Re: ignore --tdsver

On Thu, 08 Sep 2011 12:22:14 -0500
"Craig A. Berry" <[email protected]> wrote:

> 5.0 may not be a sensible default in this day and age.  But I'm not
> sure there is a different one that makes sense either.  

I'd like to riff on that a bit.  The right default might be "auto",
which is currently supported in, er, CVS HEAD.  

The recent discussion of dbsetversion() is illustrative.  The
documentation says if you don't call it, you get TDS 4.2.  That makes
sense: recompile old code, link to new library, get same behavior.  To
get new behavior, call the (previously nonexistent) function.  

But it also doesn't make sense.  Write brand new code, and to get
modern behavior you have call the (otherwise meaningless) function.
ODBC has been stuck with this for a decade, cf. SQL_ATTR_ODBC_VERSION.
You actually call the API to change the API. 

Larry Wall mentioned a similar artifact once in writing about Perl 6.
He remembered IIRC a setting in VMS mumble something like SET EXTENDED
ON to make it DTRT.  He observed that thereafter *every* script needed
that boilerplate in its preamble.  One gets the sense Larry is not a fan
of boilerplate or, for that matter, preambles.  

That's the world of binary interfaces and closed source.  Free
software lets you set the "default defaults" with ./configure.  Because
you control the source, you can set the default at compile time, which
is why the inoperability of dbset[l]version went unremarked-on for
years.

If we were to remove --with-tdsver (which we won't) we'd remove
only the ability to set the compiled-in default.  If you don't use that
configure option, you get 5.0.  It has to be *something*, even if it's
an invalid value that forces you to use one of the runtime or
programatic methods to set it to something valid.  

Which brings us back to: what should it be?  

Which makes me ask: what's a configure file for, anyway?  

freetds.conf associates a hostname, port, and TDS version with a local
name.  Hostname is converted to an IP address by DNS or similar.  Port
is discoverable on Microsoft servers using 1434.  And TDS version can
be found by starting high and working down.  Old servers reject
logins using newer versions of the protocol; from their perspective the
new version is a protocol violation.  By trying successively lower
version, a client can discover the highest one supported by a server.  

What that means to me is that FreeTDS in its next release should
normally be run without a configuration file or environment variables.
The default-default TDS version should be 0, which tells the login
function to keep trying TDS flavors as long as something accepts the
tcp/ip connection.  

That said, there is miniscule overhead in retrying logins, and it is
unnecessary if you know which TDS version the server wants, and for
some applications a few hundred milliseconds' delay to log in
matters.  The spirit of the project demands that users be able to nail
things down, even to the point of setting the compiled-in default
behavior.  

End of Riff.  Thanks for listening.  

--jkl