Re: tds_connect does too much

ZIGLIO, Frediano, VF-IT wrote:
> > 	if (tdsver == 0) {
> > 		for( i=0; i < maxver; i++ ) {
> > 			tdsver = versions[i];
> > 			int fOK = tds_login(tdsver);
> > 			if( fOK )
> > 				return fOK;
> >
> > 			if ((conn = tds_connect()) == NULL)
> > 				return tdserror();
> > 		}
> > 	}

> I don't like this that much... connect on second but not for first...

Yeah, I know.  We need logic like this:

	log_me_in
	  do {
	    connect
	    switch(next(tdsver)):
	    case TDS8
	      tds8_login 
	    case TDS7
	      tds7_login 
	    case TDS5
	      tds5_login 
	    case TDS42
	      tds5_login 
	  } while (!logged_in)

I objected to the name "tds_connect" for something that connects and logs
in.   But maybe it would be OK to have "tds_log_in" (two words: one "logs
in" to a server) both connect and log in.  

I'll work on it some and we can discuss further when we have something
concrete. 


> I never liked that name resolution is in config.c, I thing that should
> be in net.c. 

Agreed.  I'm also not crazy about defaulting to "SYBASE" as a last-ditch
name.  I don't think that does anything except confuse people.  

> To detect problems 
> related to tcp/ip connect in tsql we could test for 
> TDSSOCKET->s, is valid only if there is a connection 

Noted.  

> About naming convention we have
> 
> tds_submit/put_XXX  send packets (query.c) there is also a mix of
> submit/put/send... any  difference ??

Not much.  I like "put" for small things (e.g. fputc), "send" for blocks
of data (e.g. send(2)), and "submit" for whole packets that demand a
reply.

> tds_process_XXX     read and process packets (token.c)

At least it's regular. 

> There is also a tds8_do_login

I think I'd like to rename all those version-specific logins to just
"tds_log_inX" and forget about "send" or "do" or whatever.  

But the weather's supposed to by nice this weekend, so don't hold your
breath.   ;-)

Regards, 

--jkl