Clearly not database *drivers* if it's
going to produce this logic!) or is this the fault of pgsql-odbc for not
doing something else? It's trivial enough to patch around by judicious
use of sed, but I see that you don't use sed anywhere else in the
autoconf files. Rather than reach into my personal toolbox and apply a
band-aid, I thought I'd better just point out the problem to the list in
the hope I can it fixed properly, and save someone else the same
head-scratching I went though.

While I have your attention, and discussing deprecated things, in
configure.ac I see
AM_CONFIG_HEADER([config.h])
which I had to replace with
AC_CONFIG_HEADERS([config.h])
before autotools would let me bootstrap. Is it possible to change this
as well?

Thanks,
Matt

--
_____________________________________________
Matt Lilley
Software Engineer
SecuritEase

Tel:    +64 4 912-2100
Fax:    +64 4 912-2101
E-mail: matt.lilley@securitease.com
Web:    http://www.securitease.com
_____________________________________________

This e-mail has passed our content security scan.
It is covered by the confidentiality clauses at http://www.securitease.com/content_and_confidentiality