Re: ping DelM: coInitialize

Guillem,
No, I have not. It appears that in Delphi 5, when an Isapi is instantiated
that if an ADO connection object exists, there is some strange behavior.
During the creation of the connection object, which is not in my code but
rather in the ADO unit that is in the "uses" clause, if coinitialize has not
been called, then the object cant be created. This only happens under
certain variations of Windows Server 2K and some NT4 Server with certain hot
fixes applied. I have a web farm and this only occurs on one machine in that
web farm. If I include the "coinit" unit as the first unit in the uses
clause of the main dll, then the error goes away because that program is run
before the datamodule is created. I believe that Kevin Frevert passed that
around some 1 or 2 years ago and it has been fine. Using that unit, I have
never had to actively call coinit or couninit in any of my ISAPI programs. I
do not start separate threads in my ISAPI, they are in effect 1 thread each.
I saw some obscure references in some M$ documentation that stated
"coinitialize is required inorder to create an ADO object", leaving me with
the impression (which might be wrong), that coinit has nothing to do with
ADO execution, only with creation of objects that use ADO. THat would
properly explain the behavior that I see where a connection object on a
webmodule will cause the program to abort when the object is created and
coinit is not called. This happens even if there is no connection string and
no open, it is just during the create process.

Lots of people have said to call coinit before each execution. I think that
is not required, otherwise, they should have written the ADO unit to do it
for us. I can not imagine changing all my code to call coinit just before an
"myDataObject.open" on a stored procedure that returns data or a tadodataset
that returns a result set. Coinit is supposed to be so that ado can talk to
oledb library. If my tadodataset referes to a connection object, then I
think that the connection object is the actual pipeline to the oledb call
and that is where the coinit is required.

So I then put CoUnint followed by CoInit in my "beforedispatch" event, ...
didn't make any difference.

I am switching to Delphi2005 over the next couple of weeks and hope that the
problem will go away.
What really pisses me off is this ... if "coinit" is so damn sacred, why
doesn't it happen automatically. We invented the starter for the automobile
about 100 years ago and dont have to get out of the car and crank it by hand
every time we want to start it .. it seems to me that those guys at M$ that
claim to be so brilliant, could have applied the same principles to their
software and made the coinit built in .. computers are supposed to do the
work, not me.

I know I'm ranting ... I'm just tired of being the victim of "unannounced"
features in all this stuff. It's as bad as the old days when you bought
dBase and found out the next release was the one that really allowed you to
write your own procedures despite what the "box" said in the advertisement.

I have been in this business since the first IBM 360-30 hit the street and
on mainframes I never, ever had these kinds of problems.


.

Relevant Pages